Applying transfer learning in CNN model architectures for detecting tomato leaf disease with explainable artificial intelligence

Tomato diseases significantly impact agricultural productivity, leading to economic losses and reduced quality of crops. Given the importance of tomatoes as a major food source, timely and accurate disease detection is crucial. Traditional manual inspection is error-prone and impracti cal for large-scale farming. This study aims to enhance the accuracy and transparency of tomato leaf disease detection through deep learning techniques. A dataset of 523 images collected from Walkait Setit Humera Zone, with four irrigation sites, was augmented to 3,138 images using basic preprocessing methods to improve quality. The dataset was split into training, validation, and testing sets in a 70%, 10%, and 20% ratio. Pre-trained CNN architectures, including Mo bileNetV3, InceptionV3, and DenseNet201, were fine-tuned and implemented on Google Colab with GPU processing for disease detection. Explainable AI methods were employed to increase trust and model transparency. The DenseNet201 model, after fine-tuning, achieved the high est performance with 100 accuracy, precision, recall, and F1 score. This research significantly improves crop production and food security by providing an efficient, accurate, and computa tionally feasible solution for detecting tomato leaf diseases, thereby supporting the sustainable growth of agricultural practices. Algorithms Accuracy Precision Recall F1-Score Loss MobileNetV3 96.88 96.77 97.14 96.95 11.65 InceptionV3 98.44 98.86 97.61 98.23 4.99 DenseNet201 100 100 100 100 3.52