Research On Target Recognition Of End-sensitive Elastic Line Array Infrared Image

With the increasingly severe international situation,the frequent occurrence of local wars,and the increasingly complex battlefield environment,the intelligent and information technology of ammunition are important development trends.With the extensive use and upgrading of armored vehicles,the task of anti-armour is becoming more and more important.As an important means of anti armor weapons,the development of terminal sensitive projectile has attracted much attention.Infrared detection technology is one of the important ways to detect the target of the terminal sensitive projectile.It has the advantages of good concealment and long working time.And it also gradually replaces the point source detection with infrared imaging technology.Therefore,based on the infrared image of the terminal sensitive projectile,this paper discusses the target recognition algorithm.Firstly,based on the principle of infrared detection and the model of infrared radiation propagation,the detection band and the number of detector elements are determined,and the important parameters in the steady-state scanning stage of the detection system are analyzed.Secondly,compare different image filtering and denoising and image enhancement algorithms,determine the image preprocessing algorithm suitable for this situation,and then use Canny operator to achieve the segmentation of target and background.Then,the invariant moment feature,texture feature,HOG feature and Harris corner feature of infrared image are extracted respectively.Aiming at the problem that Harris corner detection algorithm extracts fewer corner points from infrared image of terminal sensitive elastic line array,B-spline function is introduced,the convolution of the Cubic B-spline function and the Gaussian function is regarded as a new window function,and compared with other improved algorithms.The experimental results show that the number of corner points extracted by this algorithm is large,and the algorithm has good real-time performance.Finally,the principle of support vector machine is briefly introduced,the appropriate parameters are selected,and the extracted feature vectors are divided into two groups for classification and verification.One group is Harris corner detection + SIFT feature description+ K-means clustering + BOF model + SVM classification and recognition,the other group is combination feature + data normalization + PCA dimension reduction + SVM classification and recognition.The verification of high-tower experiments shows that although the classification accuracy of the two methods is lower than that of the laboratory,they still have higher classification accuracy.