Research On Infrared Dim And Small Target Detection Method Based On Background Low Rank And Target Sparse Characteristics

With the continuous development of infrared imaging technology and the needs of modern warfare,infrared small and dim target detection and tracking technology has become the core technology of imaging guidance and warning system in military field.Due to the influence of atmospheric radiation and the long distance,infrared targets are usually small or dotted,lacking of shape and structure information and the signal-to-noise ratio is also very low,which makes it difficult to detect infrared small targets in complex background.Therefore,it is very important to detect and track the small targets robustly in complex background.This article analyzes and summarizes the advantages and the limitations of the existing detection algorithms.This paper focuses on the characteristics of infrared background and target.Based on the low-rank property of background and the sparse property of target,the recovery algorithm of low-rank background and the enhancement of sparse target is deeply investigated to improve the robustness and accuracy of the detection in different scenes.(1)An infrared small and dim target detection method is presented based on weighted nuclear norm minimization.Based on the infrared patch-image model,the image data matrix is decomposed into a low rank matrix and a sparse matrix based on robust principal component analysis(RPCA).To make up the poor performance of RPCA model in describing complex backgrounds,the weighted nuclear norm is introduced into RPCA for better description of the background's low-rank property,and the corresponding optimization algorithm is also given.Besides,an adaptive threshold segmentation method is presented,which can accurately extract small and dim targets from sparse target image.(2)The RPCA model only employs one low-rank subspace assumption of backgrounds When the background component comes from mixture of multi-low-rank subspaces,the performance of the RPCA on background suppression will be poor.Although the WNNM-IPI algorithm based on RPCA effectively lower the false alarm rate of the edge region of complex backgrounds,but it does not fundamentally solve the limitation of RPCA.In order to solve this problem,this paper proposes and implements an infrared small target detection method based on low rank representation(LRR).Considering the heavy noise and clutter in images,which can contaminate the target matrix,the noise term is added to the original model to improve its robustness to noise.And this paper also introduces WNNM into the low rank representation model to obtain a more accurate low rank representation and improve the accuracy of the algorithm.(3)In order to solve the problem that the background information of the image is lost based on the two-dimensional principal component analysis algorithm,the structure tensor is employed to make better use of the local structure information of the background.Besides the three kinds of tensor expansion matrices,the other three kinds of data matrices are added into RIPI model for better mining of the structural features.In addition,Local signal-clutter-ratio analysis is introduced into sparse reweighted strategy to enhance the true targets and avoid targets being overwhelmed by strong background edges.Experimental results show that the proposed approach endows high detection probability and robustness to noise,and outperforms state-of-the-art methods in different data sets.