Description:A SYSTEM/METHOD TO DETECT WEED IDENTIFICATION USING DEEP LEARNING
Field of Invention
The present invention is related to neural network-based field of image recognition. The first step is to use the CED artificial neural network to identify the temperature of a crop. Anything that grows other than heat is therefore considered weeds and individually destroyed, leaving only a small number of weeds. They were accurately recognized by training a second CED neural network to screen them out.
The Objectives of this Invention
The primary goal of this invention is to techno farms will profit from these most recent updates to the expert system utilized in agriculture. One of the key goals is to eradicate weeds or other undesirable plants to reduce the use of pesticides and the contamination of crops and water. One of the neural networks, CNN, first extracts the visual components using a flexible layer with ReLU functionality, then separates weeds from the plan using a high-resolution, completely integrated RELU layer.
Background of the Invention
According to (WO2018/083336A1), the steps in the aforementioned invention pertain to a method for classifying crops and weeds. They include identifying rows of crops using a convolutional encoder-decoder (CED) neural network approach that classifies all plants between the rows as weeds; and, for a small number of weeds found in the sections of the crops, differentiating the weeds from the crops. A CED neural network for crop-weed categorization receives, as input, an image of crops and learns a to-be-learned image by drawing graphic stretches on the picture being used as input at locations corresponding to the rows of the crops to acquire this skill on its own. A CED neural network for crop-row extraction receives, as input, an image of crops and acquires a to-be-learned image by doing the same thing.. Another type of application invented in (KR2018/102188521B1), The steps in the current invention comprise identifying weeds among all plants growing between rows employing Convolutional Encoder-Decoder (CED) neural network technologies; It relates to a method for distinguishing between crops and weeds that includes steps to separate weeds from the crops using an additional CED neural network. This method applies to a small number of weeds found in crop fields. A convolutional Encoder-Decoder (CED) neural network is a neural network that executes convolutional operations on each stage. It has numerous stages between the input terminal and the output terminal, and it is a structure that shrinks and expands in size as it gets to the middle stage. Another method was invented in (CN2019/110288033A), Sugarcane, the provisions of features constructed using convolutional neural network algorithms that the invention discloses, is a type of identification and localization method belonging to the scientific discipline of computer vision, passing through deep-layer convolutional neural networks. The feature-identifying processing is carried out to sugarcane image data to obtain sugarcane identification of features location model; by the image data of the input model, sugarcane surface distinctive data is achieved, and the true coordinative primarily consists of two parts: the basis and training for recognizing positioning system models, and recognition advertising, which transmits data to ancillary equipment.
According to Chinnasamy et al.'s research (2022 International Conference on Computer Communication and Informatics (ICCCI), Coimbatore, India, 2022, pp. 1-3), In a technological age, agricultural automation is becoming more and more critical. Previously, the data was recorded on a straightforward LCD screen. Still, in this article, we establish a novel idea to keep track of the water level in a particular area while also pumping water and sending it via GSM. The analyzed data can then be sent to a gardener or someone who monitors the area using GSM networks. When no water or water has been filled up in a garden or smart city that uses blockchain technology, smart farming is a device that is intended to help inform someone. The notification system can be utilized as a call for local or emergency agencies to respond in instances of emergency. According to Chinnasamy et al. ((2021), Turkish Journal of Computer and Mathematics Education (TURCOMAT), Vol.16, No.2, pp. 2858-2865), they used blockchain to overcome such security flaws, enabling a decentralized distributed blockchain protocol shared across IoT cluster chiefs. The main objective of this essay is to equip farmlands with smart greenhouses with a transportable blockchain-based infrastructure that ensures integrity and anonymity. In this case, green-house IoT sensor nodes use secure, immutable ledgers to operate as a centrally regulated blockchain to optimize energy use. Additionally, we show a vital solution that uses IoT devices and blockchain technology to provide better secure communications for Smart Greenhouse farming.
Summary of the Invention
The critical and precise relevant features are extracted in the aforementioned study using CNN classification and used to train the model efficiently. The preciseness of the suggested model is 62%. Action will be taken to increase accuracy and evaluate the model using more significant features or parameters. The methods outlined in this article would be further expanded to include other weeds and plants in an environment of autotrophs with higher bounds.
Detailed Description of the Invention
Agriculture is now a very important sector that everyone should pay attention to help farmers. The population in our country is growing rapidly in recent years due to which the demand for food is growing so directly or indirectly the farm products demand is also growing day by day. Therefore, farm yields play an important role. New methods are emerging now in the day to keep the crop high by considering the environmental impact. One of the most important steps in increasing yields is weed control as it is directly related to crop yields. So here we are considering weed identification in the farm field using an in-depth study method. Here we are using one of the deep learning techniques i.e. we are using using a convolutional neural network to identify weeds.
Convolutional neural network (CNN) falls under the artificial neural network (ANN) which performs image detection, image classification and processing of the image that is specially developed to process plant or weed images i.e. input images. The neural networks with multiple layers acts like a system of both hardware and software patterned following the process of neurons in our brain. Various old methods of artificial networks are not efficient for handling images and they impose network input images which should be reduced and its resolution decreases. CNN proposes a multi-layer perceptron-like model designed for practical and low-cost computing needs. There are three different types of layers on CNN that contain input layer, output layer and many hidden layers, where the data from the image is scanned using image pixels, these three different layers include several layers of conversion to extract. image elements and then using layers to combine to lower that image. In addition, it also includes fully connected layers and an additional layer called as normalization layers at the end of the CNN.
The convolution layer is one of the central components of the CNN architecture that plays an important role in extracting features from the input image. The sharing of link weights by all neurons in a specific feature map extracted from the image is done in this part. Neurons in a layer are only connected to a subset of the given input image which helps to optimize the number of hyper parameters in the architecture and helps to improve the computation efficiency of the model. The farmers of our country should be made aware of the different types of weeds and should give them the knowledge of the condition of weeds in their fields. This can help them to reduce the use of herbicide in large quantities in agriculture, so that they can spray it as little as possible which in turn reduces the consumption of herbicide.
The development of an automated system for an effective and precise weed control is a set of computer vision and its associated areas. The proposed model is able to group weed varieties on the basis of their similarity and also to understand the amount of leaves with a satisfactory level of accuracy. In order to identify types of weed species and to develop a robust and accurate system with plant leaves, the captured images are required to wrap natural variation in terms of ecological conditions and stages of plant developmental progress. These conditions include the setting of sunlight, the nature of the soil and the condition of the plants. A central problem with automatic leaf identification and counting is that the leaves of weeds are usually stacked on top of each other. Completely automated systems developed to measure plant count with the computer vision techniques which are limited to binary range images. Dataset: weeds and plants dataset is collected from Kaggle and various types of weeds and plants are used as features and weeds plants are used as labels. Preprocessing: before training data each image is converted to 224*224 size and re-scale image. Split data: data set is divided in to testing and validation dataset which is used for train and test. Initialize Model: In this stage VGG 19 model is initialized with required parameters and trained features and labels are fit to algorithm and model is saved to system. Prediction: User uploads weeds and plants as images as input to web application and check with model to predict result. If the output is given as weed then the machine will detect and compare the test image with predefined image and gives an output stating that it is a weed.
Pooling is process which consists of pooling layers which are actually help for down sampling the feature maps by adding up the features into a feature map. There are 2 common types of merging methods which are intermediate mergers (which summarizes the intermediate presence of the feature). Pooling is mandatory step in systems; based on convolution network will reduce the dimensionalities of the feature maps. It combines a set of values into a smaller number of values. It transforms the joint feature representation into valuable information by removing the unwanted information add keeping the useful information which helps the system. The polling operators also foreman special feature while reducing the complexity of the calculation between the upper layers by removing the contact between the convolutional layers. This layer uses a low-sample approach - on feature maps from previous layers and that will produce a new featured editable.
The result of the above data set classifies the image into particular data set. whether, it belongs to plant data set or weed data set basing upon its training the machine will recognize into which category is it. Case1: - so, in that case one if the image is recognized as a weed the machine will state that it’s a weed in the category and description. Case 2: - if the image is recognized as a plant, then the machine will automatically recognize and show it’s a plant. With this approximation of the images the user will get a better knowledge upon what type of autotroph it is training and precision are the most important part of a project the president of our project consists of around 75% true to its nature and it detects any type of weed or plant present in the new datasets which can be trained with the machine.
5 Claims & 1 Figure
Brief description of Drawing
In the figure which are illustrate exemplary embodiments of the invention.
Figure 1, The Process of Weed Identification System , Claims:The scope of the invention is defined by the following claims:

Claim:
1. A system/method for making autonomous system for agriculture using artificial intelligence and WSN, said system/method comprising the steps of:
a) The system starts with the input image data collections (1), then the unwanted data should be removed from preprocessing (2). The images are converted into the array of storage (3).
b) The system has the one more way to process the input from the existing database (4), then CNN model (5) is trained based on the two comparison values.
c) Then the CNN classification will plays role for comparison (6), if output is matched with previous medical results (7), we can classify as weed or plant.
2. As mentioned in claim 1, The first step is to gather a weeds and plants dataset is collected from Kaggle and various types of weeds and plants are used as features and weeds plants are used as labels.
3. According to claim 1, before training data each image is converted to 224*224 size and re-scale image. Then data set is divided in to testing and validation dataset which is used for train and test.
4. As per claim 1, the stage VGG 19 model is initialized with required parameters and trained features and labels are fit to algorithm and model is saved to system.
5. According to claim 1, CNNs are particularly well-suited for image analysis of weeds and plants as images as input to web application and check with model to predict result.