An Improved Faster R-CNN Algorithm for Detection of Multi-scale Tank Armored Vehicle Targets

Image target detection technology under complex battlefield environment is the key issue that needs to be solved for the future development of automatic target detection and tracking tank fire control system.In this paper,an improved algorithm is proposed for the problem of the Faster R-CNN algorithm in the detection of small-scale tank armored vehicle targets.First,a scale-dependent RPN network is introduced to set reasonable-sized sliding windows on the convolution layer of different depths according to the distribution of targets scale,so that the RPN network can extract more accurate proposals.Second,we propose a selective pooling strategy that select convolution features at appropriate layer depths based on the size of the proposals generated by the scale-dependent RPN network to perform ROI pooling,leaving enough target information for detection sub-network.Finally,in the Tank VOC4000 image library,the improved algorithm and the original algorithm are tested and compared from the perspectives of target recall rate,detection accuracy and speed.The experimental results show that the improved algorithm can achieve good detection effect on tank armored vehicle targets of various scales by using the convolution features at multiple levels.The detection accuracy and speed are better than the original Faster R-CNN and can better meet the requirements of actual application.