基于YOLOv8-SEBF的织物疵点目标检测算法OA

Aiming at the problems of weak feature extraction ability and frequent false detection phenomenon of YOLOv8 model in fabric defect detection task,a fabric defect detection algorithm based on improved YOLOv8(YOLOv8-SEBF)was proposed.To address the issue of weak feature extraction ability for fabric defects,EnhanceDATransformer was introduced,the model simulated the long-range dependence of fabric textures through Transformer,enabling global modeling and strengthen the features.For the problem of false detection,SCConv was introduced to enhance the model expression ability by combining the feature reconstruction of space and channel dimensions,and the false detection caused by texture background was significantly reduced through space-channel collaborative filtering.BiFPN was used to replace PAN in the fusion layer,and the expression ability of fusion features was enhanced through the cross-scale weighting fusion mechanism.The CIoU was optimized as Focal-EIoU,and balance of difficult&easy samples and the layer return of small defects was strengthened by introducing dynamic weighting mechanism,accuracy rate and stability of model detection were improved.The improved algorithm was tested on the actual production dataset,and the results showed that the mAP value of YOLOv8-SEBF algorithm was reached 93.8%,which was 16.0 percentage points higher than that of the original YOLOv8.Under the premise of ensuring the detection speed,the overall performance of the model was significantly improved.