Infrared Small Target Detection Based On Dehazing Enhancement And Tensor Analysis

With the progress of infrared imaging technology in recent years,infrared small target detection has gradually expanded from the military field to traffic,security protection,medical,engineering construction,emergency rescue and many other civilian fields.Most of the current research results focus on single-frame images of small infrared targets or sequences of small infrared targets,with few infrared databases,while focusing on the fusion of image features and detection objects with little research involved.In applications,the inherent imaging mechanism of infrared thermal imaging systems causes problems such as missing color information and poor quality of images,in addition to being limited by imaging distance,atmospheric transmission and other factors,resulting in low contrast,blurred images and poor visual effects of infrared images,similar to those of visible multi-fog images.In terms of detection objects,the complex appearance of small targets and the lack of textural shapes make them often submerged in complex backgrounds(e.g.thick clouds,buildings,trees and bright radiation sources).To this end,this paper focuses on the migration application of visible image defogging algorithms to infrared images from the basic characteristics of thermal imaging cameras,images and detection objects,and combines this work with tensor analysis theory to carry out research work summarized as follows:First,to address the problems of low contrast,blurred images and poor visual effects of infrared images,an infrared image clarification algorithm based on a defogging enhancement model is proposed.Inspired by the defogging of visible images,combined with the characteristics of infrared images,and based on the atmospheric scattering model as the theoretical basis,an improved dark channel a priori algorithm is used to remove the components causing infrared image blurring and achieve infrared image clarification,thus improving the visual effect of the original infrared image.Second,to address the problem of few infrared databases,infrared thermal imaging equipment and UAVs are used to construct infrared small target datasets.To improve the detection accuracy and reduce the complexity of the algorithm,an infrared small target detection algorithm based on defogging enhancement and tensor analysis is proposed.Firstly,an improved dark channel algorithm is used to defog and enhance the infrared image to improve the image clarity and enhance the low-ranking of the image,then the matching tensor frontal slices are screened to construct an infrared block tensor model,and the small target detection is transformed into a study of tensor recovery under the framework of tensor singular value decomposition;finally,a fast algorithm is used to solve the objective function.The proposed defogging model effectively solves the "fogging" problem in IR images,enhances the image low-ranking,and improves the local contrast and sharpness significantly compared with the original image.Experiments demonstrate that the proposed algorithm can effectively suppress background clutter,achieve better detection accuracy in both the constructed and public datasets,reduce the false detection rate by33% in complex backgrounds,and have good detection performance in bright background areas,indicating that the algorithm is suitable for complex scenes and eliminating potential false alarm points.