Description:FORM 2
THE PATENTS ACT, 1970
(39 of 1970)
&
The Patent Rules, 2003
COMPLETE SPECIFICATION
(See sections 10 & rule 13)
1. TITLE OF THE INVENTION
AN AUTOMATED PIPELINE DEVICE FOR ANALYSIS AND INFERENCE OF UAV-BASED HYPERSPECTRAL IMAGES OF CROP WATER STRESS
2. APPLICANT (S)
S. No. NAME NATIONALITY ADDRESS
1 NMICPS Technology Innovation Hub On Autonomous Navigation Foundation IN C/o Indian Institute of Technology Hyderabad, Kandi, Sangareddy, Telangana– 502284, India.
2 Indian Institute Of Technology Hyderabad IN Kandi, Sangareddy, Telangana– 502284, India.
3. PREAMBLE TO THE DESCRIPTION
COMPLETE SPECIFICATION

The following specification particularly describes the invention and the manner in which it is to be performed.

FIELD OF INVENTION:
[001] The present invention relates to the field of hyperspectral imaging using UAV for water stress on crops. The present invention in particular relates to a device with toolbox for end-to-end data analysis of UAV-based crop hyperspectral images for crop early water stress detection or water stress classification.
DESCRIPTION OF THE RELATED ART:
[002] The current agricultural or plant-phenotyping practices have to be changed using technological advancements for sustaining climate change. Identification of crop early water stress is crucial for identifying drought stress and to take necessary measures. Also, mapping water stressed areas with varied stress levels in the field helps with precise irrigation, saves water and time. In plant phenotyping, scientists need to assess the performance of thousands of hybrid varieties under different irrigation treatments to assess their sustainability and performance under drought. These practices are done manually or visual to date, which has many disadvantages like time taking, costly, error prone, laboursome. UAV-based hyperspectral imaging (HSI) has the potential to accelerate these practices. Hyperspectral imaging sensors are capable of identifying early water stress by capturing the subtle changes in the canopy reflectance in hundreds of spectral channels. However, analysis of hyperspectral image data is a challenge for agriculturalists.
[003] Reference may be made to the following:
[004] IN Publication No. 201741025064 relates to a high resolution imaging of the water content on crops which is obtained by collecting the images of the crops through ASTER and LANDSAT-8 satellites. The next step is to Orthorectify the image in order to correct the geometric error caused by earth rotation. Furthermore, the obtained image is rectified with lens distortion errors and correct coloring is used for balancing the image pixels by correcting shadow pixels in an image, upon this the images are to be mosaicked for a better view on the crops. This is with thermal images captured from satellites.
[005] Present invention is on with hyperspectral imaging captured from UAV platform. The water stressed crops are classified based on the spectral reflectance changes observed in the hyperspectral images, however they do on temperature changes.
[006] IN Publication No. 202341069371 discloses an agricultural drone system equipped with multispectral imaging capabilities for comprehensive crop analysis. This talks about a drone equipped with multispectral camera, which has limited spectral information in 5-10 spectral bands.
[007] Present invention is hyperspectral, where rich spectral information is provided in hundreds of spectral bands. It uses AI and Machine learning based techniques for algorithm development, which are far efficient for analysis of complex HSI data. In addition, an end-to-end tool box is provided, which can be easily used by the agriculturist.
[008] Publication No. CN103018179 relates to a device and a method for polarization detection of crop water stress. This talks about the water stress detection at leaf level, also seems to be in the indoor environment.
[009] Present invention is on analysis of canopy or field level hyperspectral data using UAV. It can have large scale mapping crop fields for different levels of water stress.
[010] IN Publication No. 201613036100 discloses a farmer interactive remotely controlled rice crop disease detection system comprising a hand-held device with display and an unmanned aerial vehicle (UAV) with a camera and processor along with embedded image processing software. Camera at UAV captures the live video of the rice field and transmits it to handheld device using ZigBee Wireless Communication. This is about crop disease detection with RGB camera. RGB images are not accurate in identifying stress or disease at early stage due to practically no spectral information.
[011] However, present invention is about water stress detection, where we use hyperspectral data in hundreds of bands, which is very efficient in detecting early stress. It provides end-to-end data analysis toolbox, which can be used by any user for their data analysis.
[012] IN Publication No. 201841001375 discloses IoT enabled control system for crop management. The claimed ground based crop management, using moisture sensor, RGB camera for disease detection. This can’t be used for large field monitoring.
[013] Present invention is related to UAV borne hyperspectral imaging, which enable to manage and map large fields for crop water stress.
[014] IN Publication No. 202311032238 disclose a robust system for monitoring water use in rainfed and dryland farming using drones, comprising of: a drone including a plurality of sensors, high-resolution digital imaging, and artificial intelligence (AI) capabilities. This uses a drone and high resolution images, which have no spectral data. It does not provide any data analysis toolbox.
[015] Whereas, present invention is about analysis of UAV-borne hyperspectral images with rich spectral as well high spatial resolution for crop water stress detection.
[016] IN Publication No. 202141010955 relates to a system and computer implemented method for determining the vegetation index of a plant and to detect the growth of plants. The system comprising, an imaging system, a plurality of brushless DC (BLDC) motors, a UAV pilot controller, a UAV IoT controller and a computing means. This talks about the vegetation index based study for plant growth detection using camera.
[017] Present invention is concerns crop early water stress detection and varied drought stress classification with hyperspectral UAV images.
[018] IN Publication No. 202211073701 discloses an unmanned aerial vehicle for farming application. The unmanned aerial vehicle comprises of cross-configured drone to take-off upon receiving instructions from a ground control center or a transmitter for flying in a complete angle, wherein butterfly net-wings are attached to proximal ends of the drone, a serial communication controller configured with a GPS guidance unit to receive real-time coordinates from a pre-loaded trajectory to navigate drone, a remote sensing camera to monitor plants closely and take images/videos in a desired frame under stable flight condition configured to deliver image, audio, and video transmission to control center, a prediction unit adapting deep learning approach to diagnose images and detect diseases and to control flight via a user interface, user interface comprises of preloaded location, which sends information to a ground telemetry module and aircraft’s air telemetry module to avoid human interference during flight. This is with drone based RGB imaging for disease detection in farming.
[019] However, the present invention is an end-to-end data analysis toolbox for UAV-borne hyperspectral images for crop water stress classification.
[020] IN Publication No. 202341007553 relates to artificial intelligence is a technique that mimics machines as human behaviors. AI has been increasingly adapted as a part of agricultural industry evolution. Using AI we can perform predictive analysis to determine the type of disease, right time to sow seeds, time of yield and suggestions to the recurring diseases based on current scenario. In agricultural development, a diversification of interdisciplinary and variety of multifield technologies like Internet of Things (IoT), bioengineering and Artificial Intelligence can be applied for the improved results. This is about an AI/ML technique such as Mask R-CNN for finding healthy and unhealthy plants based on the RGB image data.
[021] Present invention is an end-to-end HSI data analysis toolbox which consists of various progressive stages, designed for UAV captured hyperspectral images for classification and mapping of crop water stress. It also provides waveband selection algorithm, pretrained models for crop water stress.
[022] Publication No. US2021289692 relates to a multi-scale habitat information-based method and device for detecting and controlling water and fertilizer for crops in seedling stage: performing fusion analysis on multi-scale features of the water and fertilizer stress of crops on the basis of crop canopy-scale three-dimensional laser scanning information, foliage-scale polarization-hyperspectral imaging information, and micro-scale micro-CT scanning information; combining the real-time feedback of the temperature, humidity, illumination and substrate moisture content within a crop growing greenhouse; by means of multi-information fusion modeling, comprehensively determining and feeding back the water and fertilizer stress of the crops as well as water requirement and fertilizer requirement information, and providing policy information for the amount of fertilization and irrigation. This is using ground-based platform, not with drone. This is specific to seedling stage, not generalized one Also, it uses data from many sensors which will be costly.
[023] However, the present invention is about data analysis of hyperspectral images captured from UAV platform. It is efficient in water stress detection at any growth stage of the crop.
[024] Patent No. US4876647 relates to an apparatus for determining the water stress condition of an agricultural crop growing in a field includes sensors for sensing environmental and crop conditions, such as air temperature, crop canopy temperature and relative humidity, and for generating signals indicative of the sensed conditions, a microprocessor for receiving the generated signals and for calculating a crop water stress index from the signals and a visual display for displaying, in the field, the calculated index. In one embodiment, the sensors are mounted in a pistol like, hand-held housing and the microprocessor, display and a keyboard control are carried by a second housing. The microprocessor compares one or more of the sensed conditions to reasonable value limits and rejects a set of measurements containing values beyond the limits. Crop-specific data needed to calculate water stress indices for a particular crop are stored in a programmable, read only memory. When data for a different crop are needed, the memory may be removed and replaced by a memory containing the appropriate crop data, or, if erasable, the memory may be erased and reprogrammed for the different crop. The apparatus preferably includes a memory for storing a number of calculated crop water stress indices, each calculated from conditions sensed on a different day, in records including the date of the measurements. The stored indices can be retrieved and displayed with their measurement dates so that historical trends of crop water stress may be discerned.
[025] Present invention is about data analysis for crop water stress detection/classification using UAV-based hyperspectral images.
[026] Publication No. CN106067169 provides a plant water stressed state automatic monitoring method and a plant water stressed state automatic monitoring system. The plant water stressed state automatic monitoring method includes the following steps that: the infrared image and visible light image of plants and soil moisture content information of the corresponding plants are obtained; image fusion information is obtained according to the infrared image and the visible light image; the water stressed state information of the plants is obtained according to the image fusion information and the soil moisture content information; and corresponding instruction information is emitted according to the water stressed state information of the plants, so that a water replenishment device can be made to enter a corresponding working state. This is a plant water stress automatic monitoring using a visible and an infrared image, and moisture sensor. Visible image provide only spatial information, and an infrared image provides data at a particular wavelength.
[027] Present invention is concerned with hyperspectral imaging, where reflectance information in hundreds of wavelengths is used, which captures the subtle reflectance changes due to water stress.
[028] Patent No. US6597991 relates to a method and system for evaluating the water stress status of growing crops in nearly real time employing remote monitoring of entire crop-growing areas, rather than sampling or spot checks, at sufficiently high resolution to recognize features in the crop-growing area, especially to distinguish between crop foliage and non-foliage features, and requiring minimal additional measurements of environmental parameters in the crop-growing area. This is a method and system for evaluating crop water stress using remote monitoring at real-time.
[029] Present invention is a device for analysis and inference of UAV-based hyperspectral images of crop early water stress detection and varied water stress level classification and mapping.
[030] Patent No. US4755942 relates to an apparatus and method for indicating whether irrigation of an agricultural crop is required based on the measurement of crop water stress. The apparatus and method include a plurality of sensors for sensing a plurality of conditions related to the agricultural area, including crop canopy and air temperatures, and relative humidity, from which the water stress condition of the crop can be determined, and for generating signals representative of each of the sensed parameters; an electronic device connected to the sensors receiving the sensor signals for determining therefrom the water stress condition of the crop; and an output device connected to the electronic device for visually displaying the calculated water stress index. This is an apparatus and method for indication of irrigation requirement of crop. It measures the water stress condition of crop using plurality of sensors like crop canopy and air temperatures, and relative humidity. It is in no way related to hyperspectral image data or its analysis.
[031] Publication No. CN116678862 relates to a plant chlorophyll fluorescence three-dimensional imaging device and method, and belongs to the plant chlorophyll fluorescence three-dimensional field. The invention provides a plant chlorophyll fluorescence three-dimensional imaging device and a plant chlorophyll fluorescence three-dimensional imaging method in order to solve the problems that a plant chlorophyll fluorescence three-dimensional imaging method in the prior art is complex to implement, multi-angle shooting is needed, and a large amount of data is needed for image synthesis. This is a plant chlorophyll fluorescence three dimensional imaging device and method for plant health.
[032] Present invention is about drone-based hyperspectral imaging data analysis toolbox for crop water stress detection and mapping.
[033] Publication No. CN116168287 relates to a tomato plant drought stress detection method based on hyperspectral imaging, which comprises the following steps: acquiring a hyperspectral image of a tomato leaf to be identified, and extracting reflection spectrum data of the leaf according to the hyperspectral image; utilizing a genetic algorithm to screen characteristic wavelengths, and determining an optimal reflectivity image set according to the correlation between the reflectivity images corresponding to the characteristic wavelengths; deep image features of the optimal reflectivity image set are extracted by using a convolutional neural network; and fusing the spectrum and image features of the leaf, and inputting the fused spectrum and image features into a trained plant drought stress identification model for identification to obtain the drought stress level of the tomato to be identified. This is drought stress detection in tomato plants with hyperspectral imaging, and convolutional neural networks.
[034] The present invention is a device with an optimized analysis toolbox for UAV-borne hyperspectral images for water stress classification/detection in any crop.
[035] Publication No. SA113340477 relates to plant stress varies as a direct result of lack of water or nutrition or infected plants with pests or the presence of heavy metals in irrigation water. Plant stress can be measured from spectral response pattern predicted using an Airborne Hyper Spectral Imaging Camera and another one fixed on a tractor covering spectral band from 300 nm to 2500 nm. Present invention relates to crop/plant water stress classification using UAV-based hyperspectral images. Present invention specifically an optimized end-to-end hyperspectral data analysis toolbox for crop water stress classification and mapping.
[036] Publication No. US2021345567 relates to determining plant stress, including: using a computer-based camera system having thermal imaging and visual imaging to capture foliage at close proximity of at least one plant to provide high resolution images/video thereof; analyzing both thermal and visual images/video therefrom to form a composite image; determining the thermal activity of the composite image/video and photosynthesis state of the at least one plant; and deriving the plant stress from the determination. This is about plant stress detection using thermal and visual imaging at close proximity from plant.
[037] However, the present invention relates to crop water stress detection and mapping across extensive agricultural fields utilizing UAV-based hyperspectral camera data.
[038] Publication No. US2017030877 relates to a multi-sensor device comprises a housing containing multiple sensor modules for capturing and transmitting sensor data for plants in a crop. This is neither related to crop water stress nor hyperspectral imaging.
[039] Publication No. WO2021165476 relates to a plant measurement system that is able to monitor plant and/or plant organ movements in real-time. It can operate a stand-alone. Moreover, it does not rely on camera technology or other imaging techniques, but uses small scale digital sensor technology. This is plant measurement system which can monitor plant and plant organ movements in real-time using small scale digital sensor technology without using imaging techniques.
[040] Publication No. US2018018537 relates to monitoring stress symptoms in plants is crucial to maximize crop productivity. This is a method and system for hyperspectral imaging to map detailed spatial distribution of plant chlorophyll content.
[041] Present invention is about device for analysis and inference of UAV-based hyperspectral images for crop water stress classification and mapping.
[042] Publication No. CN107392892 relates to an image-based automatic identification method for drought stress on corn at an earlier growth stage. This is a visual image-based automatic drought stress identification, which may not be accurate in detecting early stress.
[043] Present invention relates to water stress detection and mapping in any crop using UAV-borne hyperspectral images, which can more accurately detect early drought stress in crops.
[044] Publication No. CN106546567 relates to a plant drought stress diagnostic method and a device based on chlorophyll fluorescence imaging technology.
[045] Present invention relates to a device for crop early water stress detection with UAV-borne hyperspectral images, which is more accurate because of rich spectral information.
[046] Publication No. KR20160052368 relates to a method of diagnosing a response of a plant to abiotic stress or an herbicide using a thermal image which uses a thermal imaging camera to acquire a thermal image and undergo an image processing step to obtain a temperature of a plant. The abiotic stress diagnosing happens using the temperature changes in thermal images.
[047] Present invention relates to water stress detection using hyperspectral camera data based on the canopy reflectance changes due to water stress.
[048] Publication No. CN203178188 relates to a plant health condition detection device based on the spectral imaging technique. A blue LED and a white LED are arranged on each edge of an equilateral triangle-shaped support in an illumination unit, and every three LEDs in the same color form an equilateral triangle, so that uniform illumination can be realized; and the two kinds of LED lamps can be switched mutually under the control of a light source controller so as to excite plant fluorescence. This patent discusses plant health detection using multispectral cameras in the visible light range,
[049] Present invention relates to analysis of UAV-based hyperspectral imaging data with rich spectral information for early water stress detection and mapping in crops.
[050] The article entitled “A review of crop water stress assessment using remote sensing” by Uzair Ahmad, Arturo Alvino and Stefano Marino; Remote Sens.,13, 4155; October 2021 talks about novel methods for evaluating crop water stress and its correlation with certain measurable parameters, investigated using remote-sensing systems. Through an examination of previous literature, technologies, and data, we review the application of remote-sensing systems in the analysis of crop water stress. Initially, the study presents the relationship of relative water content (RWC) with equivalent water thickness (EWT) and soil moisture crop water stress. Evapotranspiration and sun-induced chlorophyll fluorescence are then analyzed in relation to crop water stress using remote sensing. Finally, the study presents various remote-sensing technologies used to detect crop water stress, including optical sensing systems, thermometric sensing systems, land-surface temperature-sensing systems, multispectral (spaceborne and airborne) sensing systems, hyperspectral sensing systems, and the LiDAR sensing system. The study also presents the future prospects of remote-sensing systems in analyzing crop water stress and how they could be further improved.
[051] The above article is a review on crop water stress assessment using remote sensing, which presented various remote-sensing technologies used to detect crop water stress, however, it doesn’t talk about UAV-based hyperspectral imaging data analysis for crop water stress detection/classification. Present invention is about a device with end-to-end data analysis toolbox for analysis and inference of UAV-based hyperspectral images for crop early water stress detection, varied drought stress levels classification, and mapping water stressed field areas.
[052] The prior art inventions describe various systems and methods for water stress detection, abiotic stress detection, or plant health monitoring. Some use RGB/visible imaging, others are based on multispectral imaging, and a few utilize fluorescence-based imaging techniques. Most of these inventions are ground-based, with some even designed for indoor environments. A few employ spectral data and hyperspectral imaging; however, these are typically close-range imaging techniques. None of these inventions mention systems/methods for analysis and inference of UAV-borne hyperspectral images for crop water stress.
[053] In order to overcome above listed prior art, the present invention aims to provide an automated pipeline device equipped with an optimized end-to-end data analysis toolbox for crop early water stress detection and different drought stress intensities classification using UAV-based hyperspectral imaging data. The end user only has to give the crop hyperspectral images as input, then the inferences, water stressed areas mappings will automatically appear at the output. From these inferences, the agriculturalists can make necessary decisions. The hyperspectral images can be fed to the device, and directly get the inferences needed by the user on the hand held device. In addition, the device is equipped with many pretrained AI/ML models, from which the user can choose and work for their crop hyperspectral data. Also, the toolbox is provided with band selection stage, using which the user can identify the most suitable or optimal wavebands for water stress detection for their crop variety.
OBJECTS OF THE INVENTION:
[054] The principal object of the present invention is to provide an automated pipeline device for analysis and inference of UAV-based hyperspectral images of crop water stress and its method thereof.
[055] Another object of the present invention is to provide efficient system and method for analysis and inferencing using many data processing stages.
[056] Yet another object of the present invention is to provide end-to-end automated pipeline with no human intervention.
[057] Still another object of the present invention is to provide a portable computing device in which UAV-based HSI data can be offloaded at the field directly and inferences can be made.
[058] Yet another object of the present invention is to provide hyperspectral band selection methods, which can select optimal few bands sensitive to canopy water stress for a crop.
[059] Another object of the present invention is to provide many pretrained machine learning models for crop water stress classification, from which the user can choose and work for their crop hyperspectral data.
[060] Yet another object of the present invention is to identify crop early water stress, varied drought stress intensities classification, and mapping water stressed areas in the agricultural field using UAV-borne hyperspectral images.
SUMMARY OF THE INVENTION:
[061] The present invention relates to an automated pipeline device for analysis and inference of UAV-based hyperspectral images of crop water stress and its method thereof. This is an end-to-end automated toolbox where various data pre-processing steps such as data calibrations like reflectance and geometric calibrations, then running analysis algorithms on data hyperspectral band selection, classification, and inferences generation, water stressed areas mappings in the field are automatically done.
[062] The device and the toolbox are easy to use. For a few crop varieties, the algorithms are already trained on crop water stress data. Provision of optimal end-to-end algorithms specific to analysis of UAV-based HSI data of crop water stress. Flexibility of choosing a variety of methods at each stage of the pipeline.
[063] The present invention utilizes UAV-based hyperspectral images for crop water stress detection, providing comprehensive spectral data across hundreds of bands. The method enables detailed analysis through AI/ML techniques, offering significant advancements over other patents that rely on thermal imaging, multispectral cameras, or RGB cameras with limited spectral information. The invention also includes an end-to-end data analysis toolbox designed explicitly for UAV-borne hyperspectral data, allowing for easy and efficient water stress classification and large-scale field mapping.
[064] The user inputs the hyperspectral images into the toolbox, and all processing is automatically handled, with inferences or water stress mapping generated at the output. Additionally, several ML/DL-based pre-trained models are provided that users can fine-tune with their data to make predictions, further enhancing the usability and effectiveness of the present solution in managing crop water stress. Also, hyperspectral band selection methods are provided for selecting optimal few bands sensitive to crop water stress, which in turn reduce computational complexity and improve the accuracy of the results.
BREIF DESCRIPTION OF THE INVENTION
[065] It is to be noted, however, that the appended drawings illustrate only typical embodiments of this invention and are therefore not to be considered for limiting of its scope, for the invention may admit to other equally effective embodiments.
[066] Fig.1a shows UAV mounted with a hyperspectral camera to collect crop images, which will be offloaded to the device.
[067] Fig.1b shows various data processing stages of an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress.
[068] Fig.2 shows inter connections or flow of different steps or operations in the toolbox.
[069] Fig. 3 shows a computing device like this on which the end-to-end HSI data analysis toolbox can be deployed.
DETAILED DESCRIPTION OF THE INVENTION:
[070] The present invention provides a device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, which consists of phases right from pre-processing to classification and mapping the field with stressed areas. Figure 1a shows the block diagram of a UAV mounted with a hyperspectral camera to collect crop images of water stress, which will be offloaded to the device for water stress classification and mapping. Figure 1b shows the block diagram of various data processing stages in the device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress comprising a camera (2) from UAV (1). This user-friendly toolbox (3) consists of various data preprocessing and analysis stages connected as shown in Figure 1, for efficient data analysis and obtaining accurate inferences. Figure 2 shows flow chart of different analysis operations according to the present invention. The HSI data can be fed to the device, and directly get the inferences needed by the user on the handheld device (4).
[071] Pre-processing phase:
[072] In this phase input is raw hyperspectral images captured from a camera from UAV in .tiff, .hdr, .mat, .npy or any other format. These raw images are not useful form of data for analysis, which consists of unwanted components such as environment and solar effects, distortions due to UAV-instability etc. From this to get useful form of data, radiance/reflectance and geometric calibrations have to be performed.
[073] There are provisions for these calibrations, with a click form the user, the data preprocessing such as radiance/reflectance and geometric calibrations are automatically done. For reflectance calibration, there is an option or check button to choose calibration spectrum or the existence of a reference calibration panel. Either user has the flexibility to provide the reference calibration spectra if available, or input the image of the reference calibration panel from which reference spectrum is extracted.
[074] The execution of radiance, reflectance and geometric calibration steps and feeding the corrected image to the next stage are automatically done by the toolbox. The toolbox is interfaced to Google Earth, where the calibrated images can be viewed on the agricultural field locations.
[075] Image quality assessment (IQA) phase:
[076] Images collected from UAV platforms contain various distortions due to motion, acquisition process etc. The quality of all the images may not be up to the mark to use for further data analysis. This phase uses state-of-the-art no-reference IQA methods for evaluating the quality of hyperspectral images. Based on the quality grade, if the image passes a quality check then it will be passed to the next stage, otherwise it will be discarded.
[077] Hyperspectral Image Denoising phase:
[078] In hyperspectral images, there is some inherent noise in many wavebands. This is due to the signal-to-noise ratio decreasing with narrow wavebands. The noise needs to be removed; otherwise, it will adversely affect the data processing. In the pipeline denoising stage - is incorporated to clean the images by removing the noise. The user has the flexibility to choose a suitable hyperspectral denoising method from the provided many methods.
[079] Hyperspectral Band Selection phase:
[080] The HSI data’s spectral redundancy and high dimensionality make the analysis complex regarding computation, space, and time. Band selection or feature selection reduces dimensionality by choosing a subset of crucial wavebands for a given end task, like water stress classification here. Focusing on specific spectral bands relevant to crop water stress reduces computational and space complexity and enhances detection accuracy by excluding irrelevant bands that other factors might influence. In the pipeline, the band selection phase is incorporated, where a variety of ML-based techniques such as ranking, clustering, searching, sparsity, embedding learning, and hybrid scheme-based band selection methods are provided. The user has the flexibility to choose a method, or the toolbox automatically selects a suitable one. The user can also choose to use all the bands or specific bands if water stress-sensitive bands are known prior.
[081] Classification & Mapping phase:
[082] After band selection, the classification phase is incorporated in the toolbox, which extracts useful features using which it classifies images into different water stress categories. The toolbox provides a variety of categories of AI/ML-based classification models such as support vector machine (SVM), random forest (RF), XGBoost, Deep neural network (DNN), 1D-CNN, 1D+2D CNN, 3D CNN, 3D+2D CNN, SpectralFromer, Vision transformers, etc. The user can choose a model, or the toolbox automatically selects an optimal model for the data. The trained models are used for inference generation and stress area mapping, and models can also be saved for later use.
[083] Various stages of the toolbox are organized and implemented into 3 phases, as depicted in Figure 2.
? Pre-processing phase: The pipeline interface will be in the way that we can give the raw image as an input. Radiometric Calibration and Reflectance Correction will be done internally without human intervention. There is a provision (a check button) to choose the mean spectrum from the calibration panel from the image or an available spectrum file. If the calibration panel is available on the image, then mean spectrum can be selected from the input image that is popped. If the file option is chosen then calibration spectrum file need to be uploaded. Then Geo-mapping and rectification will be done automatically which results in mapping rectified image information with the corresponding field/target location on the Google Earth application. We can view the calibrated image, and pixel intensities of the image and can save the image for further analysis.
? Train phase: The training phase was divided into the labelling phase and model training phase as only supervised training can be done, so data need to be labelled. If there is an existing dataset then directly prediction models can be trained by navigating to the model training phase. If the dataset is not available then labelling phase need to be chosen. First, Image Quality Assessment (IQA) will be done to the input pre-processed image. If the image is clean (IQA Score < threshold), then it is used directly, or if it is less noisy (IQA Score < upper-threshold), denoising will be done. If the IQA Score > upper-threshold considered as severely noisy effected image and the image will be discarded indicating that the image is not useful for processing. Denoising can be done to reduce noise and to improve the image quality. After Denoising stage, images will be popped and data samples extraction will be done. The data sample extraction can be done in two ways depending on the prediction model we employ. In 3D patch data samples extraction, the patches of size N x K x K x B (where, N: number of samples, K: patch dimension or width, B: number of bands) are extracte. Whereas, if we use pixel/spectral based methods then K=1 is chosen, implies spectral samples of dimension N x B are extracted. The extracted data sample patches are labelled with the corresponding water stress class. After labelling phase, through Data Appending phase we can append all the data samples (N) into the dataset. After creating the dataset, we can train the dataset through Classification phase. In this phase, Initially Band Selection to be done, which selects optimal few bands sensitive to water stress, which reduce computational complexity and improve the classification accuracy. We can select all the bands or can extract bands through any band selection method like ranking based, clustering based, and hybrid band selection methods provided in the toolbox. We can also use specific bands if water stress-sensitive bands are known prior. After Band Selection stage, hyperspectral Classification models are trained using the created datasets. There is provision to choose a classification method from the list of models presented there such as Support vector machine (SVM), random forest (RF), XGBoost, Deep neural network (DNN), 1D-CNN, 1D+2D CNN, 3D CNN, 3D+2D CNN, SpectralFromer, Vision transformer (ViT), etc. The pretrained water stress classification models are saved, which can be used later.
? Test phase: In the test phase, we can predict the water stress status of crop hyperspectral images or from the dataset if available. First the image is pre-processed as mentioned earlier, and IQA will be done to check its eligibility for next stages clean images (IQA Score < threshold) can be used directly, and noisy images (IQA Score < upper-threshold) undergo denoising. Severely noisy images (IQA Score > upper-threshold) will be discarded indicating that the image is not useful for processing. The user can choose the pretrained model in the training phase or from the list of existing pretrained models from the provided repository. Using the selected pretrained model, classification results will be obtained, i.e. either water stressed or not or the water stress severity levels as output. The qualitative and quantitative results in the form of performance metrics will be generated, and the output maps will be generated and displayed on the GUI, accordingly the agriculturists can take necessary actions. Certain functionalities like Restart, Abort, and Interchange among different phases are provided, which make the toolbox more user-friendly and easier to operate.
[084] The different stages of the UAV-based hyperspectral image data analysis pipeline (figure 1) for crop water stress classification and interconnections or flow of various steps in the toolbox (figure 2) are shown. Just by feeding raw images as input, the device can predict the water status of the crop. Figure 3 shows a computing platform that looks like this, in which the optimized toolbox is deployed, and all the data processing is done. A monitor/display can be connected to the device, on which GUI, results, and mappings can be displayed.
[085] Advantages
1. End-to-end automated pipeline which takes raw UAV hyperspectral images as input and outputs the results, and mappings.
2. Efficient for analysis and inferencing using many data processing stages like data calibrations, quality assessment, denoising, band selection, etc.
3. The device is portable. At the field directly the UAV-based HSI data can be offloaded to the device and inferences can be made.
4. It can do band selection, i.e., can select optimal few bands sensitive to canopy water stress for a crop.
5. It provides many pre-trained models on different crop varieties of water stress hyperspectral data, which can be customized, fine-tuned, and used for any crop data.
[086] Numerous modifications and adaptations of the system of the present invention will be apparent to those skilled in the art, and thus it is intended by the appended claims to cover all such modifications and adaptations which fall within the true spirit and scope of this invention.

, Claims:WE CLAIM:
1. A device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images (HSI) of crop water stress, which consists of phases right from pre-processing to classification and mapping the field with stressed areas comprises a UAV (1) with hyperspectral camera (2) which is a part of the pipeline toolbox (3) wherein HSI data is fed to the handheld computing device (4) on which the optimized toolbox is deployed, where all the data processing is automatically done and directly get the inferences needed by the user, and a monitor/display is connected to the device on which the GUI and results are displayed.
2. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, herein the in-Pre-processing phase, HSI raw images captured from a camera from UAV are subjected to radiance/reflectance and geometric calibrations and feeding the corrected image to the next stage are automatically done by the toolbox.
3. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein in the image quality assessment (IQA) phase, the quality of hyperspectral images are evaluated and if the image passes a quality check then it will be passed to the next stage, otherwise it will be discarded.
4. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein in the Hyperspectral Image Denoising phase, noise is removed from hyperspectral images by incorporating efficient denoising methods.
5. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein the toolbox is interfaced to Google Earth, where the calibrated images and water stress mapping can be viewed on the agricultural field locations.
6. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein using which crop early water stress detection, varied drought stress intensities classification, and mapping water stressed areas in the agricultural field using UAV-borne hyperspectral images can be carried out.
7. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein many pre-trained AI/ML models are provided for crop water stress classification, from which the user can select models and work for their data.
8. The device equipped with an optimal toolbox for analysis and inference of UAV-based hyperspectral images of crop water stress, as claimed in claim 1, wherein a band selection stage with variety of methods is provided, using which the user can find out the most suitable or optimal wavebands for water stress detection for their crop variety.