Optimizing Rice Leaf Disease Diagnosis: Enhanced Image Processing and Lightweight CNN Model

Problem Definition

After reviewing the literature, it is evident that there are significant limitations and challenges in the existing machine learning and deep learning approaches used for detecting diseases in rice leaf plants. One major issue is the lack of effective techniques for removing noisy data from the dataset images, leading to poor image quality and subsequently impacting the accuracy of disease classification models. Additionally, the complexity and time-consuming nature of traditional disease detection models, compounded by the absence of feature selection techniques, result in the curse of dimensionality. Furthermore, the manual collection of data overlooks crucial aspects such as lighting conditions, occlusion, backdrop color, and image quality, which are essential for accurate detection. Moreover, most existing models can only detect one or two diseases, limiting their utility in real-world applications.

The inability of previous models to differentiate characteristics effectively, due to low image quality and color similarities, further hampers the classifiers' ability to learn and accurately classify diseases. These limitations collectively hinder the efficacy and accuracy of traditional disease detection models, highlighting the urgent need for an improved model that addresses these shortcomings.

Objective

The objective of the research is to develop an improved disease detection model for rice leaf plants by utilizing Contrast Limited Adaptive Histogram Equalization (CLAHE) and a light weighted Convolutional Neural Network (CNN) approach. The aim is to enhance the classification accuracy and reduce the complexity of traditional models by effectively removing noisy data from images and extracting Region Of Interest (ROI) using hybrid segmentation techniques. Through the application of CLAHE and segmentation methods, along with a light weighted CNN classifier, the research seeks to address the limitations of existing models and improve the accuracy and efficacy of disease detection in rice leaf plants.

Proposed Work

In order to overcome the limitations of the conventional rice leaf disease detection models, an effective and highly accurate disease detection model is proposed in this research that is based on Contrast Limited Adaptive Histogram Equalization (CLAHE) and Light weighted CNN models. The main objective of the proposed approach is to reduce the complexity and enhance the classification accuracy rate of rice leaf disease detection models so that Region Of Interest (ROI) is retrieved effectively. To combat this task, initially, a publicly accessible dataset of 5602 images is taken from Kaggle.com. since, the images present in the selected dataset are raw and contain a lot of noisy data that must be eliminated.

To do so, Contrast Limited Adaptive Histogram Equalization (CLAHE) technique is implemented in the proposed work, that not only improves the quality of the images by correcting the light and contrasting conditions but also enhance the edges of the images. The primary idea behind CLAHE is to use interpolation to rectify irregularities across borders while completing histogram equalization of non-overlapping sub-areas of the picture. Moreover, in order to obtain the Region of Interest (ROI) effectively from processed images, a hybrid segmentation technique based on HSV and K-means segmentation is also used in the proposed work. The HSV segmentation technique converts the processed image into the three components of Hue, Saturation and Value along with their specific range. After this, K-means segmentation is applied on HSV segmented images which further improves the quality of images and helps in extracting the region of interest (ROI) more effectively and accurately.

Moreover, a light weighted CNN classifier is also used in the proposed work for classifying and categorizing images. The processed and segmented images are subjected to the light weighted CNN model wherein images undergo through five layers for categorizing the given image as healthy or disease infected.

Application Area for Industry

This project can be applied in various industrial sectors such as agriculture, food processing, and crop management. In agriculture, the proposed solutions can help in effectively detecting diseases in rice leaf plants, thereby enabling farmers to take timely actions to prevent the spread of diseases and ensure healthy crop yield. In the food processing industry, the accuracy of disease detection models can aid in quality control and ensuring the production of disease-free products. Additionally, in the domain of crop management, the improved classification accuracy rate can assist in efficient monitoring and management of crop health. The proposed solutions address specific challenges faced by industries, such as poor image quality, complexity of traditional disease detection models, and limitations in recognizing disease characteristics.

By implementing Contrast Limited Adaptive Histogram Equalization (CLAHE) and a light weighted CNN classifier, the project aims to enhance the quality of images, reduce complexity, and improve classification accuracy. This, in turn, benefits industries by providing more accurate disease detection, efficient region of interest retrieval, and streamlined monitoring processes, ultimately leading to improved crop health and higher productivity.

Application Area for Academics

The proposed project can enrich academic research by offering a novel and effective solution to the limitations faced by traditional rice leaf disease detection models. By incorporating advanced techniques such as Contrast Limited Adaptive Histogram Equalization (CLAHE) and a light weighted CNN classifier, the project aims to enhance the classification accuracy and reduce complexity in detecting diseases in rice leaf plants. This research has the potential to contribute to the field of image processing and machine learning by providing a more accurate and efficient method for disease detection. The utilization of CLAHE for image enhancement and the hybrid segmentation technique for extracting the Region of Interest (ROI) demonstrates innovative approaches to improving the quality of images and classifying them accurately. Academically, this project can serve as a valuable resource for researchers, MTech students, and PhD scholars working in the field of agricultural technology, image processing, and machine learning.

They can leverage the code and literature of this project to enhance their own work and explore new possibilities for disease detection models in agricultural settings. The relevance of this project lies in its potential applications in agricultural research, education, and training. By developing a more accurate and efficient disease detection model for rice leaf plants, researchers and students can gain insights into new methodologies for analyzing plant health and improving crop yield. In terms of future scope, the project could be expanded to cover a wider range of plant diseases and incorporate additional features for data analysis and visualization. By continuously refining and updating the model, researchers can further enhance its performance and applicability in real-world agricultural scenarios.

Algorithms Used

The proposed approach in this research utilizes Contrast Limited Adaptive Histogram Equalization (CLAHE) to enhance image quality and improve edge detection in the dataset of 5602 rice leaf images. CLAHE corrects lighting and contrast issues in the images. A hybrid segmentation technique, combining HSV and K-means segmentation, is applied to extract the Region of Interest (ROI) effectively. The HSV segmentation breaks down the image into its components, while K-means segmentation further refines the image quality. Additionally, a light weighted CNN model is employed to classify the images as healthy or diseased, with the images passing through five layers for accurate categorization.

Keywords

SEO-optimized keywords: rice crop disease detection, plant disease detection, agricultural imaging, deep learning, lightweight deep learning architecture, convolutional neural networks, crop health monitoring, plant pathology, image classification, feature extraction, disease identification, crop disease management, precision agriculture, agricultural robotics, agricultural technology, Contrast Limited Adaptive Histogram Equalization (CLAHE), Region Of Interest (ROI), noisy data removal, K-means segmentation, HSV segmentation, light weighted CNN classifier, disease classification model, accurate disease detection, feature selection technique, curse of dimensionality, traditional disease detection models, improved disease detection model.

SEO Tags

rice crop disease detection, plant disease detection, agricultural imaging, ML approaches, DL approaches, disease detection model, noisy data removal, image quality improvement, feature selection, curse of dimensionality, data collection, disease recognition, accuracy rate, classification model, CLAHE, Contrast Limited Adaptive Histogram Equalization, ROI, Region of Interest, dataset, Kaggle, image processing, segmentation technique, HSV segmentation, K-means segmentation, light weighted CNN, image classification, healthy vs diseased plants