Tomato Leaf Disease Classification Using Siamese Network
1Mr.Anto Theepak , 2K. Jelin
1Assistant Professor, 2Student
Department of Information Technology,
Francis Xavier Engineering College, Tirunelveli, India
1antotheepak@francisxavier.ac.in, 2 jelink.ug.21.it@francisxavier.ac.in
Abstract— Tomato plants are particularly vulnerable to a range of diseases that can adversely affect both yield and quality, emphasizing the importance of timely and precise detection for effective agricultural practices. Conventional methods for disease identification depend on manual inspections, which can be labor-intensive, subjective, and susceptible to human mistakes. To address these challenges, this project presents a deep learning-based method employing a Siamese neural network that utilizes Euclidean distance for similarity assessment in classifying tomato leaf diseases. The model analyzes an input image in conjunction with a collection of reference images from a labeled dataset, extracting advanced feature representations through a pretrained convolutional neural network (CNN). The Euclidean distance computed from the extracted features indicates the closeness between the input image and the reference images, where a shorter distance signifies a closer match and a longer distance implies dissimilarity. A set threshold guarantees accurate classification by identifying the disease based on its nearest reference match. To enhance understandability, Grad-CAM (Gradient-weighted Class Activation Mapping) visualization emphasizes the crucial parts of the input leaf image that influenced the model's conclusion, helping users grasp the significant features that contributed to the classification. The system yields a comprehensive output, which consists of the image uploaded, the most similar reference image, the similarity score calculated, and the final disease classification. By harnessing deep learning, this technique boosts accuracy, efficiency, and dependability in diagnosing plant diseases, lessening the reliance on manual inspections and facilitating early detection of diseases for prompt intervention. This automated approach promotes improved crop management, enhanced agricultural productivity, and minimized economic losses, rendering it a strong and scalable solution for precision agriculture.
keywords —Deep Learning,Siamese neural network, Euclidean distance, disease classification, pretrained CNN, feature extraction, similarity measurement, threshold-based classification, Grad-CAM visualization
