Research On The Detection Technology Of Dim And Small Target For Infrared Image With Large Field Of View Under Low Sky

This dissertation，in the background of the twelfth five-year pre-research topics,under the requirement of research of high-performance infrared target search alertsystem, aiming at real-time detection of dim moving target in low-altitude sky complexbackground for a large field of view image, made a in-depth study on the field of smalltarget detection algorithm as well as proposed a dim and small target detectionalgorithm and target confirmation algorithm; These algorithms are easy to implement,have a relatively small amount of calculation, can be effectively applied to the targetreal-time precision detection in low-altitude sky complex background of this system.The thesis consists of three parts:The first part introduced the research and development status of the large field ofview image, made a in-depth analysis on the characteristics of the target, backgroundand noise using measured data and combing with the generation and radiation theory ofthem; According to their different characteristics, it established the prediction modeland defined performance evaluation rule of detection. Using the collected image, itintroduced the theory of collecting large field of view images of system and analyzedthe characteristics of large field of view image under low altitude sky.Aiming at real-time performance of detection, the second part proposed aalgorithm of target detection based on background’s classification in a large view offiled image based on departing image、entropy segmentation and adaptive thresholdsegmentation for the purpose of predigesting realization and saving amount ofcalculation of algorithm; First, made a in-depth study on commonly backgroundsuppression methods; Then proposed a cursory classification method of the backgroundarea based on departing image in overlapping block and entropy segmentation,realizedthe cursory classification between complex terrestrial background and airspacebackground; Finally, put forward a fine detection algorithm in airspace area based onstructural decomposition and a fine detection algorithm in terrestrial area based onclosing operation and gradient, as well as analyzed cost-effective performance of theterrestrial area’s fine detection. Through the measured data validation, the algorithm iseffective to detect dim and small target in a large view of field image.For the primary purpose to exclude the false alarm left from the second part, thethird part proposed a single-frame target confirming algorithm based on the pointtarget’s grayscale diffusion and dual-core judgment as well as a inter-frame targetconfirming algorithm based on similarity and continuity of the moving target.According to point diffusion effect, the single-frame algorithm designed out-core andin-core to judge contrast and similarity, then the inter-frame algorithm extracted thecharacteristics of similarity of inter-frame target in in-core, improved the continuity judgment based on method of image stream; Combining the both of results, the twoalgorithms could remove the false alarm effectively, the results of measured dataverified the feasibility of the method.