Deep Learning Architectures for Tomato Leaf Disease Segmentation: A Comparative Analysis

The growing threat of plant diseases has raised significant concerns for the agricultural sector, particularly affecting vital crops like tomatoes. These diseases, caused by various pathogens such as bacteria, viruses, and fungi, pose a substantial risk to food security by reducing crop yields. Recent years have seen an alarming rise in the occurrence of tomato leaf diseases, largely attributed to environmental changes, leading to the destruction of a significant portion of tomato harvests in many regions. The need for timely and accurate diagnosis of these diseases is now more pressing than ever, as traditional manual methods have proven inefficient and labor-intensive. Advances in artificial intelligence (AI) and deep learning (DL) technologies offer a promising solution to this challenge. Convolutional Neural Networks (CNN) has been particularly effective in automating the identification and segmentation of diseased regions on tomato leaves. By analyzing large datasets of leaf images, these models are capable of detecting disease patterns with high precision. Among the various DL models, UNet++ has demonstrated superior performance, boasting an accuracy of 98.67% in the segmentation of infected areas. Its innovative architectural design, featuring enhanced skip connections and deep supervision, allows for more detailed analysis of complex leaf patterns. Other models such as FCN, SegNet, and Mask R-CNN also contribute to this field, but UNet++ remains the most reliable for high-accuracy tasks. By facilitating early detection and precise treatment, DL models have the potential to significantly reduce crop losses, thereby enhancing productivity and ensuring more stable food supplies in the future.