Infrared Small Target Detection Technology Based On Multi-feature Extraction

Compared with visible light imaging and radar imaging systems,infrared imaging systems have many advantages.They are more penetrable,in addition,they can work all weather and have a long operating distance.Therefore,they play a vital role and have been widely applied in both the military and civilian fields,such as precision guidance,medical imaging,and transportation management systems.In modern warfare,both sides of the war conduct identification and counter-identification through various means.Therefore,detecting targets from long distance early is a guarantee for other tasks such as identification and tracking.As a key technology,infrared small target detection has undoubtedly become a research hotspot.Achieving accurate detection of small targets from long distance can provide more time for subsequent processing.With the distance increasing,infrared small targets usually lack details and texture information,and they always appear the characteristics of weak and small in the infrared images.Furthermore,the complexity of the background can also cause the targets to be easily overwhelmed.Although a lot of researchers have already carried out this subject,there are still many problems to be solved urgently.This thesis discusses the existing issues.The main research contents are as follows:(1)Aiming at the shortcomings of the existing infrared small target detection methods based on local prior and nonlocal prior,this thesis proposes a novel infrared small target detection method using both local and nonlocal prior information,which improves the shortcomings that the typical local priors-based methods are easy to misclassify some not very bright but isolated pixels as targets,and the classical nonlocal priors-based methods easily mistake some pixels in the strong edges as targets.Firstly,the proposed method utilizes dual-window to obtain the contrast map,which can suppress background clutter.Then,a multiscale sparse and low-rank decomposition method is applied to divide the contrast map into sparse target image and low-rank background image.Finally,adaptive threshold segmentation is applied.The validity of the proposed method is proved by comparing with different methods in multiple scenarios.(2)In view of lacking texture and detail information in infrared images,which makes small targets easily inundate in complex background,this thesis introduces the method of construct max-tree in mathematical morphology to infrared small target detection.The analysis found that in the infrared image,when the detail information is insufficient,the spatial information is a good supplement.Therefore,this thesis proposes a new infrared small target detection method by associating max-tree construction with sparse and low-rank decomposition method.First,the spatial information is extracted based on the max-tree,then the sparse and low-rank decomposition method is used to obtain the target image and the background image.Finally,the threshold segmentation is performed to obtain the final result.The comparison with different types of methods in multiple scenarios proves the adaptability of the proposed method.