基于XS-BiSeNetV2的城市地下管道缺陷语义分割模型  

Sematic segmentation method for urban underground pipeline defects based on XS-BiSeNetV2

作　　者：田淙文 蓝雯飞[1] 李波[1] 潘禹欣 姚为[1] TIAN Congwen;LAN Wenfei;LI Bo;PAN Yuxin;YAO Wei(College of Computer Science,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院,武汉430074

出　　处：《中南民族大学学报(自然科学版)》2025年第4期536-545,共10页Journal of South-Central Minzu University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(61976226)。

摘　　要：针对城市地下管道缺陷分割任务对实时分割的需求,基于BiSeNetV2模型提出了一种改进的分割模型XSBiSeNetV2.针对城市地下管道缺陷形态多样、空间特征复杂的问题,设计了互相关联的双分支交叉特征提取策略;针对传统跨步卷积存在的空间信息损耗,使用Haar小波变换下采样模块替换细节分支的跨步卷积,保留更多空间信息;针对轻量级模型存在的上下文特征不匹配和空间特征偏移的问题,使用上下文和空间特征校准模块提高模型的分割精度.通过实验验证了提出的缺陷分割模型的实时性和有效性,对比原始BiSeNetV2模型,mIoU提升了2.65%,mPA也提升了1.47%,且拥有每秒49帧的处理速度,具有良好的实时性.相比其他基于深度学习的实时语义分割模型,也具有一定优势.In order to meet the demand of real-time segmentation for urban underground pipeline defects segmentation task,this paper proposes an improved segmentation model XS-BiSeNetV2 based on the BiSeNetV2 model.For the problems of urban underground pipeline defects with diverse morphology and complex spatial features,an interrelated two-branch crossover feature extraction strategy is designed.For the loss of spatial information in traditional step-by-step convolution,the Haar wavelet transform downsampling module is used to replace the step-by-step convolution of detail branches to retain more spatial information.For the problems of mismatch of context features and offset of spatial features in lightweight models,the context and spatial feature calibration module is used to improve the segmentation accuracy of the model.The real-time and effectiveness of the proposed defect segmentation model is verified through experiments.Compared with the original BiSeNetV2 model,the mIoU is improved by 2.65%,and the mPA is also improved by 1.47%,and it possesses a processing speed of 49 frames per second,which provides good real-time performance.It also has some advantages compared to other real-time semantic segmentation models based on deep learning.

关 键 词：XS-BiSeNetV2模型 缺陷分割 城市地下管道 Haar小波下采样 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]