摘要
为了解决燃气管道沿线监控视频中普遍存在的背景复杂、多样的小目标和多尺度目标识别能力不足的问题,提出了基于CBAM注意力机制和4倍下采样检测头的YOLOv8改进算法,并实现了基于该改进算法的智能视频监控系统.实验结果表明,改进的算法精确度、召回率和均值平均精确度mAP@0.5分别达到了 80.2％、72.8％和80.2％的性能指标,相比原始的YOLOv8s算法分别提高了 1.9％、3.8％和1.6％.以某市燃气管道安全监管为研究案例,验证了该改进算法系统在燃气管道安全监控中的效率、准确率和时效性,为实现燃气管道安全监管的常态化、规范化和精细化,提供了一定的技术支持.
Abstract
An improved YOLOv8 algorithm,integrating the CBAM attention mechanism and a 4x down-sampling detection head,is proposed to tackle the prevalent challenge of limited capability in identifying intricate and diverse small and multi-scale targets in surveillance videos along gas pipelines.Generally speaking,the experimental results show that the improved algorithm has achieved respective performance index of 80.2％for accuracy,72.8％for recall rate,and 80.2％for mean average accuracy mAP@0.5,which are 1.9％,3.8％,and 1.6％higher than the original YOLOv8s algorithm respectively.To be spe-cific,the efficiency,accuracy,and timeliness of this system in monitoring the safety of gas pipelines have been validated through a case study on safety supervision of gas pipelines in a specific city,which serves as a certain technical basis for achieving the normalization,standardization and refinement of gas pipeline safety supervision.
维普
万方数据