信息工程大学学报2025，Vol.26Issue(4)：444-449,6.DOI:10.3969/j.issn.1671-0673.2025.04.010
面向智能监控的轻量YOLOv10目标检测算法
Lightweight YOLOv10 Object Detection Algorithm for Intelligent Surveillance
摘要
Abstract
To address the limitations of traditional manual video surveillance,such as insufficient real-time performance and inefficient detection of transient targets,an intelligent surveillance detection al-gorithm based on lightweight YOLOv10 is proposed.Firstly,depthwise separable convolutions are em-ployed to replace standard convolutions,reducing network parameters while accelerating detection speed.Secondly,the cross-stage partial bottleneck structure with dual convolution fusion in the back-bone network is substituted with an efficient multi-scale attention module,enhancing the network's sensitivity to target scale variations.Finally,an auxiliary bounding box optimization loss is integrated to enrich supervision signals and improve small-target detection performance.Experimental evalua-tions on the UA-DETRAC traffic surveillance dataset demonstrate that the proposed algorithm achieves a higher mean average precision(mAP)by 13.5,10.9,1.7,and 0.4 percentage points com-pared to these of Faster R-CNN,EfficientDet-D5,YOLOv8,and YOLOv11,respectively.With a detec-tion speed of 112 FPS and merely 2.2×106 parameters,this algorithm provides robust technical support for object detection tasks in intelligent surveillance systems.关键词
智能监控/目标检测/深度学习/YOLOv10模型/轻量化Key words
intelligent surveillance/object detection/deep learning/YOLOv10/lightweight分类
信息技术与安全科学引用本文复制引用
刘春友,唐志斌,刘智国,宋宇斐..面向智能监控的轻量YOLOv10目标检测算法[J].信息工程大学学报,2025,26(4):444-449,6.基金项目
安徽省高等学校省级质量工程项目(2023JYXM1692) （2023JYXM1692）
安徽省教育厅2024年度重点项目(2024AH050101) （2024AH050101）
