改进YOLOv4的轻量化目标检测方法

Improved Lightweight Target Detection Method for YOLOv4

针对车载平台发展过程中,在辅助驾驶环境感知方面,现有的目标检测方法对目标检测精度不高、算法推理速度慢等问题,本文以YOLOv4目标检测网络为基础,引入通道与空间注意力模块CBAM,有效提升了YOLOv4目标检测网络特征识别精度;引入Mobilenetv3轻量化网络结构替换YOLOv4的主干特征提取网络CSPDarkNet53,并利用深度可分离卷积替换整个网络的普通卷积,有效降低了YOLOv4目标检测模型大小,提升了网络模型推理速度。通过消融实验与检测结果分析,证明了改进方案的可行性。

In the development of vehicle platform, aiming at the problems of small target detection accuracy and slow algorithm reasoning speed of target detection method in auxiliary driving environment perception, this paper introduces channel and spatial attention module CBAM based on YOLOv4 target detection network, which effectively improves the feature recognition accuracy of YOLOv4 target detection network.Mobilenetv3 lightweight network structure is introduced to replace the backbone feature extraction network CSPDarkNet53 of YOLOv4,and the deep separable convolution is used to replace the ordinary convolution of the whole network, which effectively reduces the size of the target detection model of YOLOv4 and improves the reasoning speed of the network model.The feasibility of the improved scheme is proved by ablation experiment and analysis of detection results.