Dim-small Targets Detection Of Infrared Images In Complex Backgrounds

Infrared small target detection has been applied in many fields,but there still exist problems especially under complex background.A key factor is that the characteristics of the strong edge and the small target are very similar,and sometimes,some point noise in the image is also similar to the target.Therefore,it is easy to cause mistakes in target judgment.Meanwhile,the real-time performance and high detection are hard to balance.To solve these problems,the main work of this paper is as follows:In order to weaken the interference of strong edges in complex backgrounds and improve the real-time performance of the algorithm,we proposed a small target detection method based on group image-patch tensor(GIPT)model.It employs the structural information of the infrared image to promote target and background separation.Firstly,the features of the original image are depicted with the help of the structure tensor.Then the target component is assumed to be a point area by using the customized weighted sparse regularization.And the background area is modeled as line part,point part and flat part,which are recovered separately by using a novel group low rank operator.The proposed model can make the optimization process run in parallel and greatly reduce the complexity of the algorithm.The experimental results also show the huge advantages of this method compared to the other 6 algorithms in terms of running time and strong background edge suppression.The point noise contained in the complex background seriously interferes with the detection performance of infrared dim and small targets.In order to further improve the algorithm performance of the tensor model,this paper proposes an infrared dim small target detection algorithm based on super pixel segmentation.The proposed algorithm uses super pixel segmentation to construct a tensor to ensure the integrity of the local structure of the image patches,so as to better explore the low rank of the background and the sensitivity to the size of the sliding window in the tensor construction process.At the same time,it integrates the corner intensity,the consistency of the gradient direction and the local information entropy as the target prior.After that,the target weight is constructed by adding the reweighted strategy,the strong edge and noise interference are removed,and the detection model is constructed.Finally,the target detection is completed by adaptive threshold segmentation.The test results show that compared with the existing 5 algorithms,the proposed algorithm has better target enhancement ability and background suppression ability,and also has good robustness to noise.To sum up,this thesis studies the two types of interference,strong edge and point noise in complex backgrounds.By setting different target priors to suppress them,both achieve ideal results;And in order to improve the real-time performance of the algorithm,The construction method of tensors is studied,and a parallel solution strategy of grouping is proposed to speed up the algorithm and reduce the time complexity of the algorithm.