Research On Methods Of Infrared Dim And Small Target Detection Under Complex Background

Infrared dim and small target detection is a key technology in the field of precision guidance.It plays an important role in aircraft infrared search and tracking systems?IRST?,infrared imaging and guidance systems,and early warning systems for military installations.In order to achieve efficient,reliable and stable detection of infrared dim and small targets,comprehensive and thorough research on infrared dim and small target detection methods under complex background is necessary.The main work of this thesis is as follows:Firstly,a dim and small target detection method based on multiscale infrared superpixel-image model is proposed.The superpixel method is used to segment the original infrared image to obtain the superpixel images with no overlapping area,which not only makes full use of the local spatial correlation of the infrared image,but also avoids the computational burden caused by redundant information.Introducing multiscale theory followed by merging target images detected at different scales can improve the robustness of the algorithm for detecting targets of different sizes.Experimental results show that,compared with Top-Hat method,Max-Mean method,Max-Median method,two-dimensional least mean square?TDLMS?method,local saliency map?LSM?method,infrared patch-image?IPI?method,the proposed method has more advantageous background suppression effect and better adaptability to the target size.Then,a fractional-order function based reweighted low-rank and sparse approximation model for infrared dim and small target detection is proposed.Using the reweighted nuclear norm and ₁lnorm based on the fractional-order function can better approximate the rank and ₀l norm of the matrix,thus more effectively retaining background clutter such as strong edges,and suppressing non-target sparse components in the target image.Simultaneously,the novel fractional-order function based iterative reweighting mechanism can reduce the number of iterations of the algorithm and speed up the convergence,thus shortening the detection time.Extensive experimental results show that,compared with Top-Hat method,TDLMS method,weighted local difference measure?WLDM?method,IPI method,non-negative infrared patch-image model based on partial sum minimization of singular values?NIPPS?method,reweighted infrared patch-image?ReWIPI?method,the background suppression effect of the proposed method is better,the accuracy of target detection is higher,and it has certain advantages in terms of running speed.Finally,an infrared patch-group model with content awareness for dim and small target detection is proposed.Exploiting the infrared patch-group model by similar patch grouping can better satisfy the low rank hypothesis of the background patch image,thus capturing more detailed features in the background.Moreover,the partial sum of singular values with content awareness is employed to address the patch sample deficiency problem.By further combining a data-driven measure called regional redundancy,the target-background separation is achieved via an adaptive low rank and sparse matrices recovery algorithm.A large number of experimental results show that,compared with Top-Hat method,phase spectrum of Fourier transform?PFT?method,multiscale patch-based constrast measure?MPCM?method,IPI method,weighted IPI?WIPI?method,NIPPS method,the proposed method achieves a superior balance between background suppression and target detection.