Intelligent Image Segmentation Framework for Robust Plant Disease Classification

Abstract

The leaf-based disease detection has emerged as a key application of artificial
intelligence in modern agriculture, offering practical and efficient tools for monitoring crop health.
Ensuring that such systems perform reliably under real-world environmental conditions remains a major
research challenge. This study explores the use of advanced deep learning architectures for precise and
efficient detection of plant diseases, contributing to the ongoing digital transformation in agriculture. The
research focuses on enhancing the accuracy and robustness of automated disease classification in plants.
To achieve this, two disease-specific datasets were employed. A dedicated cauliflower leaf dataset was
created, comprising high-resolution images of leaves affected by Alternaria Leaf Spot and Black Rot,
enabling a detailed study of cauliflower-specific diseases. Additionally, an independent mango leaf
disease dataset was utilized to assess the framework’s generalizability across different crop species. The
proposed classification system operates in three main stages. Initially, leaf regions are segmented from
complex backgrounds to ensure that critical disease-related features are captured. Next, geometric data
augmentation techniques are applied to increase dataset diversity and improve the models’ generalization
capacity. Finally, four state-of-the-art deep learning architecturesVGG16, ResNet50, EfficientNetB3, and
MobileNetV3 Largeare employed for disease classification. Experimental results indicate that this
integrated deep learning framework offers a reliable and efficient solution for automated detection of
plant diseases across multiple crops