“Advanced Deep Learning Techniques for Plant Disease Diagnosis and Treatment”

Abstract

The pivotal role of agriculture in shaping people’s lives and economies is evident, with a substantial contribution to GSP and employment. The prevalence of diseases poses a major challenge. Novel system is introduced for the early detection of leaf diseases using Deep Learning and Tensorflow technology. The model analyses pictures from plant village, focusing on bell pepper, tomato and potato plants. Climate change soil erosion and difficulty in detecting pathogens underscore the need for early disease recognition to minimize pesticide use. The system demonstrates a remarkable 95.80% correctness in disease identification, benefiting cultivators by recommending appropriate measures in a separate context. The recognition of medicinal plant gains importance for various stakeholders. A fully automated approach combining computer vision and machine learning achieves an accuracy of 90.1% in detecting 24 different medicinal plant species, outperforming other methods. The unique image dataset created for medicinal plants in Mauritius holds the potential to enhance local knowledge, aid taxonomists and contribute to endangered species protection. Finally, a novel stochastic channel reuse SCR block is introduced for ResNet, offering a feature-reusing mechanism in plant disease severity classification. SCR-ResNet is evaluated on the agri-plant dataset and cifar-10 dataset, demonstrating superior performance over baseline models. This theory contributes a novel SCR block and the SCR-ResNet model, showcasing its effectiveness in enhancing classification accuracy for plant disease severity detection.

Introduction

I. INTRODUCTION

Farming, often referred to as the backbone of many nations, plays a pivotal role in shaping the livelihoods of millions of people and driving economic stability. India, for instance, boasts a large farming community cultivating a diverse range of crops, making agriculture the primary source of employment and a significant contributor to the country’s gross domestic product GDP. The significance of agriculture is not limited to India; it holds a crucial position in the global context, with India ranking second only to the United States in farming output.

However, this vital sector faces numerous challenges that threaten its long-term sustainability and food security. Crop productivity is influenced by a multitude of variables, from changing weather patterns to political unrest, and as the world’s population continues to grow, the pressure to produce more food intensifies. In response to these challenges, the farming industry is rapidly seeking innovative ideas to enhance crop yields, reduce production, costs and improve food grade.

 Among the significant threats to agricultural productivity are plant diseases, particularly those affecting the leaves of crops. These diseases, often spread by microorganisms like insects, pests, fungi, bacteria, and viruses, not only degrade the quality of farming products but also result in reduced harvest yields.

To ensure food security and bolster the economy, early detection, and intervention in the case of leaf diseases are imperative. The emergence of advanced farming technologies offers promising solutions to these challenges, allowing for significant increases in crop yields, cost reduction, and food quality enhancement. To harness these benefits, it is essential to monitor and control environmental factors such as temperature, humidity, and light conditions to promote crop productivity and quality.

This introduction sets the stage for the exploration of various approaches, including deep learning and image processing, that have revolutionized agriculture by providing accurate and efficient means to detect and combat plant diseases. The subsequent sections of this paper will delve into the role of artificial intelligence, specifically deep learning algorithms, in addressing the challenges posed by plant diseases in agriculture. These advanced technologies are poised to play a crucial role in ensuring the sustainability of agriculture and global food security.

II. LITERATURE SURVEY

Agriculture is a vital sector in India, and crop diseases significantly impact productivity. Early detection of leaf diseases through automated methods like image processing is crucial. Convolutional Neural Networks (CNNs) are used for disease classification in this study. The study uses a CNN-based approach for leaf disease detection, which is trained on a dataset of preprocessed leaf images. CNNs are chosen due to their ability to automatically learn features from images, reducing the need for manual feature engineering.

1. Leaf disease detection and pesticides recommendation using Deep learning algorithm by Dr.B.Srinivas Rao sir, M. Snehith Varma, J.Sandeep Kumar, S.Anil Reddy, B.Varun Tej.

The paper references various studies related to automated leaf disease detection:

• Using Local Binary Patterns (LBPs) for feature extraction and One Class Classifiers.
• Pattern recognition for crop diseases using K-Means clustering and artificial intelligence.
• Disease detection for grape leaves using SVM classification.
• Image-based disease detection using deep learning and smartphone assistance.

The paper concludes by emphasizing the effectiveness of CNN-based disease detection,    especially for large-scale farming.  The proposed system reduces the need for expensive domain experts.

2. Leaf disease detection and pesticides recommendation using Deep learning algorithm by    Mushkan Bothra, Nishitha P, Neharaj, Pavithra KS, Madhu KM, Dr. Divya BS presents a comprehensive approach to tackling the challenges in agriculture, specifically related to plant disease identification, pesticide recommendation, and overall crop health management. The authors propose a system that employs deep learning, specifically Convolutional Neural Networks (CNNs) and TensorFlow, to automate the detection of leaf diseases in three different crops (Bell Pepper, Tomato, and Potato) and provide recommendations for pesticides and other preventive measures. This research addresses the critical issue of early disease detection, which is essential for maintaining crop productivity and food security. The literature review section of the paper briefly discusses the significance of automatic disease detection systems in agriculture, the impact of plant diseases on food security, and the challenges faced by smallholder farmers due to yield losses caused by pests and diseases. It highlights the importance of early disease detection, as late diagnosis can lead to significant losses. The section also touches upon the limitations of traditional disease detection methods, such as their time-consuming and expensive nature.

3. Automatic recognition of Medicinal plants using Machine learning by Adams Begue, Venitha Kowelessur, Fawzi Mahomoodally, Sameerchand Pudaruth, Upasana Singh presents a study on the automated recognition of medicinal plants using computer vision and machine learning techniques. They discuss various studies that have employed machine learning and computer vision techniques for plant classification. These studies encompass features like shape, texture, and other attributes for plant identification. The comprehensive review of related works provides a context for their research.The paper outlines the methodology used for the study. It explains how they collected images of leaves from 24 different medicinal plant species and describes the preprocessing steps applied to the images, such as removing shadows and performing thresholding. The feature extraction process is detailed, including base features like length, width, area, and derived features based on these attributes.  The authors discuss the machine learning classifiers used for the recognition task, including Random Forest, Multilayer Perceptron Neural Network, Support Vector Machine, Naïve Bayes, and k-Nearest Neighbor. The study presents the accuracy of each classifier, with Random Forest achieving the highest accuracy of 90.1%. A confusion matrix is provided, showcasing the performance of the Random Forest classifier.

4. Stochastic Channel Resuse Residual Network for plant disease severity detection by Wenjie Liu, Guoqing Wu, Fuji Ren. The provided paper focuses on plant disease classification using deep learning, with a particular emphasis on the Stochastic Channel Reuse (SCR) block. The adoption of deep learning techniques, particularly convolutional neural networks (CNNs), has gained traction in the field of agriculture, including plant disease detection. Deep learning has shown promise in automating the detection and classification of plant diseases from images, offering a faster and potentially more accurate alternative to manual diagnosis by experts. Several deep learning architectures have been explored for plant disease detection. These architectures include well-known models like AlexNet, GoogLeNet, and ResNet, which are adapted and fine-tuned for specific plant disease datasets. The choice of architecture plays a crucial role in the performance of disease classification models. The paper introduces the Stochastic Channel Reuse (SCR) block as a novel approach to address co-adaptation issues in deep learning models. While regularization techniques like dropout and drop connect have been widely used in CNNs, the introduction of the SCR block is an innovative attempt to further enhance model performance. Evaluation metrics, such as F1 score, accuracy, precision, and recall, are commonly used to assess the performance of plant disease detection models. Researchers often benchmark their models on specialized plant disease datasets, like the Agri-Plant dataset mentioned in the paper, to demonstrate the effectiveness of their approach.

III. PROPOSED SYSTEM

The proposed system for "Advanced deep learning techniques for plant disease diagnosis and treatment" consists of several components that work together to provide a comprehensive solution for the system.

1. Image Acquisition and Pre-Processing: The initial component involves the acquisition of plant leaf images through user uploads. This module applies pre-processing techniques such as normalization, scaling, and augmentation to prepare the images for analysis. It ensures that the input data is in a consistent format suitable for deep learning models to process efficiently.
2. Leaf Identification and Disease Detection: The core component utilizes convolutional neural networks (CNNs) for the dual task of identifying the plant species and detecting any signs of disease on the leaves. The system is trained on a comprehensive dataset of plant leaves with known diseases, allowing for precise identification and classification of plant health issues.
3. Remedy Recommendation Engine: Upon successful detection of the disease, this module kicks in to suggest appropriate remedies. It houses a database of plant diseases paired with expert-recommended treatments, ranging from chemical treatments to natural remedies. The recommendations are tailored to the specific disease and plant species, ensuring relevance and effectiveness.
4. Report Generation and User Interface: This component is responsible for generating a detailed report of the diagnosis and recommended remedies. The user interface is designed for ease of use, allowing users to understand their plant's health status clearly. The report is then formatted for dissemination to the user's email.
5. Communication and Feedback Loop: The final component includes a mechanism for sending the diagnosis and remedy report to the user via email. Additionally, it allows for user feedback which can be used to refine the system's accuracy and performance. Alerts can be set up to inform users of critical updates or follow-up actions required for their plants' health

The entire system works in stages, with the primary aim of providing accurate plant disease detection and effective remedy recommendations. The first design goal of this system is to ensure the precision of plant disease identification through an advanced image processing pipeline. This involves two key features:

a. Universal Analysis Capability

• This allows any user, not just agricultural professionals, to upload an image of a plant leaf and obtain a diagnosis. This inclusivity ensures that the system can be used by a broad audience without specialized knowledge.
• The system is designed to be stateless, meaning it does not need to retain user information between sessions, ensuring privacy and ease of use.

b. Intelligent Diagnosis Delegation

• Diagnosing plant diseases accurately can be challenging due to the variability in leaf appearance and disease manifestation. Our system delegates this task to a deep learning model trained specifically to recognize a wide array of plant species and diseases
• The model acts as an independent 'auditor' of the plant's health, providing an objective assessment not influenced by human error.

c. Deep Learning Verification

• The deep learning model independently verifies the presence of disease and identifies the plant species, ensuring the accuracy of the diagnosis. This is analogous to having an expert in plant pathology scrutinize every leaf image uploaded to the system.
• The system includes a risk assessment feature that evaluates the severity of detected diseases and suggests the urgency of recommended remedies.

d. Compliance with Agricultural Best Practices

• The recommendation engine is built in compliance with agricultural best practices and phytopathological research, ensuring that the remedies suggested are effective and environmentally sustainable.
• The system also adheres to data protection standards, ensuring user data privacy and security.

e. Continuous Learning and Data Auditing

• Just as third-party auditors continuously audit data to ensure integrity, our system continually learns from new data. This adaptive learning ensures the system evolves to recognize new disease patterns and plant species.
• The system also audits its performance by comparing predicted diagnoses with user feedback and expert consultations, allowing for ongoing improvement.

In addition to the intelligent diagnosis mechanism, the system leverages cloud computing resources for scalability and reliability. This ensures that the system can handle a high volume of users and data without degradation in performance.

Conclusion

The integration of neural networks, particularly Python\'s CNN model, has demonstrated significant promise in the field of leaf disease detection, offering an impressive 96 percent accuracy rate. This technology can be further improved with the incorporation of GPUs, presenting a cost-effective alternative to relying on domain experts for the identification and management of plant diseases. Additionally, it opens up exciting possibilities, including the use of high-resolution drone-mounted cameras for aerial surveillance of vast agricultural fields, reducing manual labour and time requirements, although this may be more practical for large-scale farming. The utilization of deep learning algorithms for leaf disease detection, along with recommendations for pesticides, fertilizers, and crop management, holds substantial potential. These algorithms can leverage extensive datasets to accurately identify leaf diseases and provide tailored solutions to enhance plant health. By considering environmental factors like soil type, temperature, and rainfall, these algorithms can offer precise crop recommendations, empowering farmers to optimize yields and minimize waste. While challenges such as data availability and environmental research persist, the potential benefits make this a promising area for further investigation. In summary, the integration of neural networks and deep learning algorithms has the capacity to transform the agricultural sector, ultimately advancing global food security. This innovative approach not only aids in disease detection and management but also contributes to efficient farming practices, benefiting both farmers and consumers.

References

[1] Dr.B.Srinivas Rao sir, M. Snehith varma ,J.Sandeep Kumar ,S.Anil Reddy,B.Varun Tej ,Leaf DiseaseDetection and Remedy Recommendation Using CNN, Vol 13, Issue 05, May /2022 [2] Mushkan Bothra, Nishitha P , Neharaj,Pavithra KS, Madhu KM, Dr. Divya BS , Leaf DiseaseDetection and Pesticide Recommendation using Deep Learning Algorithm , May /2023 [3] Adams Begue, Venitha Kowlessur, Fawzi Mahomoodally,Upasana Singh , SameerchandPudaruth ,Automatic Recognition of Medicinal Plants using Machine Learning Techniques , Dec / 2017 [4] Padmavathi, K & Thangadurai, K.. (2016). “Implementation of RGB and Grayscale Images inPlant Leaves Disease Detection – Comparative Study.” Indian Journal of Science and Technology. 9. 10.17485/ijst/2016/v9i6/77739. [5] Mohanty, Sharada & Hughes, David & Salathe, Marcel. (2016). “Using Deep Learning for ImageBased Plant Disease Detection.” Frontiers in Plant Science. 7. 10.3389/fpls.2016.01419. [6] Singh, Vijai & Misra, A.K.. (2016). “Detection of Plant Leaf Diseases Using Image Segmentationand Soft Computing Techniques.” Information Processing in Agriculture. 4. 10.1016/j.inpa.2016.10.005. [7] Pawar, P. et al. “Cucumber disease detection using artificial neural network.” 2016 International Conference on Inventive Computation Technologies (ICICT) 3 (2016): 1-5. [8] Halder, Monishanker & Sarkar, Ananya & Bahar, Habibullah. (2019). “Plant Disease DetectionbyImage Processing: A Literature Review.” SDRP Journal of Food Science & Technology. 3. 534-538. 10.25177/JFST.3.6.6. [9] Khirade, Sachin D. and A. B. Patil. “Plant Disease Detection Using Image Processing.” 2015 International Conference on Computing Communication Control and Automation (2015): 768-771.

Copyright

Copyright © 2023 Prof. Avinash Utikar, Vishal Palve, Milind Pushpam, Shruti Shahane, Prerna Pawar. 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.