改进Yolov4的车辆弱目标检测算法  

Improved Yolov4 algorithm for vehicle weak object detection

作　　者：王坤 项琦鑫 WANG Kun;XIANG Qixin(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300

出　　处：《中国惯性技术学报》2023年第8期797-805,共9页Journal of Chinese Inertial Technology

基　　金：中国国家自然科学基金(62173331)。

摘　　要：针对交通场景中存在的小目标和遮挡目标等弱目标漏检、错检、检测速度慢等问题,提出一种改进Yolov4的车辆弱目标检测算法。首先设计像素重组残差模块(PS-R),通过超分辨率的方式将多个目标的中心点分散到不同网格中,保留更多的遮挡目标;其次设计特征增强注意力模块(FEAB),充分利用高层特征的语义信息和浅层特征的细粒度信息,提升弱目标检测性能;然后根据道路车辆的目标特点,k-means++结合遗传算法对车辆数据集的真实标注框进行聚类,生成更符合车辆目标的先验框;最后使用深度可分离卷积替换网络特征融合模块(PANet)中的常规卷积,提升检测速度。在车辆数据集KITTI和UA-DETRAC上进行实验验证,改进后的Yolov4算法比原始Yolov4算法精度分别提高了1.9%和2.4%,检测速度达到了61.4 fps。A vehicle weak object detection algorithm with improved Yolov4 is proposed to address the problems of missed detection,wrong detection,and slow detection speed of weak objects such as small objects and obscured objects in traffic scenes.Firstly,the pixel shuffle residual module(PS-R)is designed to retain more occluded objects by scattering the centroids of multiple objects to different grids in a super-resolution manner.Secondly,the feature enhancement attention block(FEAB)is designed to make full use of the semantic information of high-level features and the detailed information of shallow-level features to improve the performance of weak object detection.Then,based on the object characteristics of road vehicles,the genetic algorithm is combined with the k-means++algorithm to cluster the ground truth boxes of the vehicle dataset.The generated anchor boxes are more compatible with the vehicle objects.Finally,a deep separable convolution is used to replace the convention convolution in the network feature fusion module(PANet)to improve the detection speed.The detection results show that the improved algorithm achieved 1.9%and 2.4%higher than the original Yolov4 on KITTI and UA-DETRAC datasets respectively,and the detection speed is up to 61.4 fps.

关 键 词：深度学习 弱目标检测 特征增强 像素重组 深度可分离卷积 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] U495[自动化与计算机技术—计算机科学与技术]