基于深度学习利用特征图加权融合的目标检测方法  

An Object Detection Method Based on Deep Learning Using Feature Map Weighted Fusion

作　　者：张世辉[1,2] 王红蕾 陈宇翔 刘新焕 张健 何欢 任卫东[1] ZHANG Shi-hui;WANG Hong-lei;CHEN Yu-xiang;LIU Xin-huan;ZHANG Jian;HE Huan;REN Wei-dong(School of Information Science and Engineering,Yanshan University,Qinhuangdao,Hebei 066004,China;The Key Laboratory for Computer Virtual Technology and System Integration of Hebei Province,Qinhuangdao,Hebei 066004,China;Beijing Institute of Computer Technology and Application,Beijing 100854,China)

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004 [2]河北省计算机虚拟技术与系统集成重点实验室,河北秦皇岛066004 [3]北京计算机技术及应用研究所,北京100854

出　　处：《计量学报》2020年第11期1344-1351,共8页Acta Metrologica Sinica

基　　金：国家自然科学基金(61379065);河北省自然科学基金(F2014203119);装备发展部信息系统局“十三五”预研课题(31511040401)。

摘　　要：为了提高目标检测的准确性,提出了一种基于深度学习利用特征图加权融合实现目标检测的方法。首先,提出将卷积神经网络中的浅层特征图采样后与最深层特征图进行加权融合的思想;其次,根据所提的特征图加权融合思想以及卷积神经网络的具体结构,制定相应的特征图加权融合方案,并由该方案得到新特征图;然后,提出改进的RPN网络,并将新特征图输入到改进的RPN网络得到区域建议;最后,将新特征图和区域建议输入到后续网络层完成目标检测。实验结果表明所提方法取得了更高的目标检测精度以及更好的目标检测效果。In order to improve the accuracy of object detection,a method based on deep learning using feature map weighted fusion is proposed. Firstly,the idea fusing the sampled shallow feature maps and the deepest feature map in the convolutional neural network is proposed. Secondly,the corresponding feature map weighted fusion scheme is developed according to the idea of feature map weighted fusion and the specific structure of convolutional neural network,and a new feature map is obtained from the scheme. Thirdly,an improved RPN network is proposed,and the new feature map is input into the improved RPN network to obtain the region proposals. Finally,the new feature map and the region proposals are input subsequent network layers to realize object detection. The experimental results show that the proposed method achieves higher object detection precision and better object detection effect.

关 键 词：计量学 目标检测 视觉测量 深度学习 特征图加权融合 改进的RPN网络 

分 类 号：TB96[机械工程—光学工程]