Research On Infrared Dim Small Target Detection Algorithm Based On Local Spatial Features And Low-rank Background Constraint

The rapid development of intelligent technology has promoted the rapid update of modern military application technology,and put forward higher requirements for the infrared detection system’s long-distance combat and precise positioning capabilities.Infrared small target detection technology is a key technology that affects the performance of infrared detection systems,and it is a hot technology in the field of computer vision.Due to the low signal-to-noise ratio of infrared images in natural scenes,small targets imaged at long distances usually appear as weak isolated spots.In addition,bright natural backgrounds(such as thick clouds,buildings,etc.),and even sensor noise can drown out small dim targets to be detected.In order to enhance the robustness of small target detection algorithms under highly heterogeneous background clutter,it is crucial to study effective heterogeneous background suppression schemes and algorithms to extract weak and small targets.In recent years,the research results of infrared small target detection algorithms have emerged in an endless stream,but they still show certain limitations when faced with strong background clutter in real infrared conditions.In this paper,based on the low-rank prior of the infrared background,combined with the theoretical basis that small targets have more obvious characteristics in the local space,three infrared small target detection algorithms under complex backgrounds are proposed.The main research contents are as follows:1.In view of the current situation that it is difficult to completely suppress the interference of the infrared small target detection algorithm on the bright natural background,an optimal spectral scale space algorithm for infrared small target detection is proposed,including background suppression stage,feature matching stage and optimal scale selection stage.In the background suppression stage,the matrix decomposition method of the Inexact Augmented Lagrange Multiplier(IALM)algorithm is used to extract the sparse image matrix from the original image as the target foreground image.In the feature matching stage,an accurate scaling strategy is proposed,which uses 16 fine Gaussian kernel functions to convolve with the magnitude spectrum of the target foreground image to obtain a multi-scale saliency map that accurately matches the features of small infrared targets.In the optimal scale selection stage,the previous method of locating small targets with minimum information entropy was abandoned.According to the features of infrared small targets in local space,the local information entropy was redefined as the optimal scale selection method.Experimental results verify that the proposed method can suppress the interference of thick clouds and high-brightness backgrounds in a targeted manner.2.In order to solve the problems of slow running speed and edge sensitivity in the low-rank decomposition method to deal with the global infrared image,a sparse infrared small target extraction algorithm based on the rapid localization of gradient peaks is proposed.It mainly includes the following steps.First,the entire image is expanded by using circular structural elements to sharpen the edges of small objects and smooth the background noise.Then,using the theoretical prior that the gradient characteristics of the small infrared target in the local space are more obvious,the overlapping gradient information of the infrared image after expansion is calculated,and the local area with a large gradient peak is located in the original image,and it is identified as a candidate target area of ??interest.Finally,the extracted local region of interest is decomposed into low rank and sparse by the accelerated proximal gradient algorithm,and the sparse infrared small targets are extracted.The experimental results show that the algorithm runs faster than the baseline low-rank sparse decomposition algorithm,the image signal-to-noise ratio is greatly improved,and it shows superior detection performance in the comprehensive evaluation index.3.Aiming at the deficiency of the current infrared image block algorithm for detecting small targets in complex backgrounds,a robust infrared superpixel image separation model is proposed.It mainly includes the following steps.First,superpixels generated by the simple linear iterative clustering algorithm appropriately aggregate similar background components as the basic unit for subsequent processing.Specifically,this work sets an optional superpixel number range for SLIC to implement the proposed strategy robustly on low-resolution infrared images.Second,an outlier superpixel masking model is proposed to eliminate outlier superpixels with structural and non-structural extremely complex background interference components.Subsequently,a2 D Gaussian matched filter is specially introduced to blur the boundary,maximize SNR and PNHB.Finally,a singular value truncation strategy based on entropy-weighted sparsity factor(SVT-EW)is proposed to finally extract the target accurately.The entropy-weighted sparse weight factor assigns the appropriate sparse weight to the infrared small target image to achieve accurate extraction of the small target.Therefore,the threshold segmentation step acting on the background residual is omitted,and it is shown higher detection performance in the comprehensive evaluation with the comparison algorithm.