Research On Real-time Techniques Of Infrared Ship Target Detection In Complex Scenes

Infrared image auto target recognition is a hot issue studied by many scholarshome and abroad, and has a wide range of applications in military fields and manyaspects of national economy. About the issue, infrared ship auto detection under sea-skybackground is one of the important research topics in infrared auto target recognitionfields. Due to the complicity of scenes, the heavy interferences of clouds and sea waves,it is quite difficult to detect infrared ship targets automatically. The dissertation carriesout researches around this topic and is composed of three parts as follows:In Part I, sea-sky-line extraction techniques in complex scenes are studied. Insea-sky background images, sea-sky-line is an important image feature. Exactpositioning of sea-sky-line can help to improve the effect of target detection and raisethe efficiency of the running algorithm. In the light of the characteristics of complexscenes that SNR is low and that cloud&wave interferences are heavy, a sea-sky-lineextraction method based on fuzzy synthetical judgment is proposed using fuzzymathematical tools to synthesize several fuzzy features of sea-sky-line for judging. Thismethod increases the accuracy rate of sea-sky-line extraction under low SNR and clutterinterference conditions.In Part II, ship target detection techniques in complex scenes are studied. Aiming atthe characteristics of ship target regions near the sea-sky-line, an improved method ofbackground suppression based on multi-orientation gradients, a two-level iterativemethod of fast segmentation for ship target and a false alarm elimination method basedon binary tree multi-level judgment are proposed. These methods can effectively solvethe problem of sea-sky-line and sea clutter interferences in complex scenes, and detectship target fast and exactly.In Part III, the real-time application techniques for infrared ship target detectionalgorithm are studied. Aiming at the characteristics of infrared serial images that thedata amount is large and frame frequency is high, on the basis of manufacturing highlyeffective target detection algorithm, high-speed processor with large memory is chosenand miniaturized single DSP+FPGA information processing hardware system isdesigned. Through optimization of both software and hardware, the purpose forreal-time processing is attained.