基于改进BiSeNetV2的工具语义分割算法

Tool semantic segmentation algorithm based on improved BiSeNetV2

针对发电机风洞检修场景下检修工具自动化检测任务,提出了一种改进BiSeNetV2的检修工具语义分割算法。在细节分支末端增添SPPFCSPC模块融合不同尺度的特征,并将细节分支中的标准卷积替换为深度可分离卷积,显著降低计算量。在语义分支中添加高效通道注意力模块优化特征提取性能,提升分割的精度。实验结果表明改进的BiSeNetV2算法能精准检测检修工具位置信息,相比于BiSeNetV2,该算法MIoU提升了3.7个百分点,推理速度提升了16.4%。

In addressing the automated detection task for maintenance tools in the generator wind tunnel scenario,this study proposes an enhanced semantic segmentation algorithm based on BiSeNetV2.The improvement involves introducing the SPPFCSPC module at the end of the detail branch to fuse features from different scales.Additionally,standard convolutions in the detail branch are replaced with depthwise separable convolution,resulting in a significant reduction in computational complexity.To optimize feature extraction performance,an efficient channel attention module is incorporated into the semantic branch,enhanc-ing segmentation accuracy.Experimental results demonstrate that the enhanced BiSeNetV2 algorithm accurately detects the posi-tion information of maintenance tools.Compared to the original BiSeNetV2,this algorithm exhibits a 3.7 percentage improvement in MIoU and a 16.4%improvement in inference speed.