Multi-Task Learning for Tomato Crop Disease Detection and Severity Estimation using CNN Framework

The cultivation of tomatoes holds immense significance in the agricultural landscape of the Indian region, also it plays a substantial role in economic and nutritional contributions. Nonetheless, the widespread occurrence of diseases poses a notable risk to both the quantity and quality of tomato yields. This study endeavors to create a real-time application employing a deep learning mechanism in conjunction with a multi-task method. In this study, the process utilizes convolutional neural networks (CNN) and VGG16 architecture. The determination of disease severity and the stage of the disease is conducted through multi-task learning methods integrated with CNN, providing a comprehensive approach to address the challenges associated with disease threats in tomato cultivation.

I. Introduction

Automated analysis of plant leaves holds significant importance on agriculture for the early detection of plant ailments, thereby contributing to enhanced production and quality. Tomatoes, being susceptible to a variety of diseases such as bacterial and fungal infections, as well as viral strains, necessitate swift and accurate detection methods. The expeditious identification of these diseases is crucial to mitigate their detrimental impact on tomato plants and improve overall crop productivity. The initial step involves the preparation of input images, followed by the segmentation of targeted portions from the initial images. Subsequently, the dataset undergo further processing using different hyper-parameter configurations CNN-VGG16 model. This allows the CNN to capture essential features from the images, encompassing color, texture, and boundaries. While CNN models exhibit proficiency in disease categorization, accurately assessing the severity and current stage of crop diseases remains a challenge. To address this, a Multi-Task Learning strategy is employed, serving as a classifier that considers the seriousness of the ailment. The structure of the approach is detailed in Section 2, which provides an extensive analysis of previous research, followed by Section 3 that outlines the suggested methodology, with subsequent discussion of results within the same section.