基于改进YOLOv8n的管道超声图像缺陷识别方法研究

Research on Pipe Ultrasound Image Defect Recognition Method Based on Improved YOLOv8n

摘要:

管道缺陷检测作为管道安全管理的重要组成部分，对保障管道安全至关重要，超声波检测技术可用于识别和评估管道的内部缺陷；针对人工分析检测数据存在的执行效率低、漏检误检等问题，提出了一种基于改进YOLOv8n的管道超声图像缺陷识别方法；在YOLOv8的C2f模块中引入动态蛇形卷积，增强对缺陷的特征提取能力，将空洞卷积集成到SPPF模块，以减少缺陷信息损失，通过共享组卷积检测头，降低模型复杂度的同时提高对缺陷的定位能力；实验结果表明，改进的YOLOv8n算法能够实现管道超声图像自动缺陷检测，且改进后的模型与原模型相比，mAP50提升了2.1%，计算量和参数量下降了15%和7%；较其他主流检测算法，综合表现最佳。

Abstract:

Pipeline defect detection is a crucial component of pipeline safety management and plays a key role in ensuring pipeline safety. Ultrasonic testing technology can be used to identify and assess internal defects in pipelines. To address issues such as low execution efficiency and false or missed detections in manual analysis of testing data, this paper proposes a defect recognition method for pipeline ultrasonic images based on an improved YOLOv8n model. The dynamic serpentine convolution is introduced into the C2f module of YOLOv8 to enhance feature extraction capabilities for defects. Dilated convolutions are integrated into the SPPF module to reduce defect information loss, while a shared group convolution detection head is used to lower the model complexity while improving defect localization accuracy. Experimental results demonstrate that the improved YOLOv8n algorithm enables automatic defect detection in pipeline ultrasonic images. Compared to the original model, the improved model achieved a 2.1% increase in mAP50, with reductions in computational load and parameter count by 15% and 7%, respectively. The improved model outperforms other mainstream detection algorithms in overall performance.