复杂环境下的红外路况检测识别研究  

Research on Infrared Road Condition Detection and Recognition in Complex Environment

作　　者：袁子玄 廖义奎[1] Yuan Zixuan;Liao Yikui(College of Electronic Information,Guangxi Minzu University,Nanning 530006)

机构地区：[1]广西民族大学电子信息学院,南宁530006

出　　处：《现代计算机》2022年第23期1-9,共9页Modern Computer

基　　金：广西民族大学中国-东盟研究中心创新研究团队课题(TD201405)。

摘　　要：针对红外场景下存在目标模糊,难以检测的问题,提出一种基于YOLOv5所改进的算法。首先,为提高目标的检测率,在YOLOv5算法中加入CBAM注意力机制,提取注意力信息,增强红外目标在检测网络中的特征表达能力。然后,改进Bottleneck模块,有效实现特征重用,提高了模型的计算效率。接着,优化了Backbone与Neck结构部分,降低了网络的计算量。最后,改进上采样函数。研究结果证明,改进的网络在红外图像检测与识别中有着更好的检测率,计算量也更少,mAP值为81.3%,比原始网络提升了3个百分点。Aiming at the problem that the target is blurred, unclear and difficult to detect in the infrared scene, an improved algorithm based on YOLOv5 is proposed. First, in order to improve the detection rate of targets, the CBAM attention mechanism is added to the YOLOv5 algorithm to extract attention information and enhance the feature expression ability of infrared targets in the detection network. Then, the Bottleneck module is improved to effectively implement feature reuse and improve the computational efficiency of the model. Next, the Backbone and Neck structures are optimized to reduce the computational complexity of the network. Finally, improve the upsampling function. The research results show that the improved network has better detection rate and less computation in infrared image detection and recognition, and the mAP% value is 81.3%, which is 3.0% higher than the original network.

关 键 词：红外检测 YOLOv5 CBAM注意力机制 Bottleneck结构 

分 类 号：U495[交通运输工程—交通运输规划与管理] TP391.41[交通运输工程—道路与铁道工程]