Research On Infrared Dim Small Target Detection Based On Tensor Decomposition

Infrared dim small target detection is one of the hot research issues in the fields of military,traffic,surveillance,etc.However,due to the long imaging distance,the target exhibits a weak point-like characteristic,and there is no obvious texture and edge information.At present,scholars have proposed different detection algorithms for infrared weak target images in different backgrounds,but the robust and general infrared small target detection algorithm has been the goal pursued in this field.Taking the airborne infrared small target image sequence in different backgrounds as the research object,the paper applies the tensor decomposition theory to study the infrared weak target detection method and algorithm.The main work is as follows:The basic theory of tensor is introduced.The high-order structure can be used to preserve the original features of the data,and the infrared video is converted into a third-order tensor to characterize the spatial correlation and time continuity of the target.The characteristics of the infrared weak target image sequence are analyzed,which can be divided into three parts:target,background and noise.The background tensor shows low rank and the target tensor shows sparsity.The tensor decomposition can accurately mine the implicit information of the data,and then assist in separating the target from the original image sequence.Therefore,the infrared weak target detection problem is transformed into the tensor low rank sparse decomposition problem.An infrared weak target detection algorithm based on Weighted Tensor Robust Principal Component Analysis（WTRPCA）is proposed.The low-rankness of the background tensor is described by the tensor kernel norm based on the tensor t product.The weighted7)₁ norm describes the sparsity of the target tensor and is solved iteratively by Lagrange multiplier method.Experiments show that the tensor low-rank sparse decomposition in 3D mode can more effectively separate the target,and the WTRPCA algorithm with weight factor can better suppress the background and noise of the image.An infrared weak target detection algorithm based on Bayesian Tensor Low-rank Sparse Factorization（BTLSF）is proposed.Firstly,for the infrared video features,the probability distribution model of low rank background tensor,sparse target tensor and noise tensor is assumed.For the background low-rank tensor,the rank of the tensor is minimized by minimizing the number of columns of the factor matrix,which solves the problem that the rank of the tensor needs to be given in advance.The Markov model is introduced in the target sparse tensor section to perform trajectory constraints,and the false alarm is filtered out.At the same time,the noise module is modeled separately,and the Bayesian framework is used to derive the approximate expression of the noise statistic,which enhances the robustness of the algorithm to different noises.The weak target infrared video with different local signal-to-noise ratios is simulated and the algorithm is tested.When the local signal-to-noise ratio is reduced to about 1.3,the detection rate of the algorithm can still reach 0.8889.Finally,for different infrared weak target scenarios,the proposed algorithm is compared with some excellent existing algorithms.The two algorithms outperform other algorithms in target enhancement and background suppression,and can achieve infrared weakness in different scenarios.Target Detection.