Performance Analysis of Real-Time Object Detection algorithm for a Multi-Class Plant Disease Detection and Classification Using Deep Learning

The Indian economy is highly dependent on agricultural yield. Although due to plant diseases losses arise in terms of quality as well as quantity of agricultural products. In order to avoid plant diseases, it is quite necessary that they be supervised from the very beginning of the life cycle. Farmers usually observe the plants using their naked-eye which takes too much time as well as additional costs are incurred with requirement for expertise. Due to this there arises a need to automate the disease detection system that helps us automate and fasten this process using modern techniques. We present a method for identifying plant leaf diseases using the “You Only Look Once” (YOLO) algorithm in this paper. This object detection technique divides images into grids, with each cell responsible for detecting objects within its designated region. To enable real-time detection, the YOLO algorithm utilizes convolutional neural networks (CNNs). Also, we evaluated the performance of Yolov5 and Yolov7 on the Plantdoc dataset. Both algorithms are single stage object detectors. We found that Yolov7 requires significantly more processing power than Yolov5 during training. Additionally, we observed that the performance of Yolov7 is not consistent, as the mAP values fluctuated more per epoch. In contrast, Yolov5 performed much better.