基于改进YOLO的双模目标识别方法研究

摘要

According to the meets of automatic target recognition with multi-sensor image,a method of automatic target recog-nition in infrared and visible image and the fusion of decision level has been studied.For the characteristics that targets could vary in large scale,rotate rapidly,and move with time,the latest convolutional neural network YOLO is used to learn and classify the targets. And the algorithm is run in GPU to meet the need of real time computation.Considering the different imaging characteristics of infra-red and visible light,the targets in infrared image and visible image are positioned respectively.And DS theory is applied to the fu-sion of these two modes giving the results of fusion,and thus improves the ability of automatic target recognition.%针对多源图像自动目标识别的需求,对红外、可见光自动目标识别以及决策级融合方法进行了研究.针对目标所具有的变化尺度大、容易旋转、移动等特点,引入最新的YOLO卷积神经网络模型对目标进行深度学习和分类,并在GPU中完成目标识别,满足工程实时计算的需求.考虑到红外、可见光不同的成像特性,最后将红外、可见光目标反算定位对齐,并用DS证据理论对两模目标进行融合,给出融合结果,提高了系统对目标的识别能力.