Rice Leaf Disease Recognition using CNN

Abstract

The occurrence of bacterial, viral, and fungal diseases on rice leaves significantly hampers rice production, posing a challenge to meet global demand for the staple crop. While the detection of rice leaf diseases is crucial, existing methods are constrained by limitations in image backgrounds and capture conditions. Convolutional Neural Network (CNN) models have emerged as a usefull avenue for disease recognition in rice leaves, yet current approaches suffer from decreased recognition rates when applied to independent datasets and are constrained by the need for large-scale network parameters. In this project, we propose an innovative CNN-based model aimed at mitigating these challenges by reducing network parameters. Through training multiple CNN- based models to identify three common rice leaf diseases, our study aims to showcase the effectiveness and superiority of our approach compared to state-of-the-art CNN-based models for rice leaf disease recognition.

Introduction

I. INTRODUCTION

India's agricultural sector holds paramount importance in the nation's economy, given the substantial number of people engaged in this field. Crop production stands as a pivotal factor influencing domestic market dynamics, with agricultural enterprises continually seeking innovative, high-yield solutions amidst population growth, fluctuating weather patterns, and political uncertainties. The vitality of plant health in ensuring food security and agricultural sustainability cannot be overstated. However, plants are susceptible to various diseases, posing significant social and economic challenges. Crop ailments can impede growth, development, yield, and quality, constituting a primary cause of productivity decline. Early detection of diseases is imperative to prevent soil contamination, necessitating the judicious use of pesticides from the onset of infection. Rice, as a staple food for a substantial segment of the global population, plays a crucial role in ensuring food security worldwide. However, the productivity and quality of rice crops are significantly threatened by various diseases affecting their leaves. These diseases not only hinder crop growth and development but also reduces the crop yield and quality, posing substantial challenges to agricultural sustainability and food production.

Manual detection and diagnosis of rice leaf diseases have traditionally been labor-intensive and error-prone, often leading to delayed interventions and increased crop losses. Recognizing the pressing need for accurate and timely disease detection in rice plants, recent advancements in deep learning techniques, specifically Convolutional Neural Networks (CNNs), offer a promising solution.

CNNs have emerged as powerful tools in image detection, classification tasks, leveraging their ability to automatically study hierarchical features directly from raw pixel data. By training CNN models with large datasets containing images of healthy and diseased rice leaves, researchers can harness the inherent capabilities of these networks to recognize intricate patterns indicative of specific diseases with remarkable accuracy. The integration of CNNs into rice leaf disease recognition systems introduces a paradigm shift in agricultural practices, enabling automated and efficient detection of diseases. This approach not only smooths the detection process but also facilitates timely interventions, allowing farmers to implement targeted management strategies to mitigate crop losses effectively. Moreover, CNN-based disease recognition systems offer scalability and adaptability, making them suitable for deployment across diverse agricultural settings and geographical regions. By harnessing the potentiality of deep learning, these systems can contribute significantly to improving agricultural productivity, enhancing food security, and promoting sustainable farming practices. In this paper, we delve into the province of rice leaf disease recognition using CNNs, exploring the methodologies, techniques, and advancements in this field. Through comprehensive research and experimentation, we aim to play a part in to the development of robust and reliable CNN- based solutions for automated disease detection in rice plants, ultimately fostering a more resilient and productive agricultural ecosystem. Finally , the deployed CNN model can automatically analyze images of rice leaves and classify them as healthy or diseased, enabling timely intervention and management strategies to multiple crop losses. Traditional strategy of disease detection often fall short in accurately identifying these ailments, necessitating more efficient and reliable solutions. In this study, we propose to explore the potential of CNN-based models for rice leaf disease recognition, aiming to evolve a robust and effective tool for enhancing agricultural productivity and ensuring food security.

II. LITERATURE SURVEY

Here's a survey of various techniques used to optimize AES implementations for reduced area footprint:

A. Automatic Identification of Peanut-Leaf Diseases Based on Stack Ensemble

This paper addresses the critical issue of automatically identifying diseases in peanut leaves, which directly impacts crop yield and quality. It employs a hybrid approach, combining traditional machine learning with deep learning models, to enhance disease detection accuracy. Various peanut leaf diseases are considered, including rust, leaf-spot, scorch, and combinations thereof. Three data augmentation methods - image flipping, rotation, and scaling - are utilized to improve model performance. Results reflects that the deep learning model outperforms traditional methods, especially when integrated with data augmentation and ensemble techniques. After ensemble by logistic regression, ResNet50 achieves 97.59% accuracy, and DenseNet121 reaches a 90.50% F1 score, demonstrating significant improvements. Deeper neural network architectures, such as ResNet50 and DenseNet121, exhibit superior performance. All-in conclusion, this study offers valuable insights into peanut leaf disease identification, providing a reference for agricultural practices.

B. Identification and Recognition of Rice Diseases and Pests using Convolutional Neural Networks

This paper presents a solution for precise and timely recognition of diseases and pests in rice plants, crucial for mitigating economic losses in agriculture. Leveraging deep learning-based convolutional neural networks (CNNs), the study introduces two main contributions. Firstly, it fine-tunes state-of-the-art large-scale architectures like VGG16 and InceptionV3 for detecting rice diseases and pests, demonstrating effectiveness with real datasets. Secondly, to address limitations for mobile devices, the paper proposes a two-stage small CNN architecture. Comparative analysis with memory-efficient CNN architectures highlights the proposed model's ability to achieve a desirable accuracy of 93.3% while significantly reducing model size. These findings comes up with valuable insights for developing efficient disease and pestdetection systems tailored for agricultural applications.

C. Groundnut leaf disease Detection and Classification by using Back Propagation Algorithms

The quality of agricultural products often suffers due to disease attacks, primarily caused by fungi, bacteria, and viruses. Cercospora, a common leaf disease in groundnuts, exemplifies this issue. To combat such challenges, a four-step upgraded processing pattern is implemented. Initially, color renovation is applied to the RGB image, followed by conversion to HSV for improved color representation. Subsequently, plane separation and extraction of color features are conducted. Finally, disease detection is accomplished using the backpropagation algorithm. This approach streamlines disease identification, aiming to enhance agricultural product quality.

D. Plant disease Detection using Machine Learning

This paper addresses the significant challenge of crop disease detection, particularly in regions lacking necessary infrastructure. Leveraging advancements in leaf-based image classification, the study utilizes Random Forest as a technique to distinguish between healthy and diseased leaves. The proposed methodology comprises dataset creation, feature extraction using Histogram of Oriented Gradients (HOG), training the classifier, and classification. By collectively training datasets of diseased and healthy leaves, the Random Forest classifier accurately identifies disease presence in plants. Overall, employing machine learning on publicly available large datasets offers a scalable solution for detecting plant diseases, contributing to food security efforts on a massive scale.

E. Pesticide Suggestion and Crop Disease classification using Machine Learning

Crop cultivation is vital in agriculture, but food loss due to infected crops significantly impacts production rates. Detecting plant diseases early remains unexplored, posing a challenge to reducing pesticide use while improving yield and quality. A proposed system leverages enhanced Machine Learning for early leaf disease prediction, aiming to identify infected areas promptly. A color-based segmentation model categorizes the infected regions efficiently. Experimental analyses assess time complexity and the area of infection. Image processing techniques facilitate disease detection, involving steps such as image acquisition, pre-processing, segmentation, feature extraction, and classification.

F. Pest and disease detection for corn field using image analysis

This proposed system addresses the challenges faced by farmers in monitoring and managing crop pests and diseases, exacerbated by irregular climatic patterns and environmental issues.

Utilizing image processing strategy, such as Texture- based Segmentation and Simple Linear Iterative Clustering (SLIC), the system captures and analyzes periodic images of agricultural fields to identify pests and diseases at early stages. Features extracted from segmented images are used for classification, employing Binary Support Vector Machine (BSVM) and Multi-class Support Vector Machine (MSVM). Additionally, the system recommends suitable pesticides and chemicals for crop protection, considering the type of pest/disease detected and providing guidance on usage methods and quantities. By offering early detection and tailored recommendations, this system aims to mitigate crop damage and optimize yield while reducing reliance on manual monitoring and excessive chemical usage.

G. Prediction of Rice Diseases Using Convolutional Neural Network.

Efficient management of pests and diseases in rice cultivation is pivotal for optimizing crop yields. With the aid of modern technologies like smartphones, farmers now have the ability to swiftly detect and identify various rice field pests and diseases. By employing Convolutional Neural Networks programmed in R language, the analysis of images depicting diseased leaves becomes more precise and efficient. These images, sourced from the UCI Machine Learning Repository, cover three primary rice diseases: Brown Spot, Leaf Smut and Bacterial Leaf Blight. This innovative approach holds promise for early disease detection and tailored management strategies, ultimately leading to improved rice crop productivity.

H. Utilizing deep convolutional neural networks for the recognition and diagnosis of diseases in rice crops."

This research introduces an innovative approach for automatically detecting and diagnosing diseases in rice plants, leveraging deep convolutional neural networks (CNNs). Through training these networks on a dataset containing 500 diverse images of both diseased and healthy rice foliage, the model achieves an impressive accuracy of 95.48% using a 10- fold cross-validation technique. This surpasses the performance of traditional machine learning methods by a considerable margin. The findings from the simulations underscore the viability and efficacy of employing CNNs for identifying rice diseases, highlighting their promise in advancing agricultural information systems and refining crop management strategies.

I. Application of deep learning in image processing for diagnosing diseases and identifying pests in cotton leaves

This research aims to improve the detection of cotton leaf diseases and pests in Ethiopia, which is crucial for the country's cotton industry. Employing deep learning, specifically Convolutional Neural Networks (CNNs), the study developed a model capable of identifying prevalent cotton leaf issues such as bacterial blight, spider mite, and leaf miner. Through a K-fold cross-validation technique and a dataset containing nearly 2400 samples (600 images per category), the CNN model achieved an impressive accuracy of 96.4%. Developed using Python with Keras and TensorFlow backend, and implemented in Jupyter, the model indicates potential for real-time applications. This underscores the significance of IT-driven solutions in complementing traditional approaches to disease and pest detection in agriculture..

J. Detection and Classification of Leaf Diseases in Maize Plant using Machine Learning MSc Research Project Data Analytics,

Rice farming, integral to Indian agriculture, faces significant challenges from leaf diseases, impacting both yield and quality. Common ailments like Brown Spot, Leaf Blast, and Hispa exacerbate these issues. To address them, researchers have explored various machine learning and deep learning techniques for accurate disease detection. Assessment metrics like accuracy, recall, and precision were employed, highlighting the superior performance of deep learning models over traditional methods. Notably, the 5-layer convolutional model stood out, achieving an impressive 78.2% accuracy, surpassing alternatives like VGG16, which reached 58.4%. This study provides valuable insights for farmers, enabling timely disease detection and fostering healthier crop yields and economic resilience.

III. EXISTING SYSTEM

A. Disease Detection And Identification Of Rice Leaf Based On Improved Detection Transformer

The DHLC-DETR algorithm is introduced for detecting rice leaf diseases, aiming to overcome limitations such as sparse samples, high variability, and small affected regions. It employs the DHLC-FPN network to replace the backbone of DETR, enhancing feature extraction capabilities within the feature pyramid while mitigating issues related to information loss in convolutional networks. Experimental findings demonstrate that DHLC-DETR significantly improves small target detection, outperforming DETR in terms of mAP and recall. To tackle the challenge of limited sample size, image enhancement and transfer learning techniques are utilized to enhance algorithm generalization and detection performance.

IV. PROPOSED METHODOLOGY

In this project, we aim to develop a system that recognizes and predicts the rice leaf diseases. This idea useful in agriculture for farmers to recognize the diseases early and take the precautions for the recognized the diseases. Our project will be divided into few stages

A. Data Collection

Gathering data for rice leaf disease recognition using CNN involves assembling a diverse dataset containing images of rice leaves affected by different diseases alongside images of healthy rice leaves. These images can be sourced from field surveys, agricultural research centers, or publicly accessible datasets. It's crucial to ensure that the dataset encompasses a broad spectrum of disease symptoms and severity levels to enable robust model training. Additionally, metadata like disease labels, image resolutions, and environmental conditions may be collected to enrich the dataset and provide contextual information for analysis. Quality assurance procedures, including image annotation and validation, should be implemented to guarantee the accuracy and reliability of the dataset, thus enabling effective training and evaluation of the CNN model for rice leaf disease recognition.

B. Data Preprocessing and Splitting

The collected dataset undergoes preprocessing to enhance its quality and facilitate effective training of the CNN model. Preprocessing steps may include resizing images to a uniform size, normalization to standardize pixel values, and augmentation to increase dataset diversity. The dataset is divided into training, validation, and testing sets. Typically, the majority of the data (e.g., 70-80%) is used for training, while smaller portions are allocated for validation (e.g.,10-15%) and testing (e.g., 10-15%) to evaluate model performance.

C. Model Training

The splitted data will involve in the evolution and training using Conventional neural network (CNN) for predicting diseases of the rice leaf.

D. Conventional Neural Network

Convolutional Neural Networks (CNNs) are a class of deep neural networks specifically employed for processing structured grid data, such as images. They are composed of multiple layers, including convolutional layers, pooling layers, and fully connected layers, that work together to automatically learn hierarchical representations of the input data.

In a Convolutional Neural Network (CNN), there are typically several types of layers stacked together to form the network architecture. Here are the common layers found in a CNN:

1. Input Layer: This layer represents the raw input data, such as an image or a sequence of words in natural language processing tasks.
2. Convolutional Layers: These layers apply convolutional filters (also known as kernels) to the input data. Each filter detects specific patterns or features in the input data by performing element-wise multiplication and summing operations. Multiple filters are typically applied in parallel to extract different features from the input.
3. Activation Layers: Activation functions, like ReLU (Rectified Linear Unit), are employed individually on the output of convolutional layers to introduce nonlinear characteristics into the network. This facilitates the learning of intricate connections between input and output data.
4. Pooling Layers: Pooling layers downsample the feature maps produced by the convolutional layers, reducing their spatial dimensions while retaining important information. Common pooling operations include max pooling and average pooling.
5. Fully Connected Layers (Dense Layers): These layers connect every neuron from the previous layer to every neuron in the subsequent layer, allowing the network to learn high- level representations of the input data. Fully connected layers are typically used in the final stages of the network for classification or regression tasks.
6. Output Layer: The output layer in a CNN is the end layer of the network, responsible for producing the desired output of the diseases of the rice leaf according on the task being performed.

These layers are typically stacked together in a sequential manner to form the overall CNN architecture. The architecture and layer count of the Convolutional Neural Network (CNN) vary according to the specific task and the complexity of the data being analyzed.

E. Validation

The validation process for assessing the accuracy and dependability of Convolutional Neural Networks (CNNs) in recognizing diseases in rice leaves is pivotal. Through rigorous validation processes, the performance and generalization ability of CNN-based disease recognition systems are thoroughly evaluated. This typically involves partitioning the dataset into three sets which are training, validation, and testing sets to ensure unbiased evaluation. Validation metrics such as recall, accuracy, precision and F1 score are computed to quantify the model's performance. Additionally, techniques like cross-validation may be employed to validate the model’s robustness across different subsets of the data. Furthermore, validation on independent datasets helps verify the model's ability to generalize to unseen data and real-world scenarios. By meticulously validating CNN models for rice leaf disease recognition, researchers and practitioners can ensure the reliability and applicability of these systems in agricultural settings, thereby facilitating timely disease detection and management strategies for crop health and productivity.

F. Testing and Evaluation

The final trained model is evaluated on the testing data set to assess its performance in real-world scenarios. Metrics such as recall, accuracy, precision, and F1-score are calculated to quantify the model's effectiveness in classifying rice leaves into healthy and diseased categories.

In testing and evaluating rice leaf disease detection using Convolutional Neural Networks (CNN), the trained model is subjected to a separate testing dataset to assess its performance.

This evaluation typically involves measuring various metrics such as recall, accuracy, precision and F1 score to gauge the model's capacity to correctly identify different diseases present on rice leaves. Additionally, confusion matrices may be generated to analyze the mdel's classification performance for each disease class. By comparing the model's predictions against ground truth labels in the testing dataset, researchers can ascertain its efficacy in accurately detecting and classifying rice leaf diseases. This rigorous testing and evaluation process ensure the reliability and effectiveness of CNN-based approaches in aiding agricultural decision-making and crop management practices.

G. Deployment and Integration

In testing and evaluating rice leaf disease recognition using Convolutional Neural Networks, the model which is trained is subjected to a separate testing dataset to assess its performance. This evaluation typically involves measuring various metrics such as recall, accuracy, precision and F1 score to gauge the model's ability to accurately identify different diseases present on rice leaves. Additionally, confusion matrices may be generated to analyze the model's classification performance for each disease class. By comparing the model's predictions against ground truth labels in the testing dataset, researchers can ascertain its efficacy in accurately detecting and classifying rice leaf diseases.

This rigorous testing and evaluation process ensure the authentic and effectiveness of CNN-based approaches in aiding agricultural decision-making and crop management practices.

The results and experimental analysis of rice leaf disease recognition using Convolutional Neural Networks (CNNs) reveal promising advancements in automated disease detection in agricultural settings. Through extensive experimentation, CNN models have demonstrated remarkable capabilities in accurately identifying common rice leaf diseases such as Brown Spot, Leaf Blast, and leaf blight. Performance metrics including accuracy, precision, and recall have been thoroughly evaluated, showcasing the effectiveness of CNN-based approaches compared to traditional machine learning methods. Notably, deep learning models, particularly the 5-layer convolutional model, have shown superior performance, achieving impressive accuracy rates exceeding 78%. Furthermore, comparative analyses against established models like VGG16 have highlighted the efficacy and efficiency of CNN architectures tailored specifically for rice leaf disease recognition tasks. These findings underscore the potential of CNNs in revolutionizing disease management practices in agriculture, enabling farmers to swiftly detect and mitigate crop diseases, ultimately leading to healthier crop yields and enhanced economic stability in rice production.

Conclusion

In summary, we have utilized a Convolutional Neural Network method to classify different types of diseases affecting rice leaves, focusing on Bacterial leaf blight, Leaf Blast, and brown spot. Our study employed a dataset comprising diverse rice leaves afflicted with these diseases, which we processed using a custom-designed 5-layer convolutional network. Our findings indicate that this 5-layer CNN model excels in accurately identifying diseased rice leaves compared to other models. We also noted that by fine- tuning training parameters such as learning rate and optimizer methods, we achieved significant accuracy even with a simpler model containing fewer layers. Enhanced detection of infections can greatly aid farmers in safeguarding their crops. Moving forward.Our aim is to expand our research to encompass a broader range of diseases and algorithms, ultimately streamlining disease detection processes to be more comprehensive, efficient, and rapid.

References

[1] Shreya Ghosal, Kamal Sarkar, “Rice Leaf Disease Classification using CNN with Transfer Learning”, IEEE Conference Publishers, IEEE Xplore. [2] Ramakrishnan, M. (2015, April). “Groundnut leaf disease detection and classification by using back propagation algorithms” [3] 0p [4] Vimal K. Shrivastava, Monoj K. Pradhan, Mahesh P. Thakur, “Application of Pre-Trained Deep Convolutional Neural Networks for Rice Plant Disease Classification”. IEEE Conference Publishers, IEEE Xplore. [5] R. Swathika, S. Srinidhi, N. Radha, K.Sowmya, “Disease Identification in Paddy Leaves Using CNN Based Deep Learning”, IEEE Conference Publishers, IEEE Xplore. [6] Rahman, C. R., Arko, P. S., Ali, M. E., Khan, M. A. I., Apon, S. H., Nowrin, F., & Wasif, A. (2020). “Identification and recognition of rice diseases and pests using convolutional neural networks”. Biosystems Engineering, [7] Rishabh Sharma, Vinay Kukreja, Rajesh Kumar Kaushal, Ankit Bansal. \"Rice Leaf Blight disease detection using multi- classification deep learning model.\" IEEE Access, 342-344. DOI:10.1109/ICRITO56286.2022.9964644 [8] A.K.M Salam Hosain, Md. Humaion Kabir Mehedi, Tamanna Jahan Jerin, Md. Manik Hossain(2022),\"Rice Leaf Disease Detection with Transfer Learning.\" IEEE Access, DOI: 10.1109/IICAIET55139.2022.9936780 [9] A. Howard, M. Sandler, B. Chen, W. Wang, L.-C. Chen, M. Tan, G. Chu, V. Vasudevan, Y. Zhu, R. Pang, H. Adam, and Q. Le, ‘‘Searching for MobileNetV3,’’ in Proc. IEEE/CVF Int. Conf. Comput. Vis. (ICCV), Oct. 2019, pp. 1314–1324. [10] Pallapothala Tejaswini, Priyanshi Singh, Monica Ramchandani, Yogesh Kumar Rathore(2023)” Rice Leaf Disease Classification Using CNN”. IOP Conference Series [11] Bhattacharya, S., Mukherjee, A., Phadikar, S.: A deep learning approach for the classification of rice leaf diseases. In: Intelligence Enabled Research, pp. 61–69. Springer (2020) [12]

Copyright

Copyright © 2024 Chinthalapati Meghana, Senapathi Vennela, Sivapuram Sree Rama Gayatri, Dr. T. D. Bhatt. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.