基于改进YOLOv8的烟丝检测的实例分割算法OA

Aiming at the problems of difficult segmentation of slender tobacco shreds,blurred edges and low detection efficiency,a lightweight instance segmentation algorithm based on the improved YOLOv8-EffDS-YOLO was proposed.The lightweight network EfficientNet-B2 was introduced as the backbone network in EHDS-YOLO.Combined with the long strip-shaped feature of tobacco shreds,a feature fusion module including dynamic serpentine convolution and convolutional block attention module was designed to improve the adaptability and response speed of the algorithm to tobacco shreds of different scales.Meanwhile,an edge-weighted loss function was adopted to improve the accuracy of edge segmentation.The experimental results show that,compared with the original YOLOv8 instance segmentation algorithm and other comparison methods,the improved algorithm has significant improvements in both recall rate and mAP@0.5 index,and the processing speed also meets the actual production requirements.