Leaf Disease Detection Using Deep Learning and ML Techniques

Abstract

Crop yield losses from plant diseases are estimated to be between 10% and 16% of total food production each year, making them one of the most enduring and economically devastating threats to global agricultural productivity. Particularly in rural and resource-constrained farming communities, conventional diagnosis methods rely on laborious, expert-led manual inspection that is limited in accuracy, scalability, and accessibility. In order to automatically classify plant diseases from digital leaf images without the need for specialized agronomist intervention, this paper presents a comprehensive AI-Assisted Plant Disease Detection System that uses Convolutional Neural Networks (CNNs). The system is implemented as a responsive Django web application with real-time inference, and it is trained on the popular PlantVillage benchmark dataset, which includes over 54,000 annotated leaf images covering 38 disease categories across 14 plant species. Four UML design artifacts—a system architecture diagram, a use-case diagram, an activity diagram, and a CNN layer architecture diagram—are used to formally document the entire system, offering a strict specification of both structural and behavioral system properties. Experimental results show that deep learning-based detection is a viable, low-cost substitute for traditional agronomic field inspection by confirming accurate real-time classification of several plant conditions, such as Pepper Bell Bacterial Spot, Tomato Early Blight, and Potato Late Blight.