Enhancing Plant Disease Detection: A Deep Learning Approach and Comparative Analysis of Pretrained Models with Explainable AI Visualization.
Abstract: Accurately detecting plant diseases is essential for safeguarding food security and promoting agricultural sustainability. This study presents an advanced deep learning approach using Convolutional Neural Networks (CNNs) to classify plant leaf diseases. Utilizing the Plant Village dataset, a custom CNN model was developed, which achieved a superior accuracy of 96.11%, surpassing popular transfer learning models such as ResNet152-v2 (94%), VGG16 (91%), and Inception-v3 (88%). Key aspects of this work include the implementation of robust data preprocessing techniques like image augmentation and the design of an optimized CNN architecture to enhance performance while maintaining computational efficiency. Furthermore, explainability was explored using Layer CAM, which provided more detailed visual explanations than Grad-CAM, thereby increasing the model’s transparency and trustworthiness. These results confirm the effectiveness of the proposed method for accurate plant disease identification, positioning it as a promising tool for agricultural disease diagnosis. Future efforts will focus on expanding the dataset, improving model generalization, and deploying the model for real-time field applications.
Background: In recent years, the application of Artificial Intelligence (AI) has revolutionized various fields by offering advanced solutions to complex problems. AI encompasses a range of techniques and technologies designed to enable machines to perform tasks that typically require human intelligence. Machine Learning (ML) and Deep Learning (DL) are two of the many subfields of artificial intelligence (AI) that have drawn a lot of attention because of their capacity to analyze massive datasets, find patterns, and make predictions. Applications for these technologies can be found in a variety of fields, such as agriculture, healthcare, and finance. Plant disease control and identification are essential to agriculture in order to maintain crop health and productivity. Plant diseases pose a threat to food security and can result in significant financial losses. Manual inspections are a common component of traditional plant disease detection techniques, although they are time- and labor-intensive. With the introduction of AI, disease detection procedures might be improved and automated, giving farmers and agricultural specialists access to timely and reliable information.
Objective: The primary objective of our research is to develop an advanced deep-learning model to enhance the classification of plant leaf diseases using images from the Plant Village dataset. Plant diseases pose a significant threat to global agriculture, requiring timely and accurate diagnosis to mitigate their impact. This research aims to create a reliable diagnostic tool that surpasses traditional methods in accuracy and efficiency. To achieve this objective, our research is structured around several key steps. Firstly, we will preprocess and augment the Plant Village dataset to optimize it for machine learning applications. This preparation is crucial to ensure The quality and diversity of the dataset improved the performance of our models. Next, we will design and implement a custom Convolutional Neural Network (CNN) tailored for plant disease classification. This custom model will be built To efficiently capture the intricate patterns and features contained in plant leaf photos, thereby improving classification accuracy. In parallel, we will train and evaluate three pre-trained deep learning models: VGG16 [16], Inception V3 [17], and ResNet152-v2 [15]. These established architectures are known for their robustness and high performance in image classification tasks. By employing these models, we aim to set a benchmark for our custom CNN. A comprehensive comparison between the custom CNN model and the pre-trained models will be conducted to assess their performance. Important metrics including accuracy, precision, recall, and F1 score will be the main emphasis of this comparison, giving a thorough grasp of the advantages and disadvantages of each 6 model. Our goal is to identify the most effective approach for plant disease classification, considering both accuracy and practical applicability. Furthermore, we will fine-tune the hyperparameters of all models, exploring various configurations to achieve optimal performance. This tuning process will involve adjusting learning rates, batch sizes, and other critical parameters, ensuring that each model is operating at its best. Through this research, we aim to make a significant contribution to agricultural disease management. An advanced deep learning model, refined through rigorous analysis and optimization, promises to improve diagnostic accuracy and support farmers in maintaining crop health. This research not only advances the subject of AI in agriculture, but it also lays the path for more efficient and scalable solutions to one of the most important issues in global food production.