Research On Small Target Detection For Space-based Infrared Image

With high penetration and detectability,infrared imaging has been widely applied in remote sensing,aerospace,surveillance and other fields.As the imagery distance is usually far,the target generally appears as a small speckle in infrared images,lacking of available image information and seriously influenced by complex background and inevitable noises during imaging process,which brings great challenges to target detection.This paper concentrates on the problem of small target detection for space-based infrared image,investigates the imaging models and imaging characteristics of single sampling system and scanning oversampling system,and proposes target detection method respectively.The main works of this paper are as follows:In Chapter 2,the imaging models and imaging characteristics of infrared single sampling system and scanning oversampling system are studied.As imaging characteristics of target and background differ greatly for ranged detection system,which performs a significant impact on target detection,therefore this chapter mainly analyzes the single sampling system and TDI(Time Delay Integration)scanning oversampling system,builds corresponding imaging models and simulates the imaging characteristics of targets and background respectively.In Chapter 3,the point target detection method utilizing multi-label Markov Random Field(MRF)based on target similarity measure is studied.Conventional binary-label MRF-based detection methods generally regard the detection as a problem of binary image labeling,winner-take-all discrimination commonly leads to the accumulation of errors during the iterative labeling process,which affects the detection performance remarkably.For this problem,this chapter introduces target similarity measure in the image labeling process,formulates multi-label generative MRF model and realizes the detection of targets utilizing the adaptive pointwise filter based on target similarity measure,which enhances the adaptability to complex background,reduces the false alarms and improves detection performance effectively.In Chapter 4,the point target detection utilizing super-resolution strategy for infrared scanning oversampling system is studied.The specially designed double-alignment detector in oversampling system leads to oversampling in both spatial and temporal domain with quadruple image information,which is beneficial for target detection.Concerning how to utilize the quadruple information,existing methods mainly investigate the aliasing characteristics of the scanning oversampling mechanism,however the aliasing accounts for the degradation of image information,leading to quadruple image information not effectively utilized.For this problem,this paper proposes a point target detection method utilizing super-resolution strategy,realizes de-aliasing and resolution enhancement through super-resolution process.Meanwhile,the designed adaptive MRF-based regularization utilizing the multi-label MRF model proposed in Chapter 3 preserves and further aggregates the target energy superiorly,improving the detection performance remarkably.