Infrared False Alarm Source Detection Based On Spatial-temporal Tensor Model

Infrared imaging target detection is one of the important detection modes for earth observation and space-based early warning system.In order to improve the accuracy of infrared target detection,detecting false alarm sources has become an important auxiliary mean.The scenario in infrared image is complex and changeable,and the forms of false alarm sources are different.It is difficult to build a detection model to detect different kinds of false alarm sources,or different forms of the same kind false alarm sources.Although the imaging of infrared false alarm sources is different,and different false alarm sources have different morphological textures,the infrared false alarm source images have the characteristics of low rank and sparseness.Therefore,it can be considered to combine tensor recovery theory and robust principal component analysis to build a false alarm source detection model.In this thesis,combined with the prior information extracted by the structure tensor,or the mask generated by the prior information,and then combined with the spatial-temporal tensor model,the infrared false alarm source detection method based on spatial-temporal tensor is established.The problem of false alarm source detection is transformed into a problem of establishing an optimization model for numerical solution.In addition,this thesis also studies the false alarm source classification algorithm.Considering the characteristics of infrared data,an infrared false alarm source classification scheme based on support vector machine is adopted.The research content of this topic is mainly composed of the following parts:(1)The imaging characteristics of infrared false alarm sources are analyzed,and the basic theory of visual detection and classification of false alarm sources is studied,such as morphological processing,matrix tensor recovery,tensor expansion and singular value decomposition.Finally,the robust principal component analysis is introduced into the low rank sparse decomposition of tensor.(2)An infrared false alarm source detection method based on spatial-temporal tensor model with fractal feature constraint is proposed.In order to solve the problems of inaccurate rank representation of background and poor sparsity of false alarm source in traditional methods,a non-convex tensor rank substitution based on Gamma function is introduced.The prior information is extracted by the structure tensor based on fractal features,and the false alarm source is detected by combining the spatial-temporal tensor model.Compared with other basic methods,the proposed algorithm has better detection ability.(3)An infrared false alarm source detection method based on visual saliency mask combined with non-convex tensor rank substitution is proposed.Aiming at the problem that the fractal feature can not represent the large volume of false alarm source well,the non-convex tensor rank substitution in Laplace function is introduced.By calculating the visual saliency of the false alarm source image,a mask containing the false alarm source region is obtained.At the same time,the construction method of spatial-temporal tensor is improved to make it more consistent with the characteristics of infrared images.Experiments show that the proposed model has a significant advantage in detecting large false alarm sources.(4)An infrared false alarm source classification algorithm based on support vector machine is proposed.Using the characteristics of samples,multi-classification support vector machine is trained to classify false alarm sources.Considering the characteristics of the infrared false alarm source data set,texture features are used to construct feature vectors to achieve high-precision classification of false alarm sources in complex scenes.