Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model

Abstract

In agriculture, promptly and accurately identifying leaf diseases is crucial for sustainable crop production. To address this requirement, this research introduces a hybrid deep learning model that combines the Visual Geometric Group Version 19 (VGG19) architecture features with the transformer encoder blocks. This fusion enables the accurate and précised real-time classification of leaf diseases affecting grape, bell pepper, and tomato plants. The system integrates a CSI camera (102) to capture field images (101), which are processed by the NVIDIA 128-core Jetson Nano GPU (103). The images undergo a reprocessing (105) pipeline that includes labelling, resizing, segmentation, and augmentation to enhance the dataset's quality. The data is then split (106) into training (70%), validation (20%), and testing (10%) datasets to ensure robust model development. At the core of the system is the DSC-TransNet model (107), which performs feature extraction (108) and selection (109) to identify diseases in the crops. The model is trained, validated, and tested (110), and the best version is selected for real-time classification (111). The evaluation ensures that the model is both accurate and generalizes well to new data. A 7-inch interactive display unit (104) provides an intuitive interface for visualizing the results of the disease detection process, allowing users to see whether crops are healthy or infected. This system leverages advanced Deep learning and image processing techniques to create a precise and scalable solution for crop disease management. By combining real-time image input, robust pre-processing, and state-of-the-art modelling, it offers an efficient tool for agricultural applications, improving crop health monitoring and fostering sustainable farming practices.

Specification

Description:Technical Field

The area of invention is related to plant pathology and agricultural technology, with a particular emphasis on automated methods for detecting crop diseases. For precise disease classification of leaves, it employs deep learning, computer vision, and real-time picture processing. By developing a more effective method of tracking the condition of plants in real time, this breakthrough hopes to improve sustainable crop production.

Background details

The background on the invention is intended to facilitate an understanding of the present invention. However, it should be appreciated that the discussion is not an acknowledgment or admission that any of the material referred to was published, known, or part of the common general knowledge in any jurisdiction as of the priority date of the application. The details provided herein the background if belongs to any publication is taken only as a reference for describing the problems, in general terminologies or principles or both of science and technology in the associated prior art.
The rising prevalence of plant diseases, the effects of climate change, and pest infestations have all contributed to a greater demand for environmentally friendly and economically viable farming methods in recent years. Rapid spread and substantial output loss caused by plant diseases, especially those affecting leaves, pose a danger to food security and the livelihoods of farmers. In order to keep crops healthy, decrease chemical treatments, and promote sustainable agricultural techniques, early identification and management of plant diseases are crucial.
In recent days, farmers are worried about diseases harming their plants in their agriculture farms. There are numerous systems available in the market to address said problem. Following are some of the problems noticed in the existing systems:
Farmers traditionally rely on visual inspections to detect diseases in their crops. This process can be time-consuming, especially in large fields or greenhouses. Additionally, human eyes might not catch subtle signs of diseases in the early stages.
If diseases like bacterial spots, leaf spot, and black measles, are not detected early, they can spread rapidly and damage a significant portion of the crop. Delayed detection can lead to substantial losses for farmers. Traditional methods of disease management involve applying pesticides or treatments across the entire field, even when only a portion of the crop is affected. This can be costly and potentially harmful to the environment. Modern agriculture benefits greatly from data-driven decision-making. Farmers need data to optimize their practices and maximize crop yields.
In light of the foregoing, there is a need for a system and method for real-time disease classification in grape and bell pepper plants that overcomes problems prevalent in the prior art
The prior art [IN202141003130] merely discloses the detection of diseases in leaves. A generic framework for rapid leaf disease detection is realized by its new architecture, which is based on deep learning models and transfer learning. It uses a supervised learning strategy for disease prediction and training, taking real-time pictures of leaves using satellite or remote sensing technology. The capacity to identify leaf diseases in a particular crop is a boon to technology-based agriculture. With little tweaks to the settings, it can identify leaf diseases in a wide range of crops because to its modular design. The innovation is compatible with all current PA applications that are utilized in the agriculture field.
The prior art [IN202341036734] merely discloses a method for pomegranate plant disease detection with deep learning models. This method automates the process of disease identification by utilizing deep learning and computer vision techniques. This allows for early detection and prompt intervention. The technique requires amassing a large dataset of pictures of pomegranate leaves, both healthy and diseased, with a range of disease types and severity levels represented. Following this, the pictures undergo preprocessing to standardize pixel values and improve data quality. In order to train a deep learning model-more precisely, a convolutional neural network (CNN)-to recognize pomegranate leaves in photos, transfer learning is employed. In order for the CNN to classify diseases, fully connected layers are inserted. Using an optimization approach, the trained model is fine-tuned before being tested on a new collection of photos of pomegranate leaves.
The prior art [IN202111055682] merely discloses deep convolutional neural network (CNN) models to detect and diagnose apple diseases from their leaves, capitalizing on the remarkable progress that convolutional neural networks (CNN) have made in machine vision. There is a large computational cost and a large number of parameters needed when CNN models are built from the ground up. As a result, we sped up training and reduced computational cost by using typical CNN pre-trained networks, but with optimization. The adopted models were educated using a public dataset that comprised three distinct apple disease classes and a single healthy class.
The prior art [IN202241039603] merely discloses an IoT system that uses deep learning models to detect, identify, and alert users to plant diseases. Utilizing Deep Learning algorithms for agricultural disease identification is the main emphasis of the idea. It is the job of the predictive unit to foretell when plant diseases would strike. If there is a difference between the plant's growth and its yield, the Internet of Things (IoT) device will notify the farmers.
Results from the prior art search show that deep learning has been successfully used to detect plant diseases in a variety of crops and using different methods. In order to monitor plant diseases, presents an Internet of Things (IoT) system that uses deep learning to foretell when diseases may spread and notify farmers in advance. Taken together, these techniques highlight how AI is becoming more important in precision agriculture for both targeted and broad disease detection.

Summary of the Invention

For large-scale farming operations in particular, the labor-intensive, time-consuming, and expensive hand inspection of plants by agricultural specialists is a common traditional method for disease identification. In addition, smallholder farmers in rural regions may not have access to agricultural specialists, making physical inspection impractical. Artificial intelligence (AI) and machine learning (ML) approaches, particularly deep learning models, are being investigated more and more as potential automated solutions to these problems; these algorithms can accurately identify plant illnesses from photos.
When it comes to picture classification, deep learning models-specifically CNNs such as the VGG19 architecture-have been absolutely crushing it. New developments in transformer models for natural language processing, however, have increased performance in capturing data's complicated linkages. A potential option to improve the accuracy and speed of recognizing leaf illnesses in plant disease detection is to integrate the strengths of CNNs and transformers.
An accurate hybrid deep learning model for real-time disease detection in crops including tomato, bell pepper, and grape plants is the goal of this project. The model will be built by combining VGG19 architecture with transformer encoder blocks. The suggested system implements data augmentation, real-time feedback mechanisms, and image processing pipelines to provide farmers with scalable, resilient, and field-applicable disease management. Possible benefits of this novel method include improved crop health monitoring, faster reactions to disease outbreaks, and more sustainable farming techniques.
This invention presents a state-of-the-art method for detecting agricultural diseases. It uses a hybrid deep learning model that combines VGG19 characteristics with transformer encoder blocks to accurately classify leaf diseases in real-time. Images of tomato, bell pepper, and grape plants are captured and processed by the system using an NVIDIA Jetson Nano GPU and a CSI camera. Data quality is improved by a thorough pretreatment pipeline, while illness detection and feature extraction are carried out by a DSC-TransNet model. For strong model performance, the data is partitioned into training, validation, and testing sets. A user-friendly 7-inch display gives them the results of disease categorization in real-time. Supporting sustainable agriculture practices, this integrated system provides an accurate, efficient, and scalable solution for crop health monitoring.
For sustainable crop management in agriculture, it is vital to diagnose leaf diseases quickly and accurately. This study introduces a hybrid deep learning model for accurate, real-time leaf disease classification in tomato, bell pepper, and grape plants. The model integrates VGG19 features with transformer encoder blocks. An NVIDIA 128-core Jetson Nano GPU (103), in conjunction with a CSI camera (102), processes field photos (101).
Labeling, scaling, segmentation, and augmentation are all steps in the preprocessing (105) procedure that images go through in order to make the dataset better. To guarantee the model's robust performance, the data is divided into three sets: training (70%), validation (20%), and testing (10%). The DSC-TransNet model (107), which is at the heart of the system, is responsible for accurately identifying diseases through feature extraction (108) and selection (109).
After the model has been trained, validated, and tested (110), the one with the best performance is selected for use in real-time classification (111). Verification and flexibility to fresh data are hallmarks of the assessment procedure. Viewing disease detection findings, which show whether crops are healthy or sick, is made possible by an interactive7-inch display unit (104).
This system offers a scalable, accurate solution for crop disease management by integrating advanced deep learning with image processing. It supports effective crop health monitoring and promotes sustainable farming methods.

BRIEF DESCRIPTION OF THE ACCOMPANYING DRAWING

The utility model Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model of the present disclosure will now be described with the help of the accompanying drawing, in which:
Figure 1 illustrates a perspective view of Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model

LIST OF REFERENCE NUMERALS

101 - Field images
102 - CSI camera
103 - NVIDIA 128-core Jetson Nano GPU
104 - Display unit
105- Preprocessing
106 - Data
107 - DSC-TransNet model
108 - Feature extraction
109 - Selection
110 - Training, validation, testing
111 - Classification

DETAILED DESCRIPTION

A principal aim of the present invention is developed with the deep learning model that combines VGG19 architecture with transformer encoder blocks to enable precise and real-time classification of leaf diseases in grape, bell pepper, and tomato plants, thus supporting sustainable crop production.
A further aim of the invention is to recycle the used solar panel in an appropriate manner. Used solar panels can be recovered or recycled using a variety of methods, some of which are known as applications. Methods may involve inspecting a previously used solar module for defects, either on its own or in conjunction with data (such as that which is included in a received shipment) pertaining to the module's characteristics, including but not limited to panel size, width, length, height, glass thickness, and so on. At one or more stages of refurbishing or recycling, discarded solar modules may undergo different cleaning procedures according to particular applications.
The Camera Serial Interface (CSI) camera (102) is responsible for capturing real-time, high-quality images of crop leaves in the field.
The field photos (101) refer to the raw images captured by the CSI camera in the agricultural setting. These images contain visual data of leaves from various crops, specifically grape, bell pepper, and tomato plants.
The NVIDIA Jetson Nano GPU (103) is a compact and powerful 128-core GPU designed for AI-driven edge computing, allowing real-time data processing and machine learning model inference directly on-site.
Pre-processing (105) is a crucial step that prepares the raw field images for model training and testing.
Data Splitting (106) splits the data into three sets to ensure a balanced and effective training process Training Set (70%): Used to train the model by allowing it to learn the distinguishing features of diseased and healthy leaves. Validation Set (20%) helps in fine-tuning model parameters and monitoring its performance to avoid overfitting. Testing Set (10%) provides an independent dataset to evaluate the final model's accuracy and ensure it performs well on new, unseen data.
DSC-TransNet Model (107) is a hybrid deep learning model specifically designed for leaf disease detection. It combines the Visual Geometric Group Version 19 (VGG19) architecture, which is adept at image feature extraction, with transformer encoder blocks that enhance the model's capability to understand complex patterns.
Feature Extraction (108) allows in extracting high-level features or patterns from images that correspond to different diseases.
Feature Selection (109) allows in selecting the most relevant features that aid in accurately identifying the disease, improving model efficiency and precision.
Model Training, Validation, and Testing (110) allows in training the DSC-TransNet model with the labeled dataset, adjusting it based on validation set performance, and evaluating its accuracy with the test set.
Real-Time Classification (111) selects the best-performing version of the DSC-TransNet model after evaluating its accuracy and efficiency.
Interactive 7-Inch Display Unit (104) provides a user-friendly interface for farmers or operators to view real-time disease detection results. It visually indicates whether the crops are healthy or infected, making it easy for users to interpret the system's output and respond accordingly.
, Claims:1. Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model comprising of:
- A CSI camera (102) configured to capture field images (101) of plants, including grape, bell pepper, and tomato; NVIDIA Jetson Nano GPU (103) for processing the captured images in real-time; preprocessing pipeline (105) for preparing the images, which includes steps of labeling, resizing, segmentation, and augmentation to enhance the quality of the dataset;
- A hybrid deep learning model that combines VGG19 architecture features with transformer encoder blocks to form a disease classification model, the hybrid model comprising a DSC-TransNet model (107) for feature extraction (108) and feature selection (109), used to classify the disease state of the crops;
- A system for splitting the preprocessed image data (106) into training, validation, and testing datasets, where 70% is allocated for training, 20% for validation, and 10% for testing; a model training and evaluation system (110) to select the best-performing model for real-time disease classification (111); A 7-inch interactive display unit (104) for visualizing the classification results and indicating whether the plants are healthy or diseased.
2. Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model as claimed in claim 1, wherein the preprocessing pipeline further comprises image augmentation techniques to improve model robustness and ensure the data is suitable for deep learning-based classification.
3. Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model as claimed in claim 1, wherein the DSC-TransNet model performs real-time classification with a high degree of accuracy by utilizing transformer-based feature selection for enhanced disease identification.
4. Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model as claimed in claim 1, wherein the NVIDIA Jetson Nano GPU (103) processes the image data for real-time disease detection, ensuring scalability and field-level deployment.
5. Real-Time Leaf Disease Detection in Bell pepper and Grape Using a Hybrid Deep Learning Model as claimed in claim 1, wherein the interactive display unit (104) provides an intuitive user interface to display real-time classification results, enabling on-site agricultural decision-making for disease management.

Documents