MCSG-YOLOv8: A Lightweight Underwater Pipeline Defect Detection Algorithm

In the field of urban underwater pipeline defect detection, traditional detection algorithms face many challenges, such as low detection accuracy in complex environments, difficulty in meeting real-time requirements, and insufficient lightweight models. In response to these issues, this paper proposes a lightweight detection algorithm MCSG-YOLOv8 based on Yolov8. The algorithm, by introducing the CSPPC module, GhostConv module, and MLLA attention mechanism, significantly reduces the model size, parameter quantity, and computational amount while maintaining accuracy, enhancing real-time performance, and making it more suitable for resourceconstrained underwater detection scenarios. To comprehensively evaluate the proposed method, we have constructed a dedicated self-built dataset and conducted in-depth comparisons with the YOLOv8 baseline model. Experimental results show that while the precision of MCSG-YOLOv8 is improved by 2.42 percent, the model size is reduced by 30.88 percent, the parameter quantity is reduced by 31.61 percent, GFLOPs is reduced by 21.67 percent, and FPS is increased by 8 frames. This research not only provides new insights into the optimization of target detection algorithms in theory but also offers significant technical support for the precise detection and efficient identification of municipal pipeline defects in practical applications, which has profound significance for ensuring the safe and stable operation of urban pipeline systems.