Severity Analysis of Leaf Disease Identification in Pear Using Deep Object Detection Methods

Abstract

Early-stage leaf diseases in plants can significantly impact crop yield and quality. If left untreated, these diseases can spread rapidly. This will lead to widespread damage and potential crop loss. Early detection allows for targeted disease control measures to minimise negative effects on plant health and optimise crop production. This study examines the effectiveness of various object detection methods for identifying early-stage leaf diseases, with a particular focus on pear leaf disease. Advanced machine learning algorithms, including R-CNN detectors and YOLO models, were employed to analyse plant leaf images. The YOLOv8s model emerged as the most effective with an mAP of 88.3. This may be attributed to its robust architecture and its ability to extract features effectively. The cascade-rcnn r50 fpn model demonstrated strong performance, due to the effectiveness of its multi-stage region proposal network and feature pyramid networks. YOLOv8s and cascade-rcnn r50 fpn showed promise in detecting mild disease symptoms, but further research is needed to cover a wider range of plant diseases and make them accessible for farmers to use