Automated Multi-Crop Leaf Disease Detection Using Deep Learning Models and YOLO Based Object Detection

Abstract

Crop diseases have been  the biggest challenges in attaining food security, thus highlighting importance of automated detection systems for crop diseases at early stages. Manual visual inspection is not effective, appropriate, or feasible for cultivating large areas of crops, relying on expert knowledge that is mainly based on observation. This article provides a comparative deep learning approach in accurately plant leaf detect diseases using image data, thus addressing the challenges associated with traditional visual inspection methods. This article presents a proposed framework that compares the performance of a number of pre training models, including Res Net, VGG, Mobile Net, Dense Net, Efficient Net, all have been pre trained  to the image datasets of leaf images, along with a custom CNN approach to accurately detect plant leaf diseases using image data. Standard image preprocessing techniques have been applicable to enhance the performance of proposed approach, thus eliminating the chances of overfitting, leading to increased accuracy in classifying leaf images under varying environmental conditions. In addition, real-time object detection is achieved using YOLO, thus accurately localizing disease areas on leaf images. In addition, to enhance real-world applicability, the proposed approach is implemented using a web-based system, thus making it feasible for integration with existing systems in agriculture. Experimental results show that transfer learning models outperform the custom CNN approach in terms of accuracy, whereas YOLO is used to accurately localize disease areas on leaf images.