CUSTOMIZED U-NET CNN MODEL FOR TOMATO LEAF-BASED DISEASE CLASSIFICATION.

The pests and agricultural diseases are estimated to be responsible for the loss of ten percent of the annual productivity throughout the globe. Most farmers are unable to identify visually, whether a crop has been infected with a disease or not. This is due to a lack of crop morphological knowledge, or a lack of technical knowledge about disease forecast, or temporal features understanding in protecting the crops from being afflicted in the future. The high variability and complexity of leaf diseases, which make accurate identification difficult, is the principal barrier in tomato leaf disease detection. Convolutional Neural Networks (CNN) and U-Net architecture-based models can be implemented to address this issue. In our method, deep learning associated image processing techniques facilitate the preprocessing of tomato leaf images, enhancement of leaf image quality for feature extraction, and extraction of pertinent information for accurate leaf disease diagnosis. CNN enables efficient feature extraction, whereas U-Net enables accurate segmentation and localization of disease areas in tomato leaf images, resulting in enhanced disease detection and diagnosis. Our model uses images of tomato leaves, to construct a deep learning model, which uses a sequential model to identify illnesses, which affect plants. After using the U-Net architecture for image segmentation and localization, the convolutional neural network is capable of carrying out classification of tomato leaf diseases with an accuracy of 92.77% on training dataset, 94.85% on testing dataset and 83% on random generated dataset. [ABSTRACT FROM AUTHOR]