Tank Armor Based On Faster R-CNN Target Detection Research

With the continuous development in the field of computer vision,image-based target detection technology has become one of the topics of focus in the field of scientific research.At present,the technology has been widely used in video surveillance,artificial intelligence,image recognition and other fields.In the military field,target detection technology also has important research significance.In military operations,due to the complexity of the ground battlefield environment and the camouflage behavior of tank armor itself,it has increased the difficulty of detecting tank armor targets.Therefore,the recognition of tank armor images in complex environments has important research significance.The research object of this article is the image of tank armor in complex environment,which first analyzes the current status of target detection research based on tank armor in complex environments,and then conducted in-depth study on Faster R-CNN algorithm(Faster Region-Based Convolutional Neural Networks).And on the basis of Faster R-CNN algorithm,respectively proposed improved algorithms for feature extraction and Region Proposal Network(RPN).Finally,the two improved algorithms are experimentally verified on the data set,which proves the feasibility of the improved algorithm.The improved feature extraction algorithm is based on the characteristics of only one feature extraction network in the original Faster R-CNN algorithm,which proposes to apply two feature extraction networks on the regional recommendation network and the classification regression network,and conduct experimental detection on the image of the tank armor.Experimental results show that the detection accuracy of the improved algorithm is 3% higher than the original algorithm of Faster R-CNN.The improved algorithm of the proposed region is an improved algorithm proposed for the problems of poor quality and many invalid regions output by the Faster R-CNN algorithm.Experimental results show that the proposed region output by the improved algorithm is significantly less than the Faster R-CNN algorithm,and the improved algorithm saves 9ms on the detection time of each image,thus proving the efficiency of the improved algorithm.Experimental results show that,based on the characteristics of the original Faster R-CNN algorithm,respectively proposed improved algorithms for feature extraction and Region Proposal Network,eventually,the research objects can be accurately and efficiently identified.