Enhanced leaf disease detection: UNet for segmentation and optimized EfficientNet for disease classification

Abstract

This manuscript delineates the code developed for a published scholarly article aimed at supporting researchers in addressing plant leaf disease detection and classification (PLDC) challenges while evaluating the efficacy of various deep learning models. Furthermore, the research incorporates preprocessing strategies, correlation, segmentation employing the UNet model, feature extraction methods and EfficientNet model. The software model generates graphs such as confusion matrix, ROC curve (Receiver Operating Characteristic), and visual representations of loss and accuracy graphs. The initial research was disseminated in the Multimedia Tools and Applications journal, and the accompanying dataset was also introduced in the Data in Brief journal. • Optimized EfficientNet deep learning model. • The U-Net model is used for image segmentation where the weights are updated to extract the specific region from leaf to classify the disease. • Feature extraction technique is used to extract the features like texture, shape, color etc. from the segmented image. • Our software model learns the features to solve over-fitting issues and vanishing gradient issues. • Unwanted errors are also addressed by our model for classifying the diseases.