Potato Disease Detection Algorithm Based on Improved YOLOv8

Address the challenges posed by the diverse range of potato fruit defects, which encompass difficulties in detection and limited accuracy, an improved YOLOv8 potato disease detection algorithm is proposed. Firstly, for the lack of spatial information capturing ability of the original YOLOv8 model, the DRB idea is introduced to improve the C2f module, and then the original C2f module is improved by the improved C2f-DRB module. Secondly, for the problem that small targets of potato diseases are not easy to be detected, the CARAFE up-sampling operator is introduced to improve the model's ability of extracting features of the disease, and to enhance the effect of the disease detection. Finally, the original IoU loss function of the model is improved using the Inner IoU loss function, which enhances the generalization ability of the model. Relevant comparison experiments were constructed for ten kinds of diseased potatoes, and the results showed that the improved YOLOv8 algorithm improved 2.4%, 1.7%, and 1.5% in precision, recall, and mAP50, respectively, which proved the proposed algorithm had better performance, and verified the effectiveness and feasibility of the proposed algorithm.