DeepResNet18: Model of Tomato Leaf Disease Prediction to Initiate Sustainable World

To achieve sustainability, our aim is to get healthy food which achieves from disease free crop. We consider tomato, a nutrient and delightful fruit, for our research work. Healthy plant grows healthy fruit or crop. Early disease prediction of leaf helps to sustain plant for providing wholesome meal. Our study presents an end-to-end deep learning system for detecting tomato leaf diseases using the ResNet18 architecture. This architecture is pre-trained on ImageNet. We have proposed DeepResNet18 model, fine-tuned on the Plant Diseases Dataset from Kaggle, focusing on three classes: healthy tomato leaves, bacterial spot, and tomato mosaic virus, with 5,418 training and 1,354 validation samples. The fully connected layer of ResNet18 was replaced with a custom classifier head. It uses ReLU as an activation function and a 0.4 dropout rate. Data preprocessing involved resizing images to 224x224 pixels, applying augmentation techniques like random horizontal flip, rotation, resized crop, and color jitter for training, and normalization for validation and testing. The model was trained for 100 epochs. We have used Adam Optimizer. The learning rate that we set to is 0.001. The model gives a validation accuracy of 99.85% at epoch 73 and a test accuracy of 99.23%. Accuracy, precision, recall, F1 score helps to establish the performance of our proposed model. DeepResNet18 offers a highly accurate, automated solution for early tomato leaf disease detection, suitable for agricultural applications.