Enhancing Tomato Leaf Disease Detection with Deep Learning and Augmentation Techniques

Tomato leaf diseases significantly impact crop yield and agricultural sustainability, especially in developing countries. Early and accurate detection of these diseases is crucial for timely intervention and minimizing losses. In this study, we explore and compare the performance of multiple real-time object detection models—YOLOv8n, YOLOv9c, YOLOv10ny, YOLOv11n, YOLOv12n, and RT-DETR—for the automatic identification of seven common tomato leaf diseases alongside healthy leaves. A custom-labeled dataset consisting of 737 high-resolution images was collected. Data augmentation was applied to the training set, producing three additional augmented versions of each image using the Albumentations library to enhance model robustness and generalization. The following pre-processing steps were applied to each image to standardize the input across all models. All models were trained for 100 epochs on identical configurations, and their performance was evaluated using mean Average Precision at IoU thresholds of 0.5 and 0.5:0.95. The results indicate that YOLOv9c achieved the highest [email protected] of 76.5%, while RT-DETR showed superior balance between precision and recall. This comparative study provides insights into the trade-offs between lightweight model architectures and detection accuracy, contributing toward efficient real-time disease diagnosis systems for smart agriculture applications.