Research on pipeline crack recognition algorithm based on deep learning

With the continuous development of science and technology, the city’s daily maintenance and inspection work is also constantly changing, urban pipeline maintenance and review of this important work related to the daily life of residents is very important, the more common pipeline inspection technology is CCTV inspection technology. The commonly used CCTV inspection technology also relies heavily on human subjective judgment, so the YOLO algorithm, which is highly efficient and high-speed in the field of vision, is used to detect and identify the common interfaces and two kinds of cracks in the pipe wall. Firstly, the CBAM attention mechanism is introduced into the Backbone structure in order to improve the extraction of key feature information in the local area, thus improving the detection ability of small targets. Then, the original C3 module of the Backbone structure is replaced by the C2F module, which can enhance the expressive ability of the network model to improve the detection accuracy, and the mAP value finally reaches $\mathbf{9 8 . 5 \%}$.