摘要
针对Faster R-CNN内部网络结构对特征图信息利用不充分这一问题,对网络内部结构增加一条自下而上的反向侧边连接路径,对目标检测方法做出优化,采用公开的数据集MS-COCO对其进行训练和测试。实验证明,不同IoU阈值获取的数据相较于改进前的Faster R-CNN模型检测框架,其中包围盒和目标检测准确率都得到了一定程度的提高,尤其对于小、中等尺寸目标的检测准确率提高较多。
For inefficient use of feature diagram information in Faster R-CNN internal network,we add a reverse side connection path from up to down in the network structure to optimize target detection.Public data sets MS-COCO are used for training and testing.Comparing with the Faster R-CNN model detection framework,experimental results show that the accuracy of both bounding box and target detection of our method is improved with different IoU threshold,especially,for the smaller and medium-sized targets.
作者
郭昕刚
张培栋
梁锦明
王帅
GUO Xingang;ZHANG Peidong;LIANG Jinming;WANG Shuai(School of Computer Science & Engineering, Changchun University of Technology, Changchun 130012, China)
出处
《长春工业大学学报》
CAS
2020年第5期474-480,共7页
Journal of Changchun University of Technology
基金
吉林省发改委基金资助项目(2019C040-3)。
