Feature Extraction and Segmentation Methods in Plant Disease Detection: A Multimodal Approach

Abstract

Plant disease detection is essential for improving agricultural productivity. Deep learning models have shown great potential in identifying plant diseases because they can leverage large datasets. However, while efficient, traditional machine learning methods often face challenges with generalization when trained on small datasets using basic features like shape, color, and texture. A promising approach to overcome this is the combination of deep feature extraction with machine learning classification, enabling more accurate disease detection. Traditional classifiers trained on smaller datasets can still offer viable solutions in resource-limited environments. By extracting critical features and employing classical techniques, these models remain practical for constrained settings. Integrating deep learning models with traditional methods allows for better handling of disease variability across plants and conditions, enhancing adaptability and accuracy. This review explores deep learning and traditional machine learning approaches for feature extraction and segmentation in plant disease detection. It highlights how combining deep feature extraction with machine learning classification improves accuracy and addresses the challenges posed by limited datasets. The potential of multimodal techniques for enhanced detection is also discussed, leading to more robust and scalable solutions for plant disease management.