Research On Multi Sensor Target Recognition Technology Based On Machine Learning

Target recognition technology of multi-sensor which based on machine learning is studied in this paper.This paper makes differences on bag of words(BoW)model which can turn SIFT with high level vectors into feature vectors with low level,and this improvement is applied to target recognition of combining global feature and with local feature and recognition of fusing infrared and visible images.Firstly,target recognition which based on global feature is studied.Theory of support vectors machine(SVM)is researched and multi-classifier which based on SVM is constructed.When SVM classifies LBP features extracted from visible images,phenomenon that some samples cannot be recognized appears.To solve this problem,two strategies are put forward.Secondly,target recognition which based on local feature is studied.Improvements on BoW model which based on k-means clustering algorithm are made.Based on existing research works of k-means,related improvement is lunched and result of experiment taken on UCI dataset proves effects of advanced k-means algorithm by this paper is better than before.Result of experiment that prove the target recognition rate is improved by using advanced k-means algorithm when the advanced algorithms are applied to BoW model.Thirdly,target recognition based on feature level and decision level is studied respectively.Experiment of combining local feature and global feature using visible images under decision level is lunched.In this time,SIFT feature and LBP feature are combined.Two kind of combine methods which can improves the finally recognition rate are put forward during the experiment.Another experiment where infrared and visible images are fused is lunched.Result of this experiment proves using two kind of sensors can improve final recognition rate under the premise of unequal number of samples.