基于改进YOLO v5的城市地下管网缺陷识别算法

Defect recognition method of urban underground pipe network based on improved YOLO v5 algorithm

摘要:

城市地下管网系统作为保障城市排涝安全的重要市政基础设施，在长期超负荷运行过程中普遍存在着诸多病害问题。传统检测技术CCTV依赖专业人员的专业技术以及先验经验，因此为实现自动化的城市地下管网缺陷病害，一种城市管网缺陷病害检测算法被提出并成功运用于实际工程中。采用自适应CA注意力机制，有效弱化复杂背景的负面影响；缺陷分类与回归的解耦关键方法，使得检测部分充分利用缺陷纹理和边缘信息，从而提高小尺寸缺陷的精度；SIoU损失函数的运用为算法引入角度项权衡，有效加快收敛速度。经实验测试得到71.1%的平均精确度，较原始算法提高5.3%，并满足了实际工程上的应用。

Abstract:

As an important municipal infrastructure that ensures the safety of urban drainage, urban underground pipe network systems commonly suffer from many disease problems during long-term overload operation. Traditional detection technology CCTV relies on the professional skills and prior experience of professionals. Therefore, in order to realize automated urban underground pipe network defects and diseases, an urban pipe network defect and disease detection algorithm was proposed and successfully used in actual projects. The adaptive CA attention mechanism is adopted to effectively weaken the negative impact of complex backgrounds; the key method of decoupling defect classification and regression enables the detection part to make full use of defect texture and edge information, thereby improving the accuracy of small-sized defects; The application of SIoU loss function introduces angle term trade-offs into the algorithm, effectively speeding up the convergence. After experimental testing, an average accuracy of 71.1% was achieved, which was 5.3% higher than the original algorithm and satisfied the practical engineering application.