摘要
针对安防场景中红外图像对比度低、目标轮廓不明显导致目标检测效果差的问题,提出一种基于改进YOLOv7的红外安防目标检测算法。采用递归门控卷积改进主干网络,增强对输入图像高阶信息交互能力;使用SimAM注意力机制构建ELAN-S模块,降低信息丢失率的同时减少网络参数;使用K-means++聚类算法优化锚盒尺寸,提高检测精度。对InfiRay公开数据集进行数据增强和模型验证实验,结果表明,提出的算法在保持较高检测速度前提下,平均精度均值达到了87.15%,相对于原YOLOv7网络与其他主流算法有明显提高,证明改进方法先进有效。
To address the problem of poor object detection due to low contrast and inconspicuous object contours in infrared images in security scenes,an improved YOLOv7-based infrared security object detection algorithm is proposed.The recursive gated convolution is used to improve the backbone network and enhance the ability to interact with higher-order information of the input image;the ELAN-S module is constructed using the SimAM attention mechanism to reduce the information loss rate while reducing the network parameters;the anchor box size is optimized using the K-means++ clustering algorithm to improve the detection accuracy.Data enhancement and experiments are conducted on the InfiRay public dataset,and the results show that the algorithm proposed in this paper has an mAP value of 87.15% while maintaining a high detection speed,which is a significant improvement compared with the original YOLOv7 network and other mainstream algorithms,proving that the improved method is advanced and effective.
作者
韩瑶
骆晓玲
程换新
沈静
HAN Yao;LUO Xiaoling;CHENG Huanxin;SHEN Jing(Qingdao University of Science and Technology,Qingdao 266061,China;Hubei University,Wuhan 430062,China)
出处
《激光杂志》
CAS
北大核心
2024年第5期55-61,共7页
Laser Journal
基金
国家海洋局重大专项项目(国海科字[2016]494号No.30)。
