摘要
针对车辆类型检测受物体遮挡以及车辆重叠等影响,导致车辆类型的检测难度大的问题,本文提出了一种基于改进Faster RCNN模型的检测方法,在特征提取网络中嵌入卷积模块的注意力机制模块结构,使得特征提取网络可以重点关注与目标相关的有用信息,并弱化其它的无用信息,还引用Soft-NMS算法优化NMS算法,减少重叠或相邻目标漏检和错检问题。测试结果表明,与未改进的Faster RCNN模型做对比,改进后Faster RCNN模型的MAP值由84%提升至89%,证明使用该方法使检测精度有一定的提升。
Vehicle type detection is affected by object occlusion and vehicle overlap,which makes vehicle type detection difficult.This paper proposes a detection method based on the improved Faster RCNN model,which embeds the Convolutional Block Attention Module(CBAM)structure of the convolution module in the feature extraction network,so that the feature extraction network can focus on useful information related to the target and weaken other useless information.It also refers to the Soft-NMS algorithm optimization the NMS algorithm to reduce the missed detection and false detection problems of overlapping or adjacent targets.The test results showthat compared with the unimproved Faster RCNN model,the MAP value of the improved Faster RCNN model is increased from 84%to 89%,which proves that the detection accuracy can be improved to a certain extent by using this method.
作者
魏相站
邵丽萍
周骅
WEI Xiangzhan;SHAO Liping;ZHOU Hua(College of Big Data and Information Engineering,Guizhou University,Guiyang 550025,China)
出处
《智能计算机与应用》
2020年第7期97-100,103,共5页
Intelligent Computer and Applications
基金
贵州大学培育项目(黔科合平台人才[2017]5788-60)
贵州大学引进人才培育项目(贵大人基合字[2015]53号)
