Plant Disease Detection Using Machine Learning

Abstract

Agricultural productivity is significantly affected by crop pathogens and pests, with unpredictable climatic conditions intensifying the challenge. This growing threat to global food security highlights the need for efficient plant disorder detection methods. Traditionally, naked eye observation is used for identifying plant disorders, but it requires continuous expert presence and struggles with visually similar symptoms. Existing automated detection approaches often rely on simple background datasets like Plant Village, which may not perform well in real-world conditions with complex backgrounds and occlusions. To address these limitations, non-invasive, image-based detection methods are proposed. These approaches are simple, effective, and eliminate the need for in-situ data collection. The models are tested on various datasets, using handcrafted feature extraction techniques and novel image segmentation methods that calculate super pixels. To enhance performance, thermal imaging is employed for detecting water stress, an abiotic disorder. Additionally, a residual module-based deep learning model is developed to improve generalization and mitigate overfitting, surpassing traditional feature-based methods. These approaches show promise in accurately detecting both biotic and abiotic plant disorders, contributing to early intervention and improved crop management. The integration of advanced imaging and deep learning techniques holds potential for practical, real-world agricultural applications.