Improvement and Implementation of a Potato Recognition Algorithm Based on YOLOv8

To address issues such as poor recognition accuracy and low efficiency caused by the large number of potato targets that need to be processed at once during potato identification, this paper proposes an improved YOLOv8-based potato recognition algorithm. The improvements include: introducing DynamicConv in the backbone network to enhance model representation capability; replacing the C2f module in the backbone network with C2f_SAConv to strengthen feature extraction in object detection and segmentation tasks; incorporating Slim-neck in the Neck layer to reduce model parameters and floating-point operations; and adding an EMA attention mechanism to improve the model's focus on different potatoes. The improved YOLOv8 algorithm achieves an average recognition accuracy of 89.7%, reduces computational load by 27.2%, and facilitates deployment on resource-constrained embedded devices.