A NOVEL IMAGE SEGMENTATION TECHNIQUE FOR IMPROVING PLANT DISEASE CLASSIFICATION WITH DEEP LEARNING MODELS

Abstract

This study proposes a deep learning-based framework for automated plant leaf disease detection by integrating image segmentation and classification into a unified pipeline. The model is built on a U-Net-based architecture, which performs pixel-level segmentation using an encoder–decoder structure with skip connections to capture both spatial and contextual features. A Kaggle-based leaf disease segmentation dataset with annotated images and masks is used, along with preprocessing techniques such as resizing, normalization, and data augmentation to improve generalization. The training process employs the Adam optimizer with a combination of Binary CrossEntropy and Dice loss functions to enhance segmentation accuracy. Additionally, a classification module is incorporated to determine whether a leaf is healthy or diseased. Experimental results demonstrate high classification accuracy and effective leaf region segmentation, although fine-grained lesion detection remains limited due to class imbalance and background complexity. Overall, the proposed approach highlights the potential of deep learning techniques in precision agriculture and provides a scalable solution for automated disease detection and crop monitoring.