GPNet:轻量型红外图像目标检测算法  

GPNet:Lightweight infrared image target detection algorithm

作　　者：李现国[1,2] 曹明腾 李滨 刘意 苗长云 LI Xian-Guo;CAO Ming-Teng;LI Bin;LIU Yi;MIAO Chang-Yun(School of Electronics and Information Engineering,Tiangong University,Tianjin 300387,China;Tianjin Key Laboratory of Optoelectronic Detection Technology and System,Tianjin 300387,China)

机构地区：[1]天津工业大学电子与信息工程学院,天津300387 [2]天津市光电检测技术与系统重点实验室,天津300387

出　　处：《红外与毫米波学报》2022年第6期1092-1101,共10页Journal of Infrared and Millimeter Waves

基　　金：国防科技创新特区项目,天津市重点研发计划科技支撑重点项目(18YFZCGX00930)。

摘　　要：针对资源受限的红外成像系统准确、实时检测目标的需求,提出了一种轻量型的红外图像目标检测算法GPNet。采用GhostNet优化特征提取网络,使用改进的PANet进行特征融合,利用深度可分离卷积替换特定位置的普通3×3卷积,可以更好地提取多尺度特征并减少参数量。公共数据集上的实验表明,本文算法与YOLOv4、YOLOv5-m相比,参数量分别降低了81%和42%;与YOLOX-m相比,平均精度均值提高了2.5%,参数量降低了51%;参数量为12.3 M,检测时间为14 ms,实现了检测准确性和参数量的平衡。A lightweight infrared image target detection algorithm GPNet is proposed to address the need for accurate and real-time target detection in resource-constrained infrared imaging systems.The feature extraction network is optimized using GhostNet,feature fusion is performed using an improved PANet,and a depth-separable convolution is used to replace the ordinary 3×3 convolution at specific locations to better extract multi-scale features and reduce the number of parameters.Experiments on public datasets show that the algorithm in this paper reduces the number of parameters by81%and 42%compared with YOLOv4 and YOLOv5-m,respectively;the average mean accuracy is improved by 2.5%and the number of parameters is reduced by 51%compared with YOLOX-m;the number of parameters is 12.3 M and the detection time is 14 ms,which achieves a balance between detection accuracy and number of parameters.

关 键 词：红外图像 目标检测 YOLO GhostNet 参数量 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术] TN219[自动化与计算机技术—计算机科学与技术]