Description:FIELD OF THE INVENTION
The present invention relates to a system for automatic detection, identification and classification of microbes and a method thereof. More particularly, the present invention discloses a hybridized deep learning method with fuzzy linguistic model as a decision support system for the automatic detection, identification and classification of microbes to assist users such as microbiologists.

BACKGROUND OF THE INVENTION
Microbes or microorganisms are extremely small living things that are not visible to the naked eyes but are visible with special equipment such as microscopes, electron microscopes etc. The term “microbe” includes bacteria, fungus, viruses, algae, etc. A majority of microbes are exceedingly toxic and cause a range of life-threatening illnesses to different living things including HIV, anthrax, toxoplasmosis, tuberculosis, and plague.

However, some of these microbes are crucial to human life and have a number of practical applications in areas including industry, agriculture, medicine, and others. They are useful for keeping track of environmental changes, wastewater management, the processing of food, healthcare, etc. The study of microbes is vital for the analysts working in clinical and medical research domains, agriculture, and food and beverage production.

Microbes are studied under a microscope using culture techniques to improve understanding of their biological, genetic, and physiological traits. Traditionally there are various methods which are employed to identify and classify microbes like staining methods, biochemical testing, motility testing etc. These methods require trained persons to carry out detection and identification, besides being expensive, time-consuming and do not achieve high accuracy. Besides these, organisms have plenty of morphological similarities, making it difficult to categorize them at times.

There are a number of patents and non-patents literature currently available which are focused on improvements over the traditional identification methods. For example, in a study by Khutlang R, Krishnan S, Dendere R, Whitelaw A, Veropoulos K, Learmonth G, Douglas TS, titled “Classification of Mycobacterium tuberculosis in images of ZN-stained sputum smears”, in IEEE transactions on information technology in biomedicine, 2009 Sep 1, the authors present methods for the automated identification of Mycobacterium tuberculosis in images of Ziehl –Neelsen (ZN) stained sputum smears, obtained using a bright-field microscope. They suggested an artificial neural network-based technique in 1998. Edge-based segmentation, utilizing the clever edge detector, was followed by the extraction of shape data using the discrete Fourier transform. Then, the data is trained and classified using a multi-layered feed-forward network. The method proposed here is restricted to identification of only TB bacterium and is cost prohibitive.

Another non-patent literature titled as “Deep Learning-enabled Detection and Classification of Bacterial Colonies using a Thin Film Transistor (TFT) Image Sensor”, ACS Photonics (2022), by Yuzhu Li, et al, demonstrates the use of a TFT-based image sensor to build a real-time CFU detection system to automatically count the bacterial colonies and rapidly identify their species using deep learning. This method, however, specifically focused on E. coli and other coliform bacteria and does not identify other species of bacteria as well as viruses and fungi.

Both non-patent and patent literature in the existing state-of-art have several drawbacks in the detection, identification and classification of microbes using machine learning and other intelligent systems since there is limited level of accuracy. For instance, the publication by Chen et.al titled, “Effect of ultrasonic and ozone pretreatment on the fate of enteric indicator bacteria and antibiotic resistance genes, and anaerobic digestion of dairy wastewater, Bioresource Technology, 2021 Jan, discloses a machine learning based method for the image detection of wastewater bacteria species. The iterative threshold approach and edge detection based on mathematical morphology were combined in the process to segment images. The training rate of the suggested technique and the conventional backpropagation means or algorithm were also contrasted by the authors. However, an accurate reading is not achieved under this method as the integration of a software framework with optimization in deep learning for the detection, identification and classification of microbes is not initiated.

Another drawback in the existing state-of-the-art is that although the neuro-fuzzy system can overcome technical obstacles, it has certain limitations, including a very slow learning rate and confinement to local minima as is seen in the document titled “Deep convolutional neural networks for image classification: A comprehensive review. Neural computation”, 2017 Aug 24, by Rawat W & Wang Z.

Numerous neural network techniques have already been used in a variety of predictive modelling scenarios in earlier studies. US-31143003-A titled “System and automated and remote histological analysis and new drug assessment”, discloses a quantitative approach to automate the diagnosis process, conventionally carried out by pathologists, by use of advanced methods supported by Al software tools. The desired feature vector is classified using pattern recognition methodologies supported by support-vector machine (SVM) technology or neural network or fuzzy logic or similar Al methodology. However, these techniques have the limitations of slow learning rate.

US-201816754324-A titled “Decision support system and method for water treatment”, discloses a decision support system that uses machine learning applied to historical data from a selected water system or data from other water systems to modify the rules or algorithms used to analyze current data from a selected water system. This has a drawback of not using real time data and the results are not optimized. Consequently, a smart decision-making model is required which is convenient, employs optimized models for prediction of the microbes.

Ordinarily, laboratory-based techniques are used for direct detection and identification of microbes. Direct approaches are of limited use and cannot be used for on-field detection, despite delivering reliable data whereas, for on-field detection, indirect approaches are used. Optical sensors are used for indirect detection techniques including hyperspectral imaging, fluorescence imaging, and thermography as is described in a paper titled “Plant Viral Disease Detection: From Molecular Diagnosis to Optical Sensing Technology—A Multidisciplinary Review. Remote Sensing”, 2022 Mar 23 by Wang et al.

Another drawback in existing the state-of-the-art is that intricacy of the data gathered and the volume of data collected are the limitations of different optical sensing systems. These strategies demand expensive setup and computing costs, as well as expertise in data analytics and statistical methodologies with the help of sensors, to be used effectively. Numerous sensors are employed to detect different environmental conditions. Different sensors like optical sensors and fluorescent sensors are emphasized by many other developments in the detection of microbes; however, these are expensive as disclosed in the paper titled “Plasmonic nano-antimicrobials: properties, mechanisms and applications in microbe inactivation and sensing Nanoscale”, 2021 by Erramilli et al.

In view of the above there arises a need for a detection, identification and classification system which can automatically detect microbes without much human intervention and can solve complex problems. There is a requirement of a low-cost method and system for the detection, identification and classification of microbes which is not expensive, overcomes the issues in categorizing and is highly accurate.

Accordingly, the present invention discloses a cost-effective system for automatic and fast identification of a wide range of microbes to include bacteria, virus and fungi which uses an improved, fast, and adaptive back propagation technique for classification.

OBJECT OF THE INVENTION
In order to overcome the shortcomings in the existing state of art, the main object of the present invention is to provide an automatic hybrid intelligent system for detection, identification and classification of a wide range of microbes using sensors and deep learning means.

Yet another object of the invention is to provide a system for quick and easy detection, identification and classification of microbes with minimum human intervention.

Yet another object of the invention is to provide a method for detection, identification and classification of microbes with minimum human intervention to deliver accurate and faster results at affordable cost.

Yet another object of the invention is to provide an automatic system for detection, identification and classification of microbes which would aid in various practical applications in the fields of microbiology, food and beverage, industries, clinical and medical research, healthcare, agriculture hygiene and sanitation etc.

SUMMARY OF THE INVENTION:
Accordingly, the present invention discloses an automatic system for detection, identification and classification of microbes which employs an intelligent and optimal Deep Learning approach in sensor based networks.

The present invention provides an automatic system for detection, identification and classification of a wide range of microbes using deep learning models which is not possible in the existing state of the art.

The present invention discloses an efficient and optimized neuro fuzzy-based method for efficient performance with an average classification accuracy of greater than 98%. The present invention is capable of predicting multiple classes of microbes. The system’s robustness has been determined by assessment of the performance using measures like precision, F1 Score, recall, and specificity.

The present invention discloses a feasible method for faster detection, identification and classification of microbes at an affordable cost. This is achieved as a result of integrating Fuzzy sets and deep learning techniques which provide a hybrid intelligent system optimized with the backpropagation algorithm acting as an ANFIS optimization parameter.

The system comprises of four modules including a sensor based module, a fuzzy module, a network module, and a decision module. Initially, sensors are used to acquire real-time data, and a detailed analysis is carried out to get insights from the different datasets which has been collected.

Th method of the present invention comprises of acquiring real time data, learning from the data after data wrangling, creating the deep learning model and coming to a decision regarding the type of microbe.

Further the method in the present invention has the ability to handle machine learning problems with large datasets and high dimensional parameter spaces. The method combines at least two optimization methods to deal with sparse gradients, and to deal with non-stationary objectives. For example, AdGrad which is one of the optimization methods to deal with sparse gradients and RMSProp which is yet another optimization method to deal with non-stationary objects, are used in the present invention The method is straightforward to implement and requires little memory. The experiments confirm the analysis on the rate of convergence.

This system and the method of the present invention are capable of being employed in various practical applications in the fields of microbiology, food and beverage, industries, clinical and medical research, healthcare, agriculture, hygiene and sanitation etc.

The system in the present invention primarily centers on the problem-solving approach for microbiologists who need a fast and accurate method detection, identification and classification of microbes with least human intervention.

BRIEF DESCRIPTION OF DRAWINGS
Figure 1(a) displays the architecture of the neuro fuzzy framework.
Figure 1(b) displays the flowchart depicting the overall framework of the system.
Figure 2 displays applying the MATLAB simulated rule surface in the fuzzy environment.
Figure 3(a) displays the Fuzzy based neural network architecture.
Figure 3(b) displays Flowchart illustrating a neuro fuzzy approach.
Figure 4 displays different rule-based systems using fuzzy controllers.
Figure 5 displays the resultant decision tree after pruning.
Figure 6 displays the correlation map.
Figure 7 displays the confusion matrix.
Figure 8 displays the plots for accuracy v/s loss value.
Figure 9 displays the error rate V/s K value.
Figure 10 displays the comparison chart for MAE values.
Figure 11 displays the comparison chart for RMSE values.
Figure 12 displays the comparison chart for R2 values.
Figure 13 displays the comparison based on precision values for optimization.
Figure 14 displays the comparison based on recall values for optimization.
Figure 15 displays the comparison based on F1 score values for optimization.

DETAILED DESCRIPTION OF THE INVENTION WITH ILLUSTRATIONS AND EXAMPLES
While the invention has been disclosed with reference to certain embodiments, it will be understood by those skilled in the art that various changes may be made, and equivalents may be substituted without departing from the scope of the invention. In addition, many modifications may be made to adapt to a particular situation or material to the teachings of the invention without departing from its scope.

Throughout the specification and claims, the following terms take the meanings explicitly associated herein unless the context clearly dictates otherwise. The meaning of “a”, “an”, and “the” include plural references. Additionally, a reference to the singular includes a reference to the plural unless otherwise stated or inconsistent with the disclosure herein.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this disclosure belongs. The system, methods, and examples provided herein are only illustrative and not intended to be limiting.
The abbreviations used in the invention are represented in table 1 as below:
Table 1: Legend of abbreviations
S.no. Particulars Legend
1 Adaptive-network-based fuzzy inference system ANFIS
2 Back-propagation Neural Network BPNN
3 Back-propagation BPA
4 Rectified linear activation unit ReLU
5 Deep Learning Neural Network DLNN
6 Adaptive moment estimation. ADAM
7 Matrix laboratory MATLAB
8 Support vector machine SVM
9 My structured query language. MySQL
10 Classification and regression trees. CART
11 Probabilistic programming system R2
12 Root Mean Square Error RMSE
13 Mean Absolute Error MAE

Some of the technical terms used in the specification are elaborated as below:
• Fuzzy logic-Fuzzy logic, also known as fuzzy models or fuzzy sets are mathematical means of representing vagueness and imprecise information. These models have the capability of recognizing, representing, manipulating, interpreting, and using data and information that are vague and lack certainty. It is designed to solve problems by considering all available information and making the best possible decision given the input.
• Deep learning-Deep learning is a machine learning technique wherein a computer model learns to perform classification tasks directly from images, text, or sound. Deep learning models can achieve state-of-the-art accuracy, sometimes exceeding human-level performance. Models are trained by using a large set of labeled data and neural network architectures that contain many layers.
• Neural network- Neural networks also known as artificial neural networks (ANNs) or simulated neural networks (SNNs), are a subset of machine learning and are the basis of deep learning algorithms or means that use interconnected nodes or neurons in a layered structure. These networks are comprised of node layers, containing an input layer, one or more hidden layers, and an output layer. Each node, or artificial neuron, connects to another and has an associated weight and threshold. If the output of any individual node is above the specified threshold value, that node is activated, sending data to the next layer of the network.
• Neuro fuzzy logic/system- Hybridization of fuzzy logic and neural network results in a hybrid intelligent system that combines the two techniques. Neuro-fuzzy hybridization is widely termed as fuzzy neural network (FNN) or neuro-fuzzy system (NFS) in the literature. It incorporates the human-like reasoning style of fuzzy systems through the use of fuzzy sets and a linguistic model consisting of a set of IF-THEN fuzzy rules.
• Decision tree- A decision tree is a non-parametric supervised learning algorithm, which is utilized for both classification and regression tasks. It has a hierarchical tree structure, which consists of a root node, branches, internal nodes and leaf nodes and is more effectively used for decision support predictive modelling. Up until the required function is achieved, it maps the class labels into several levels based on each attribute in the training data.
• Back propagation neural network-Backpropagation is a process involved in training a neural network. It involves taking the error rate of a forward propagation and feeding this loss backward through the neural network layers to fine-tune the weights. Backpropagation is the essence of neural net training.
• Biosensor- A biosensor is an analytical device, used for the detection of a chemical substance, which combines a biological component with a physicochemical detector. It uses a living organism or biological molecules, especially enzymes or antibodies, to detect the presence of chemicals. It is a sensor with biological sensing element.

The reference numerals used in the present invention are tabulated below in table 2.
Table 2: Legend of Reference numerals
Ser no. Item description Reference numerals
1 System S

2 Sensor based module SM
Biological sensitive element SM1
Transducer SM2
Amplifier SM3
Processor SM4
Display unit SM5

3 Fuzzy module FM
Feature extraction submodule FM1
Cluster based pre-processing submodule FM2
Fuzzy system generator FM3
Fuzzy inference system FM4
Fuzzifier FM5
Knowledge based module FM6
Information system FM7
Rule-based system FM8
Inference Engine FM9
Defuzzifier FM10

4 Network module NM
Back propagation neural network (BPNN) or neural network NM1
Learning means NM2
Hybrid means (Deep learning means) NM3
Neuro fuzzy inference system NM4
Optimizer NM5

5 Decision module DM
Decision tree rules DMT

6 Sensor Data Acquisition/ sensors SD
Optical sensors SD1
Electrochemical sensors/Impedance sensors SD2
Enzyme based biosensors SD3
Temperature sensors SD4
Humidity sensors SD5
pH sensors SD6
CO2 sensors: SD7
Microcontrollers SD8
Storage device SD9

7 Hardware Design HD
Printed circuit board (PCB) HP
Power supply HS
User interface/ Communication interface HI

8 Deep Learning Model Development DLM
Means to interface with sensors SI
Data mining tools MT

9 GPU GPU

10 Application/Output MA
Mobile application, web application, smart application MAS

11 Operation mode OM
Static mode OM1
Dynamic mode OM2

12 Sample set X

The present invention discloses an automated hybrid system (S) with an intelligent and optimal deep learning approach in sensor based networks for detecting, identifying and classification of microbes. The three key components of the system (S) comprise of:
• A neural network (NM1) implemented for the classification of microbes to include the pre-processing of the dataset.
• An exploratory data analysis tool for visualizing and analyzing the dataset to summarize the important characters.
• A prediction module for identifying and classifying microbes.

Methodology
The system (S) and the method of working of the said hybrid system (S) for automatic detection, identification and classification of microbes using Deep learning techniques is as described below.

As per an embodiment of the invention the overall architecture of the decision support neuro fuzzy framework is depicted in figure 1(a). The System (S) consists of at least four modules, namely a sensor based module (SM), a fuzzy module (FM), a network module (NM), and a decision module (DM). The said sensor based module (SM) comprises of a sensor (SD) or a group of sensors (SD) or a group of sensors (SD) with combined functionalities capable of sensing environmental, microbiological or chemical values in a given sample, with components of a biological sensitive element (SM1), a transducer (SM2), amplifier (SM3), processor (SM4), and a display unit (SM5). The fuzzy module (FM) comprises of a feature extraction submodule (FM1), a cluster based pre-processing submodule (FM2), fuzzy system generator (FM3), and a fuzzy inference system (FM4). The fuzzy inference system (FM4) comprises of a fuzzifier (FM5), a knowledge based module (FM6) to include an information system (FM7) and a rule-based system (FM8), an inference Engine (FM9) and a defuzzifier (FM10). The network module (NM) comprises of a neural network (NM1), a learning means (NM2), and a hybrid means or deep learning means (NM3) along with an optimizer (NM5). The fuzzy module (FM) and the network module (NM) are connected by the learning means (NM2) and the rule-based system (FM8) for generating the fuzzified rules. The neural network (NM1) in the network module (NM) is a back propagation neural network (BPNN) and is integrated with a decision-making module (DM). These components provide an automated tool that can help to identify and classify not only a variety of microbes but also species of the said microbes with least human intervention.

An aspect of the invention provides the decision tree path and symbolic rules in neural fuzzy decision tree techniques by using the MATLAB simulation. The key elements of the decision module comprise a method for creating the decision tree rules (DMT) and the decision-making process. Pre-trained networks for classifying text are used and are typically trained using sensor based input data. The architecture based on fuzzy neural networks is designed using the most recent advancements in the field. Thereby a hybrid fuzzy decision tree-based backpropagation method has been used in this system model which assists in making the decision in a top-down fashion. In this hybrid model a statistical learning approach is applied for classification and prediction.

The flow of data and processes including the details of hardware, devices and modules used in the invention are provided in a Flow chart representing the overall framework of the system (S) as shown in figure 1(b). In an exemplary embodiment, the various hardware and devices which together form the system (S) along with the working of the invention and the method thereof are illustrated below.

The system (S) comprises of a set of sensors (SD) capable of sensing environmental, microbiological or chemical values in a given sample, a microcontroller (SD8), and a printed circuit board PCB (HP) to house the sensors (SD), one storage device (SD9), power supply (HS), User interface or communication interfaces (HI) and GPU (GPU). The microcontroller (SD8) is attached to the storage device (SD9) for storing data. The sensors used in the system (S) in an embodiment are listed below.
- Optical sensors (SD1) capable of hyperspectral imaging or thermography or fluorescence detection.
- Electrochemical sensors/Impedance sensors (SD2).
- Enzyme-based biosensors (SD3).
- Temperature sensors (SD4).
- Humidity sensors (SD5).
- pH sensors (SD6).
- CO2 sensors (SD7).

Sensor Data Acquisition (SD) is the first step in the workflow which entails acquiring of data from sensors (SD) that can detect environmental, microbiological and chemical value changes indicating microbial growth. The sensors (SD) are deployed in contact or in contact less way with the sample or culture on which detection, identification and classification of microbes was to be carried out. Data is acquired from these sensors (SD) and sent to the microcontroller (SD8) that has a means (SI) for interfacing with sensors, for further storage and pre-processing. The microcontroller (SD8) ideally has a number of input/output pins that can be used to interface with different types of sensors.

Features from acquired data in microcontroller (SD8) get extracted by the feature extraction submodule (FM1) of the fuzzy module. The extracted features of the acquired data are then preprocessed that includes data cleansing or refining by the pre-processing submodule (FM2) of the fuzzy module (FM). The preprocessed data is subjected to fuzzy system generator (FM3) to obtain fuzzy sets based on input features from trained data during pretraining. These features extracted during pretraining were used to develop the deep learning model (DLM) that accurately classified the acquired data into classes or species of microbes to indicate the presence or absence of microbes. The preprocessed data with fuzzy sets applied is sent to the fuzzy inference system (FM4) as crisp input to obtain crisp output values. The crisp inputs are sent to fuzzy inference system (FM4) where these are converted to fuzzy inputs by fuzzifier (FM5). The fuzzy inputs are then sent to inference engine (FM9) by fuzzifier (FM5) that is controlled by rule-based system (FM8) and information system (FM7) components of the knowledge based module (FM6) wherein fuzzy operators are applied and the fuzzy data is converted to crisp output values by defuzzifier (FM10).

The network module (NM) comprises of a neural network (NM1) integrated with the fuzzy module (FM) with a rule-based system (FM8) and one learning means (NM2) to form a neuro fuzzy inference system (NM4). This system applies neural network (NM1) to fuzzy inference system (FM4) for generating fuzzified rules to predict class labels with respect to input features from trained data during pretraining stored in information system (FM7) using data mining tools (MT). The preprocessed data thus obtained gets standardized. The network module (NM) utilizes an optimizer (NM5) to select the best solution as final parameters for the neuro fuzzy inference system (NM4).
The deep learning model of the invention has a means to act as a user interface (HI) for user interaction with the product or output of system. The neuro fuzzy inference system (NM4) used was chosen from a group of ANFIS, FuNe, Fuzzy RuleNet, GARIC, NEFCLASS, NEFCON. The optimizer (NM5) used along with the neural network (NM1) also uses an activation function from a group of ReLU, sigmoid etc.

The decision module (DM) applies decision tree rules (DMT) to the crisp output values from the fuzzy module (FM) and to the standardized data with class labels from the network module (NM), to classify the acquired data into various classes of microbes and predict species of microbes to be displayed as results on an application (MA).

The application (MA) would facilitate display of the results on any smart devices like mobiles, tablets, smart watches or PC allowing users to easily access and analyze the acquired data. The application (MA) was selected from a group of mobile applications, web applications, smart applications (MAS) etc. capable of functioning on wireless or wired smart devices. The application (MA) was predesigned to provide real-time updates on the detection, identification and classification of microbes, as well as provide historical data and analytics.

The method for automatic detection, identification and classification of microbes used in the present invention comprises of the steps mentioned below.
- Acquiring data from at least one sensor (SD01, SD02, …, SD0n) deployed in sample, to detect changes in environmental, microbiological and chemical values that indicate microbial growth in said sample.
- Sending said acquired data to said microcontroller (SD8), that runs on said means to interface with sensors (SI), for further storage and preprocessing of said acquired data.
- Extracting features from said acquired data in microcontroller (SD8) by employing feature extraction submodule (FM1).
- Subjecting said extracted features of said acquired data to further preprocessing to include data cleansing or refining using preprocessing submodule (FM2).
- Subjecting said preprocessed data to fuzzy system generator (FM3) to obtain fuzzy sets based on input features from trained data during pretraining.
- Sending said preprocessed data with said fuzzy sets applied, as crisp input to fuzzy inference system (FM4) to obtain crisp output values.
- Generating fuzzified rules to predict class labels with respect to input features from trained data during pretraining stored in information system (FM7), using data mining tools (MT) at the network module (NM), to obtain standardized data.
- Employing optimizer (NM5) to choose the best solution as final parameters for said neuro fuzzy inference system (NM4) of said network module (NM).
- Applying crisp output values from said fuzzy module (FM) and standardized data with class labels from said network module (NM) along with decision tree rules (DMT), to divide said standardized data into various species of microbes.
- Predicting species of microbes as results based on output from decision module (DM).
- Displaying of said results to provide real-time updates on the detection of microbes along with providing historical data and analytics on application (MA) using a GPU (GPU) that also has user interface or communication interface (HI) to allow interaction of users with product or output of system.

The method of pretraining of fuzzy module (FM) comprises of steps as summarized below.
- Obtaining a plurality of datasets of microbes from repositories or sensors as training datasets.
- Preprocessing and refining of said training datasets of microbes.
- Analyzing said training datasets of microbes to list important features.
- Defining universe discourse and linguistic intervals.
- Establishing fuzzy set and membership functions.
- Determining topology, initial weights, bias and threshold length.
- Performing fitness evaluation.
- Adjusting weights, bias and threshold based on the calculation of errors.
- Defuzzifying based on interval once iteration is reached.
- Optimizing said neural network (NM1) using optimizer (NM5) with activation function.
- Applying classification and regression analysis to datasets for correlating microbiological and chemical features of microbes.
- Predicting class labels as per input features.
- Forecasting of microbes based on back propagation.
- Testing using test dataset to obtain trained data.

This system (S) correlates microbiological and chemical features of microbes in a hybrid neuro fuzzy method to predict a wide range of species of microbes using a simple, accurate, quick, portable, non-laboratory arrangement without human intervention at affordable cost. Once the sensors acquire data from the sample or culture, an output bringing out the results in the form of detected microbes and their classification is obtained instantly on the application (MA) through the above-described automated process of the invention.

When the sensors are selected, the hardware is designed to include the printed circuit board, PCB (HP) to house the sensors, microcontroller (SD8) or other processing unit, power supply (HS), and user interface or communication interfaces (HI). After the hardware is designed, the software is developed to provide the Deep Learning Model Development (DLM). This process includes developing the deep learning means also referred to as hybrid means (NM3) that would be used for detecting microbes, as well as the software or means that would interface (SI) with the sensors and process the acquired data. The software or means would include a user interface (HI) for interacting with the product or output of the system. The extracted features are used to develop a deep learning model that accurately classifies the acquired data into categories to indicate the presence or absence of microbes.

Sensors are analytical instruments that recognize molecules using a biological identification system. The ability of sensors to detect minute metabolites and their sensitivity, specificity, scalability, and cost-effectiveness are key factors in their development. This is largely dependent on how well a sensor’s fundamental components such as a biological receptor and a transducer (SM2) element are combined. The sensitivity of the sensors is determined based on the shape, dimensions, materials, operation mode, detection scheme, surrounding medium, and functional layer.

In the present invention, the properties of microbes are captured using the biorecognition element by monitoring the rapid changes in metabolite levels in real-time. The detection of metabolic activity takes place at the sensor surface, causing the sensor to deform or change its resonance frequencies. Both operational modes (OM), namely static (OM1) and dynamic modes (OM2) have been utilized by sensors (SD) to detect the microbes. In the dynamic mode (OM2), thermal noise drives the cantilever at its resonance frequency which changes depending on the position of the sensor and the mass load when bacteria adhere to this surface. Thus, the performance of the cantilevers in this mode is primarily determined by their resonance frequency, effective mass, and the quality factor of the surrounding medium. In static mode (OM1), the bio recognition results in a change in the surface on just one side of the sensor, which causes the transducer (SM2) deflection.

In one of the embodiments of the invention, sensors (SD) are initially used to acquire real-time data, and a detailed analysis is carried out to get training insights from at least four different datasets which have been collected. All the data is analyzed based on the same parameters, but the input values are taken from different environments. It is thereafter converted into the desired format using data wrangling. A classification and predictive model are then applied followed by regression analysis.

A multiple set of datasets can be utilized for learning and training of the models of the system (S). The data could be used from real time sensors or from datasets from repositories. In one of the embodiments, four different datasets were used for learning and classification. These dataset details provide for the classification of microbes as given in table 3 below.

Table 3: Dataset details
Dataset No of instances No. of attributes
UCI repository 221579 20
Kaggle repository 255296 26
China National GeneBank database 30529 24
Real time sensor-based data 50728 20

The invention discloses a hybridized deep learning method with the fuzzy linguistic model as a decision support system. It correlates the microbiological and chemical features of the microbes in diverse fields, which decreases the overall time spent by the industry persons on the data analysis. The neural architecture is employed in the hybridized approach to predict class labels with respect to the input attributes resulting in data being standardized. Herein, the prediction is done based on the class labels in the trained data and which form the classes or categories of different microbes. This method has shown superior performance compared to the existing manual analysis.

Figure 2 depicts the said system’s architecture in view of the fuzzy linguistic model with rule base surface creation. Its three major elements are a back-propagation network with deep architecture, a fuzzy inference system, and a data mining pool for analytics. The classification and prediction processes are assisted by analytical tools. The rule-based system uses the fuzzy inference module, fuzzification, defuzzification, and knowledge base modules which make up this system. For evaluating the quality of microbes’ properties, the deep neural network module (NM) integrates with decision tree (DMT) objects and serves as the expert system’s interface.

To generate decision trees, an aspect of the invention presents a hybrid strategy that makes use of fuzzy logic and back-propagation. The fuzzy decision tree-based induction method utilized reduces classification ambiguity for various attributes. This hybrid approach, boundary situation issues while dealing with continuous characteristics and prediction issues with high accuracy. Here the lambda cut method is utilized for defuzzification. In an embodiment of the invention, the rule base is simulated using the MATLAB tool and the surface is embedded with neural network (NM1) structure as shown in figure 3(a).

In the present invention, an efficient and optimized neuro fuzzy-based method delivers efficient performance and rule consistency of the generated fuzzy recommendations, which are crucial for data mining. This technique is employed especially for the multiclass classification of microbes in order to improve diagnostic outcomes. The network module (NM) of the invention integrates fuzzy sets and optimization techniques for the prediction. The network module (NM) comprises of neuro fuzzy inference system (NM4) which applies neural network (NM1) to fuzzy inference system (FM4). ANFIS is an example of neuro fuzzy inference system (NM4) used in the invention. The said hybrid intelligent system with the backpropagation means, acts as an ANFIS optimization parameter. The backpropagation means optimizes the fuzzy parameters during the training phase working as hybrid means (NM3). Thus, said network module (NM) is configured to optimize fuzzy parameters from fuzzy inference system (NM6), of trained data during pretraining, to work as hybrid means (NM3) in a neuro fuzzy inference system (NM4) or equivalent.

The neural network (NM1), BPNN as used in the present invention both during pretraining and running of the system (S), provides a supervised learning architecture and has an input layer, an output layer, and typically hidden layers according to the deep architecture similar to a feed-forward neural model. Each of these levels contains nodes that are connected to other nodes in the layers. Each connection has a weight attached to it. Figure 3(a) illustrates the back-propagation neural network’s topology. The fundamental processing of components in each node, the summation and an activation function is shown in figure 3(a).

To overcome the problem of a very slow learning rate and confinement to local minima in case of using a neuro fuzzy system, the use of the back-propagation method optimizes the ANFIS model’s parameters and is employed to anticipate the values. In the present invention, rectified linear unit (ReLU) is the activation function used in the methodology and an ADAM optimizer is used to select the best solution as the final parameters for the ANFIS model.

The BPNN (NM1) is exploited to derive performance derivatives depending on the bias variables analogous with each neuron. Here, epoch and weight-bias components are precisely tuned, which is preceded and in accordance with gradient descent and the naive approach.

The present invention also discloses the resolving of the number of neurons and hidden layers in each layer. The number of neurons and hidden layers are determined by the model input and output variables. The optimized number of neurons in the input layer, output layer and hidden layers ranges can range from 1 to 1000 depending on the number of features in data. For exemplary results, the following requirements were considered to regulate the exact number of neurons in the corresponding hidden layers:
- The number of neurons in the hidden layer is not greater than the size of the output layer and to be less than the size of the input layer.
- It can have two thirds of the total number of the input and output layer.
- It may not be greater than two x number of the input layer.
- It can be the sqrt (neurons in the input layer x neurons in the output layer).

A total of 24 parameters for the input layers and 9 classes for the output labels are considered. Accordingly, the model herein is designed with 24 neurons in the input layer and 9 neurons in the output layer based on the aforementioned specifications. However, an iterative selection is used to fix the optimized number of hidden layer nodes. Accordingly, the network design took the input 24X 24 and produced a vector entry for each class. Each of the hidden layers having the dimension 14 X 14 neurons deep in the network were added for better outcomes. This was mapped to an output layer of 9X 9. The network has fully linked layers by the time it is finished, resulting in a vector with one entry for each class.

In yet another embodiment of the invention, a set of parameters are compiled, and the information standardized within a range {0,1} by using the min-max procedure in the fuzzified linguistic approach. The resulting network is then fed into a uniform distribution to train the neural network. The network’s structure is then improved by upgrading all the bias-weight values in each iteration with respect to the loss value. This process is continued up to the specified number of epochs which also satisfies the least amount of error. By minimizing the error rate, this forecasting approach improves speed and accuracy. The primary goal was to create the most appropriate training architecture and associated training strategy for prediction with the help of the gradient descendent learning rule.

The minimum of a function is reached by the iterative gradient descent procedure used in the invention with the use of derivatives, the Gradient Descent Algorithm enables the system (S) to make the required decisions quickly and effectively. During gradient descent, it is aimed to find the derivative with respect to weight (w) and bias value (b) and is represented using equations (1), (2).
∂J/ ∂w = 2 ∗ Error ∗ ∂error /∂w ……………. (1)
∂J /∂b = 2 ∗ Error ∗ ∂error/ ∂b ……………… (2)
where, the gradient of error with respect to w and b are shown in equations (3),
(4)
∂error/ ∂w = ∂ (y ′ − y)/ ∂w …………………... (3)
∂error /∂b = ∂ (y ′ − y)/ ∂b …………………… (4)
For getting optimized results the error to be achieved is
∂error /∂w = x ………………………………… (5)
∂error /∂b = 1 …………………………………… (6)
By using this approach, the direction to minimize the error is to be found out. Thus, the learning rate determines how large a step is to be taken which is represented in (7) given below:
∂J /∂b = 2 ∗ Error ∗ learning rate ……………… (7)

In the fuzzy decision tree approach, correctness of algorithm or means depends on the formulation of the membership functions. Membership functions, which are arbitrary curves, are used in fuzzy logic. There are numerous membership functions, including triangular, trapezoidal, and Gaussian, available. In this invention, the surface roughness is modelled using triangular membership functions. The triangular membership function is selected in this instance of the hybrid approach because it is understandable by nature and appropriate for the situation holding better accuracy.

From the generalized membership function the said approach has tried all three methods and the best result are seen from triangular membership function regarding the number of rules generated. Table 4 illustrates the membership functions for different microbes as per an embodiment of the invention.

Table 4: Analogy of membership functions for different microbe dataset
Dataset Triangular Trapezoidal
Gaussian
Rules Accuracy Rules Accuracy Rules Accuracy
Microbe1 22 92 17 82 7 78
Microbe2 18 89 12 81 5 79
Microbe3 13 87 9 76 4 74
Microbe4 16 92 13 79 6 67

The simulation in the present invention based on the triangular membership function in the MATLAB tool is depicted in Fig. 3(a) in the earlier section. The triangular membership function can be calculated as in (8) and (9):

0 if x ⩽ vs;
µvs (Fj) = (F j,s−Fj)/( F j,s−F j,vs) if vs < x ⩽ s;
1 if x ⩾ s. ……………………………………………. (8)

(Fj−Fjmin )/(Fj,vs−Fj,min ) if vs < x ⩽ s;
µs (Fj) = 1 if s ⩽ med;
(Fj, med−Fj)/ (Fj, med−Fj, ss) if med < x ⩽ l;
0 if x ⩾ l. …………………………………………. (9)

The logic rules are outlined in the fuzzification module by using important methods that provide an antecedent and consequent form. The fuzzy rules are defined and combined based on different conditions and membership functions and the output results are modified based on these rules. Then the last step is defuzzifying the output distributions. Some of the results for fuzzy controller fuzzification and defuzzification were obtained in MATLA for the combination of different attributes. The figure 4 displays the final crisp output value after the defuzzification process. Then the controller calculates the results based on the other input values.

The fuzzy decision tree technique and back-propagation are combined in the present invention to forecast the outcomes. The means used determines and selects the important parameters to be tested which are created by the likelihood value of membership functions and statistics in a node. Neural networks are well suited for classifying phenomena into specified groups because they can model complex nonlinear relationships. However, the accuracy of neural network (NM1) outputs is constrained, does not tolerate zero error, and only permits the least square of the error values. In addition, the time needed to train a neural network (NM1) can be considerable. By employing fuzzy sets to define inputs and outputs, fuzzy logic systems directly handle input and output imprecision and give programmers more freedom to formulate system descriptions at the proper level. A set of rules for neuro fuzzy approach is formulated and is depicted as a flowchart in figure 3(b). The input comprises input vector, input layers, hidden layers and random assigned weights and the output comprise of class labels. After obtaining the training set data, the universe discourse is defined, the linguistic intervals membership function and the threshold length are defined. Next fuzzification steps are applied. This is done by establishing fuzzy sets and membership functions. The topology, initial weights and bias and threshold lengths are determined. Fitness evaluation is performed and once the iteration is reached, error minimization is performed, and the weights are updated. Next defuzzification is done based on interval and forecasting done on Back Propagation.

The Decision module (DM) of the present invention engages a decision-making framework to make decisions on ambiguous and uncertain circumstances. The predictive analytics is incorporated into the decision-making framework using information system and data mining methodology. Decision tree rules (DMT) are used to divide the data into various groups. The massive tree leaf nodes represent the final groups and unsplit nodes.

In the said methodology of the invention, each data set is divided into a sample set X, based on important features. The splitting condition determines which class these samples fall into. The highest information gain for the corresponding attribute is used as a criterion for decision-making. Then standardization is applied that produces a tree-like structure as a final result. The entire architecture of the decision support framework is depicted in Fig. 1(a).

The results in the form of microbiological characteristics of microbes, which are gathered in real-time using sensors (SD) are employed as input neurons. In this invention, the inputs were fixed during the testing and training phases. In an instance, using Python framework as the runtime environment, the hybrid fuzzy back-propagation technique is trained. Trial and error could be used to optimize the selection of the network’s architecture in relation to hidden layers.

In an embodiment of the invention, the parameters of the dataset are selected from the group of solidity, eccentricity, equivalent diameter, extrema, filled area, extent, orientation, Euler number, bounding box, convex hull, major axis, minor axis, convex area, centroid, perimeter, etc. The data was entered into a MySQL database to make data mining operations easier to complete. The dataset’s facts are used in detecting, identifying and categorizing the pattern. The features work together and are important for the mapping of the microbes’ classification and prediction. The process of cleansing data identifies the necessary properties. Both the noise in the dataset and the irrelevant attributes are removed. Using computational approaches, it is possible to select the relevant attributes.

The selection of microbial application in various industries based on various chemical and microbiological aspects is the subject of decision support systems. Additionally, it determines which type of application is appropriate for each microbe. Backpropagation and a fuzzy decision tree technique are the basis for making this choice. Based on the attributes the type of business is forecast that would emerge from the hybrid means (NM3). The results from an exemplary embodiment in terms of the decision tree are shown in figure 5. The resultant tree is a pruned one and is determined based on the Gini index as the key factor. As per an embodiment of the invention, the correlation map and the confusion matrix are as depicted in figures 6 and 7.

Figures 8(a) to 8(d) depict the relation between loss factor and iteration for each training epoch. To learn back-propagation, in an embodiment of the invention, randomly 60 % is selected as a training set, the remaining 20 % utilized as test data and the other 20% for validation. The training ended after 20k epochs to achieve the maximum accuracy without overfitting and underfitting. The test set’s classification accuracy remained at 98.75% at this time. The empirical results highlight that the validation loss is flattening after the last epoch, and the difference between the training and validation loss widens. The dataset gathered from different sensors is used to test the implementation of the said model. The improved hybrid algorithm’s means implementation is tested with dataset obtained with the help of sensors. The method herein compares with the C4.5, ID3, Naive Bayes method and CART algorithms. The complexity analysis and the performance comparison are shown in table 5.

Table 5: Performance comparison of microbe data and complexity analysis

Algorithm Time Complexity Microbe1 Microbe2 Microbe3 Microbe4
P R F1 P R F1 P R F1 P R F1
ID3 O(mn^2) 0.84 0.82 0.83 0.79 0.77 0.78 0.85 0.84 0.84 0.74 0.72 0.73
C4.5 O(mlogn) 0.83 0.85 0.84 0.87 0.85 0.86 0.85 0.83 0.84 0.73 0.75 0.74
CART O(mnlogn) 0.82 0.81 0.81 0.85 0.83 0.84 0.85 0.80 0.83 0.85 0.84 0.82
Linear NN O(mn) 0.58 0.56 0.57 0.58 0.56 0.57 0.63 0.62 0.62 0.59 0.56 0.52
MLP O(mn) 0.79 0.79 0.78 0.72 0.69 0.71 0.72 0.71 0.71 0.69 0.72 0.74
Naive Bayes O(mn) 0.80 0.78 0.79 0.82 0.83 0.82 0.83 0.83 0.83 0.81 0.82 0.83
Fuzzy inference O(logn) 0.92 0.91 0.91 0.90 0.89 0.89 0.93 0.92 0.92 0.92 0.93 0.95
FuzzyDT O(logn) 0.95 0.94 0.94 0.89 0.88 0.88 0.93 0.92 0.92 0.92 0.95 0.94

In a preferred embodiment the model is cross validated with a group of eight samples and the samples are subdivided into training, testing, and validation sets. For training phase, the corresponding weight values are estimated in such a way that the error rate should be minimum and make sure to be reflective of the same statistical data and shows the generalization ability. Here in the implementation results, the error rate is minimum at the beginning, and it starts to increase when overfitting occurs. The analysis of performance based on the different number of hidden layers is also shown in table 6.

Table 6: Performance statistics for different hidden layer architecture
Model Architecture Phase RMSE MAE R^2
Single HL Training 6.71 4.60 0.87
Testing 6.82 4.43 0.84
Validation 7.94 5.61 0.72
Two HL Training 5.89 3.57 0.89
Testing 5.61 3.48 0.87
Validation 4.92 4.21 0.83
Five HL Training 4.55 3.38 0.91
Testing 4.61 3.13 0.89
Validation 3.87 3.78 0.87

Constrained Optimization approaches to nonlinear predictive model have been tested in the invention. The model of the present invention is tested with different numbers of hidden layers and is evaluated with regard to R2, RMSE, and MAE in order to select the best model as said in the earlier section. The model performs well in the validation set with consistent results. The training and testing data that have minimal hidden nodes are considered to be optimal. In a preferred embodiment the trained model with 5 hidden layers and 14 neurons in each hidden layer, outperforms the other combinations in terms of performance and provides the best accuracy as well. A graph is plotted with respect to the error rate and k value and it depicts that the prediction accuracy of the network is slightly better than that of the network with a single hidden layer and is depicted in figure 9.

The comparison of the present invention method with various decision tree methods is done that depicts the results effectively. The graphs in figures 10 to 12 display the performance comparison for MAE, RMSE and R2 for various hidden layer configurations. The performance analysis is carried out based on the factors of accuracy, specificity, sensitivity, precision, recall and F1 score is also depicted in figures 13, 14 and 15.

The outcomes of an embodiment of the invention based on the framework of the different predictive models prove the efficacy of the methodology. The hybridized approach is faster as compared with the other existing approaches and is more scalable in pruning and thereby improves accuracy. A greater convergence rate is achieved in the present invention with a substantial reduction in the computation time and the feedforward phase will never lead to be a local minimum which is determined based on the fuzzy linguistic variables. The invention provides a heuristic approach with sigmoid as the activation function.

As per an embodiment, for getting better convergence rate, the back propagation means is combined with different optimization techniques and the optimized results are obtained for Gradient Descent. In the second part epoch is fixed according to the convergence rate and learning rate commences from 0.001. The result obtained is shown in table 7 and it is clear that the best result is obtained for a learning rate of 0.003. Here the epochs are fixed based on the optimized convergence rate and the graph shows the respective results for 20K epochs.

Table 7: Comparison of evaluation results for different optimizers for microbe dataset
Dataset Optimization Activation Function Accuracy P R F1
Microbe1 AdaDelta Sigmoid 49.86 0.43 0.46 0.45
Gradient Descent Sigmoid 79.63 0.72 0.88 0.79
RMSProp ReLU 87.59 0.89 0.84 0.87
Momentum Sigmoid 85.47 0.79 0.85 0.82
Adam ReLU 93.57 0.97 0.98 0.97
Microbe2 AdaDelta Sigmoid 50.26 0.49 0.53 0.51
Gradient Descent Sigmoid 81.03 0.76 0.74 0.75
RMSProp ReLU 86.52 0.84 0.89 0.87
Momentum Sigmoid 87.63 0.94 0.96 0.95
Adam ReLU 94.53 0.97 0.97 0.97
Microbe3 AdaDelta Sigmoid 53.78 0.65 0.74 0.69
Gradient Descent Sigmoid 82.02 0.74 0.75 0.74
RMSProp ReLU 87.53 0.89 0.80 0.84
Momentum Sigmoid 88.46 0.95 0.93 0.94
Adam ReLU 95.67 0.91 0.96 0.93
Microbe4 AdaDelta Sigmoid 48.74 0.51 0.48 0.49
Gradient Descent Sigmoid 89.23 0.78 0.82 0.80
RMSProp ReLU 92.47 0.91 0.98 0.94
Momentum Sigmoid 91.68 0.89 0.95 0.92
Adam ReLU 96.5 0.98 0.99 0.98

An approach for gradient descent optimization techniques is called Adaptive moment estimation (ADAM) which is effective when dealing with complex problems involving a lot of data or parameters as it uses minimal memory. It combines, intuitively, the algorithms for gradient descent with momentum and Root Mean Square Propagation. By using the exponentially weighted average of the gradients, this approach is used to speed up the gradient descent algorithm. The technique converges faster towards the minimum when averages are used.

Algorithm 2 for the Adam Optimizer applied in the present invention is as follows. Depending on the number of training steps, batches are equated to minibatch Generator (X, Y, Batch-size). For Number of Batches, miniBatch X, miniBatch Y are equated to batches[j]. Forward Propagation using miniBatch X is done to calculate y’. Cost is calculated using miniBatch Y. Backward Propagation is done to calculate the derivative dW and db. The parameters are updated using the fuzzy rules.

The strengths of the aforementioned approaches are carried over and improved upon by Adam Optimizer to produce a gradient descent that is more optimal. Here, in order to overcome the local minima problems, the gradient descent rate has to be adjusted so that there is minimal oscillation when it reaches the global minimum with a big step size. After each iteration, the gradient descent is adjusted to ensure that the process is controlled and unbiased.

The required and adequate conditions for the convergence of networks using different training patterns are shown in the analysis. A MATLAB based simulator tool is also used here to support these theoretical conclusions. The invention has the ability to properly detect decision boundary patterns using membership functions, which is tough to do when using attribute-based classification approaches and is particularly enabled by fuzzy rules. This system’s development made use of the idea of deep transfer learning to extract reliable characteristics, which were then used to classify the microbes with the inclusion of additional classification modules. In comparison to all state-of-the-art approaches, the system (S) achieved a 98 % classification accuracy with hyperparameters such as lambda cut of 0.6, 0.003 as the learning rate, and optimizer (NM5) as ADAM.

The method combines the advantages of two recently popular optimization methods: the ability of AdaGrad to deal with sparse gradients, and the ability of RMSProp to deal with non-stationary objectives. The method is straightforward to implement and requires less memory. The invention discloses data that confirms the analysis on the rate of convergence. An aspect of the invention tries to clip the gradients with an upper or lower bound with a guarantee of convergence. The system (S) is trained with a higher batch size, more epochs and with a decayed learning rate. This method is aimed towards machine learning problems with large datasets and high dimensional parameter spaces. The training network is optimized with an average accuracy of 98%. By combining fuzzy approach with the neural network (NM1), the generalization capability is increased while the threshold is decreased. Accordingly, an improved, fast, and adaptive back propagation technique for classification was modelled in this invention.
The said deep learning model has performed incredibly well in the automatic detection, identification and classification of microbes in numerous applications in various industries. The empirical results point out that the invention outperforms several existing methods in terms of average accuracy of more than 98% in predicting each class of microbes.

EXAMPLES
The present invention shall now be explained with accompanying examples. These examples are non-limiting in nature and are provided only by way of representation. While certain language has been used to describe the disclosure, any limitations arising on account of the same are not intended. As would be seeming to a person skilled in the art, various working alterations may be made to the method in order to implement the inventive concept as taught herein. The figures and the preceding description give examples of embodiments. Those skilled in the art will appreciate that one or more of the described elements may well be combined into a single functional element. Alternatively, certain elements may be split into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, order of steps of method or processes of data flow described herein may be changed and is not limited to the manner described herein. Moreover, the actions of any flow diagram need not be implemented in the order shown; nor do all of the acts need to be necessarily performed. Also, those acts that are not dependent on other acts may be performed in parallel with the other acts. The scope of embodiments is by no means limited by these specific examples.

In an exemplary embodiment, the various components which together form the system (S) along with the working of the invention and the method thereof are illustrated below.

The system (S) comprised of a set of sensors (SD), a microcontroller (SD8), and a printed circuit board PCB (HP) to house the sensors (SD), one storage device (SD9), power supply (HS), communication interfaces (HI) and GPU (GPU). For instance, NVIDIA GeForce RTX 3090 which features 10496 CUDA cores, 328 Tensor cores, and 82 RT cores was used as a suitable GPU option for the present deep learning application. It had 24GB GDDR6X memory and a memory bandwidth of 936 GB/s. The sensors used in the system (S) in this embodiment are listed below.
- Optical sensors (SD1) for instance, FluoroCycler XT PCR instrument with integrated fluorescence detection or LightCycler 480 PCR instrument with fluorescence detection.
- Electrochemical sensors/Impedance sensors (SD2), for instance, ECIS (Electric Cell-Substrate Impedance Sensing) system.
- Enzyme-based biosensors (SD3), for example Biosensor platform based on glucose oxidase or Nucleic acid-based biosensors or GeneXpert System.
- Temperature sensors (SD4) for instance, DS18B20 which is a digital temperature sensor.
- Humidity sensors (SD5), for instance, DHT11 which is a digital humidity and temperature sensor.
- pH sensors (SD6), for instance, PH-4502C.
- CO2 sensors (SD7), for example, CCS811 an indoor air quality sensor or SCD30 which is a CO2, temperature, and humidity sensor.

The sensors were deployed in contact or in contact less way with the sample or culture on which detection, identification and classification of microbes was to be carried out. Data was acquired from these sensors (SD) which could detect environmental, microbiological and chemical changes indicating microbial growth. The acquired data was sent to the microcontroller (SD8) that had a means for interfacing with sensors, for further storage and pre-processing. Arduino Uno R3 was the microcontroller board used for sensor data acquisition. It had a number of input/output pins that were used to interface with different types of sensors.

Features from acquired data in microcontroller (SD8) got extracted by the feature extraction submodule (FM1) of the fuzzy module. The acquired data was then preprocessed and included data cleansing or refining by the pre-processing submodule (FM2) of the fuzzy module (FM). The preprocessed data were sent to the Fuzzy inference system (FM4) as crisp input to obtain crisp output values.
The preprocessed data is subjected to fuzzy system generator (FM3) to obtain fuzzy sets based on input features from trained data during a prior training phase. These features extracted during training phase were also used to develop the deep learning model that accurately classified the acquired data into categories to indicate the presence or absence of microbes. The preprocessed data is sent to the fuzzy inference system (FM4) as crisp input to obtain crisp output values.

The network module (NM) comprised of a neural network (NM1) BPNN integrated with the fuzzy module (FM) with a rule-based system (FM8) and one learning means (NM2) to form a neuro fuzzy inference system (NM4). This system applied neural network (NM1) to fuzzy inference system (FM4) for generating fuzzified rules to predict class labels with respect to input features from trained data during prior training phase stored in information system (FM7) using data mining tools (MT). The preprocessed acquired data thus obtained got standardized. The network module (NM) having ReLU as the activation function utilized ADAM as an optimizer (NM5) to select the best solution as final parameters for the neuro fuzzy inference system (NM4).

The deep learning model of the invention had a means to act as an interface (HI) for user interaction with the product or invention or output of system. The model developed in the invention used ANFIS as the neuro fuzzy inference system (NM4).

The decision module (DM) applied decision tree rules (DMT) to the crisp output values from the fuzzy module (FM) and the standardized data with class labels from the network module (NM), to classify the standardized acquired data into various classes of microbes and predict species of microbes to be displayed on application (MA). The application (MA) was predesigned to provide real-time updates on the detection, identification and classification of microbes, as well as provide historical data and analytics.
, Claims:We claim:
1. A system (S) for automatic detection, identification and classification of microbes wherein said system (S) comprises of
- at least one sensor based module (SM01, SM02, …, SM0n), said sensor based module (SM) comprising of at least one sensor (SD01, SD02, …, SD0n) capable of sensing environmental, microbiological or chemical values in a given sample;
- at least one fuzzy module (FM01, FM02, …, FM0n), each said fuzzy module (FM) having undergone pretraining, comprising of submodules of
• at least one feature extraction submodule (FM11, FM12, …, FM1n),
• at least one cluster based pre-processing submodule (FM21, FM22, ..., FM2n),
• at least one fuzzy system generator (FM31, FM32, …, FM3n), and
• fuzzy inference system (FM4) which includes at least one fuzzifier (FM51, FM52, …, FM5n), at least one knowledge based module (FM61, FM62, …, FM6n) which includes at least one information system (FM71, FM72, …, FM7n) for storing trained data and at least one rule-based system (FM81, FM82, …, FM8n), at least one inference engine (FM91, FM92, ..., FM9n) and at least one defuzzifier (FM101, FM102, ..., FM10n);
- at least one network module (NM01, NM02, …, NM0n), said network module (NM) comprising of
• at least one neural network (NM11, NM12, …, NM1n), which includes input layer, output layer, and a plurality of hidden layers having optimized number of neurons in each layer,
• at least one learning means (NM21, NM22, …, NM2n),
• at least one hybrid means (NM31, NM32, …., NM3n) for neuro fuzzy inference system (NM4), and
• at least one optimizer (NM51, NM52, …NM5n) to provide the best solution as final parameters for said neuro fuzzy inference system (NM4);
- decision module (DM), said decision module (DM) comprising of decision tree rules (DMT);
- at least one microcontroller (SD81, SD82, …, SD8n) or other processing units;
- hardware devices comprising of at least one printed circuit board (HP1, HP2, …, HPn) to house said sensors (SD), said microcontroller (SD8) or other processing units, at least one storage device (SD91, SD92, …, SD9n), power supply (HS), and at least one user or communication interface (HI1, HI2, …HIn); and
- application (MA) with GPU (GPU), selected from a group of mobile applications, web applications, smart applications (MAS) capable of functioning on wireless or wired devices selected from a group of smart devices such as mobiles, tablets, smart watches, PC
wherein
- said sensor module (SM) is configured for acquiring data from at least one sensor (SD) deployed in said sample, for detecting changes in environmental, microbiological and chemical values over a period of time, to indicate microbial growth in said sample and sending said acquired data to microcontroller (SD8), that runs on means to interface with sensors (SI), for further storage and preprocessing of said acquired data,
- said fuzzy module (FM) is configured for extracting features from said acquired data in said microcontroller (SD8) by employing said feature extraction submodule (FM1), subjecting said extracted features of said acquired data to further preprocessing using said preprocessing submodule (FM2) to yield preprocessed data, subjecting said preprocessed data to fuzzy system generator (FM3) to obtain fuzzy sets, sending said preprocessed data along with said fuzzy sets as crisp input to fuzzy inference system (FM4) to obtain crisp output values,
- said network module (NM) is configured for applying neural network (NM1) to fuzzy inference system (FM4) for applying fuzzified rules to predict class labels to standardize said preprocessed data to obtain standardized data with class labels,
- said decision module (DM) is configured for
• applying decision tree rules (DMT) to said crisp output values obtained from fuzzy module (FM) and said standardized data from neural network (NM1), to classify said standardized data into various classes of microbes and predict species of microbes, and
• displaying results using a GPU (GPU) on application (MA) for user interaction
wherein
- said neural network (NM1) of said network module (NM) and said fuzzy module (FM) are integrated to form a neuro fuzzy inference system (NM4) or equivalent using said rule-based system (FM8) of fuzzy inference system (FM4) and said learning means (NM2) of network module (NM),
thereby enabling said system (S) to predict a wide range of species of microbes using a simple, accurate, quick, portable, non-laboratory arrangement without human intervention at affordable cost.
2. The system (S) as claimed in claim 1, wherein said sensor (SD) comprises of at least one sensor (SD01, SD02, …, SD0n) or a group of sensors (SD) with combined functionalities, capable of sensing values including environmental values, microbiological and chemical values in said sample, said sensors(SD) comprising of optical sensors (SD1) capable of hyperspectral imaging or thermography or fluorescence detection, electrochemical sensors/Impedance sensors (SD2), enzyme based biosensors (SD3), temperature sensors (SD4), humidity sensors (SD5), pH sensors (SD6), CO2 sensors (SD7), other chemical or biological sensors
each said sensor (SD) having at least one biological sensitive element (SM11, SM12, …., SM1n), at least one transducer (SM21, SM22, …., SM2n), at least one amplifier (SM31, SM32, ..., SM3n), at least one processor (SM41, SM42, …, SM4n), at least one display unit (SM51, SM52, …, SM5n) and at least one means to interface with sensors (SI1, SI2, …, SIn);
3. The system (S) as claimed in claim 2, wherein in said sensor based module (SM), the detection of metabolic activity takes place at said sensor (SD) surface, causing said sensor (SD) to deform or change its resonance frequencies.
4. The system (S) as claimed in claim 3, wherein said sensor (SD) utilizes operational modes (OM), to include static mode (OM1) or dynamic mode (OM2).
5. The system (S) as claimed in claim 1, wherein said crisp input are sent to said fuzzy inference system (FM4), where said crisp inputs are subjected to fuzzifier (FM5) controlled by rule-based system (FM8) and information system (FM7) components of the knowledge based module (FM6) to obtain fuzzy inputs, said fuzzy inputs being then sent to said inference engine (FM9) where said fuzzy inputs are subjected to fuzzy operators and thereafter to defuzzifier (FM10) to obtain crisp output values for further processing in the decision module (DM) for detection, identification and classification of microbes.
6. The system (S) as claimed in claim 1, wherein neural network (NM1) of neuro fuzzy inference system (NM4) employs activation function, said activation function being selected from a group of ReLU, Sigmoid.
7. The system (S) as claimed in claim 6, wherein said activation function employed by said neural network (NM1) of said neuro fuzzy inference system (NM4) is ReLU.
8. The system (S) as claimed in claim 1, wherein the neuro fuzzy inference system (NM4) used is chosen from a group of ANFIS, FuNe, Fuzzy RuleNet, GARIC, NEFCLASS, NEFCON.
9. The system (S) as claimed in claim 8, wherein said neuro fuzzy inference system (NM4) employed is ANFIS.
10. The system (S) as claimed in claim 1, wherein said optimizer (NM5) used to select said best solution as final parameters for said neuro fuzzy inference system (NM4) is ADAM.
11. The system (S) as claimed in claim 1, wherein optimized number of neurons in the input layer, output layer and hidden layers ranges from 1 to 1000.
12. The system (S) as claimed in claim 1, wherein method of pretraining of said fuzzy module (FM) comprises of steps of
- obtaining a plurality of datasets of microbes from repositories or sensors as training dataset;
- preprocessing and refining of said training datasets of microbes;
- analyzing said training datasets of microbes to list important features;
- defining universe discourse and linguistic intervals;
- establishing fuzzy set and membership functions;
- determining topology, initial weights, bias and threshold length;
- performing fitness evaluation;
- adjusting weights, bias and threshold based on the calculation of errors;
- defuzzifying based on interval once iteration is reached;
- optimizing said neural network (NM1) using optimizer (NM5) with activation function;
- applying classification and regression analysis to datasets for correlating microbiological and chemical features of microbes;
- predicting class labels as per input features;
- forecasting of microbes based on back propagation; and
- testing using test dataset to obtain trained data.
13. The system (S) as claimed in claim 12, wherein said features extracted from trained data during pretraining are subjected to deep learning model development (DLM) to classify said acquired data obtained from said sensor module (SM) into species of microbes.
14. The system (S) as claimed in claim 12, wherein said network module (NM) is configured to optimize fuzzy parameters from fuzzy inference system (FM4), of trained data during pretraining, to work as hybrid means (NM3) in a neuro fuzzy inference system (NM4) or equivalent.
15. The system (S) as claimed in claim 1, wherein said neural network (NM1) deployed during pretraining or running of system is BPNN.
16. A method for automatic detection, identification and classification of microbes wherein said method comprises of steps of
- acquiring data from at least one sensor (SD01, SD02, …, SD0n) deployed in sample, to detect changes in environmental, microbiological and chemical values that indicate microbial growth in said sample;
- sending said acquired data to microcontroller (SD8), that runs on means to interface with sensors (SI), for further storage and preprocessing of said acquired data;
- extracting features from said acquired data in said microcontroller (SD8) by employing feature extraction submodule (FM1);
- subjecting said extracted features of said acquired data to further preprocessing to include data cleansing or refining using preprocessing submodule (FM2);
- subjecting said preprocessed data to fuzzy system generator (FM3) to obtain fuzzy sets based on input features from trained data during pretraining;
- sending said preprocessed data with said fuzzy sets applied, as crisp input to fuzzy inference system (FM4) to obtain crisp output values;
- generating fuzzified rules to predict class labels with respect to input features from trained data during pretraining stored in information system (FM7), using data mining tools (MT) at the network module (NM), to obtain standardized data;
- employing optimizer (NM5) to select best solution as final parameters for said neuro fuzzy inference system (NM4) of said network module (NM);
- applying crisp output values from said fuzzy module (FM) and standardized data with class labels from said network module (NM) along with decision tree rules (DMT), to divide said standardized data into various species of microbes;
- predicting species of microbes as results based on output from decision module (DM); and
- displaying of said results to provide real-time updates on the detection of microbes along with providing historical data and analytics on application (MA) using a GPU (GPU) that also has user interface or communication interface (HI) to allow interaction of users with product or output of system.
17. The system (S) as claimed in claim 1, wherein said system (S) has practical applications in the fields of microbiology, food and beverage industries, clinical or medical research, healthcare, agriculture, hygiene and sanitation etc.