基于YOLOv7的道路目标检测改进算法研究  

On Improved Road Target Detection Algorithm Based on YOLOv7

作　　者：何泽江 吴杰 谢曙芝 HE Zejiang;WU Jie;XIE Shuzhi(Changsha Nanfang Professional College,Changsha Hunan 410219,China;Central South University of Forestry and Technology,Changsha Hunan 410004,China)

机构地区：[1]长沙南方职业学院,湖南长沙410219 [2]中南林业科技大学,湖南长沙410004

出　　处：《北京工业职业技术学院学报》2025年第2期21-26,共6页Journal of Beijing Polytechnic College

基　　金：2023年教育部产学合作协同育人项目(230826311807159);2022年湖南省研究生科研创新项目(CX20220735)。

摘　　要：自动驾驶视觉感知算法需要在合理延迟的基础上,准确地检测道路目标。针对该问题,基于YOLOv7,提出一种改进的道路目标检测算法。算法使用SODA10M道路目标检测数据集,确定了损失函数各部分损失的最佳权重,验证了最佳锚框数量和在不同检测尺度上的锚框分布,在不降低检测精度的前提下提高了检测速度;改进了一种加权特征融合ELAN结构,促进了网络模型中提取特征的动态融合;在检测头前采用ASFF网络实现提取特征的跨尺度融合,增强了网络的特征提取能力。实验证明:改进后算法的平均检测精度mAP、精确率P和召回率R,优于改进前的算法以及与之对比的其他目标检测算法。Autonomous driving visual perception algorithms need to accurately detect road targets based on reasonable delay.To address this problem,an improved road target detection algorithm based on YOLOv7 is proposed.Using the SODA10M road object detection dataset,the optimal weights of the losses in each part of the loss function are determined,and the optimal number of anchor frames and the distribution of anchor frames at different detection scales are verified,which improves the detection speed without decreasing the detection accuracy.A weighted feature fusion ELAN structure is improved,which facilitates the dynamic fusion of the extracted features in the network model.The ASFF network is used before the detection head to achieve cross scale fusion of feature extraction,which enhances the feature extraction capability of the network.The experimental results show that the average detection accuracy mAP,precision P,and recall R of the improved algorithm are superior to the original algorithm and other object detection algorithms compared to it.

关 键 词：深度学习 道路目标检测 自动驾驶 YOLOv7算法 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]