摘要
针对YOLOv4目标检测器存在信息利用率不足的问题,提出了一种新的基于改进的路径聚合和池化YOLOv4的目标检测方法 YOLOv4-P。为了充分利用路径聚合可以有效防止信息丢失这个特点,对YOLOv4的路径聚合网络进行改进,利用主干特征提取网络的第二个残差块,新增一个检测层,加强融合浅层特征层。另外,使用K-means聚类对数据集重新进行处理,获得合适的先验框尺寸。此外,图像经过主干特征提取网络后的感受野比理论感受野小,为了增大感受野,在主干特征提取网络的后端加入金字塔池化模块,利用4种不同尺度的金字塔池化引入不同尺度下的特征信息。最后,在PASCAL VOC2007和VOC2012进行仿真实验,实验结果表明,提出的YOLOv4-P有效提高了检测精度。
Aiming at the problem of insufficient information utilization in the YOLOv4 object detector,an improved object detection method based on the improved path aggregation and pooling of the YOLOv4model,namely YOLOv4-P,is proposed.First,we improve the path aggregation network of YOLOv4since the path aggregation network can effectively prevent information loss.The second residual block of the backbone feature extraction network is used to add a detection layer to strengthen the fusion of shallow feature layers.Second,the datasets are reprocessed using the K-means clustering to obtain a suitable prior box size.Third,since the size of the receptive field of the image after passing through the backbone feature extraction network is smaller than that of the theoretical one,a pyramid pooling module is added to the back end of the backbone feature extraction network and four different scales of pyramid pooling are used to introduce features at different scales to improve the receptive field.The simulation experiments on PASCAL VOC2007 and VOC2012 show that the proposed YOLOv4-P can effectively improve the detection accuracy.
作者
杨真真
郑艺欣
邵静
杨永鹏
YANG Zhenzhen;ZHENG Yixin;SHAO Jing;YANG Yongpeng(Key Lab of Broadband Wireless Communication and Sensor Network Technology,Ministry of Education,Nanjing University of Posts and Telecommunications,Nanjing 210003,China;School of Network and Communication,Nanjing Vocational College of Information Technology,Nanjing 210023,China)
出处
《南京邮电大学学报（自然科学版）》
北大核心
2022年第5期1-7,共7页
Journal of Nanjing University of Posts and Telecommunications：Natural Science Edition
基金
国家自然科学基金(61501251)
南京邮电大学宽带无线通信与传感网技术教育部重点实验室开放研究基金(JZNY202113)
南京邮电大学科研项目(NY220207)资助项目。
