摘要
针对无人驾驶场景下目标检测算法误检率高的问题,设计一种改进YOLOv3的多目标检测算法。该文在原始特征提取网络Darknet53中引入分组卷积核替换标准卷积核,降低了卷积操作的计算量;改进原始YOLOv3的特征融合方法,使不同尺度的特征层融合更加充分,对遮挡目标和小目标的检测效果有明显提升;构建CIoU位置损失函数,提示网络收敛效果。实验结果表明,改进的YOLOv3算法平均精确度提高了1.71%,误检率降低了12%,明显优于原始算法。
Aimed at the problem of high false detection rates of object detection in unmanned driving scene,a multi-target detection algorithm with improved YOLOv3 is proposed.The groups convolution kernel was introduced into the original feature network Darknet53 to replace the original convolution kernel,which reduced the complexity of convolution operation.The original feature fusion was improved to make the fusion of different scales more fully,and it improved the detection effect of occluded and small targets.The CIoU loss function was constructed to make the network convergence better.Experimental results show that the average accuracy of the improved YOLOv3 algorithm is increased by 1.71%,and the false detection rate is reduced by 12%,which is significantly better than the YOLOv3 algorithm.
作者
牛文杰
伊力哈木·亚尔买买提
Niu Wenjie;Yilihamu·Yaermaimaiti(School of Electrical Engineering,Xinjiang University,Urumqi 830046,Xinjiang,China)
出处
《计算机应用与软件》
北大核心
2024年第8期282-288,共7页
Computer Applications and Software
基金
国家自然科学基金项目(61866037,61462082)。