Semantic segmentation algorithm fused with wavelet domain information in vaccine glass vial defect detection
Confronting the challenges of irregular defect shapes and high inter-class similarity in vaccine vial defect detection,this study introduces a semantic segmentation network that integrates wavelet domain information.The network incorporates a discrete wavelet transform branch,enabling effective fusion of wavelet domain information with the convolutional backbone's feature maps,thereby enhancing the segmentation integrity of irregular defects.In the design of the backbone network,dynamic snake convolution is introduced to refine the segmentation of irregular defects.Furthermore,the backbone network structure is optimized to reduce computational resource dependence while increasing segmentation precision.A dedicated vaccine vial defect segmentation dataset was constructed for a series of comparative experiments with several existing convolutional semantic segmentation networks and vision Transformer networks.Experimental results demonstrate that the proposed method outperforms the compared networks in accuracy,integrity,and refinement,confirming its effectiveness in the task of vaccine vial defect detection.
万方数据
维普
