The Methods Of Dim Target Detection And Performance Evaluation

The detection of small target in infrared image has been the subject of intense investigation in pattern recognition. Normal methods based on the structure of the target are not applicable because there is no information of the shape or the structure as the image of the target is very small in the detector. So the detection of small target is of great difficulty.There are a lot of algorithms developed to solve this problem. But the question is there is not a method to analyze and compare quantitatively the performances of the methods, in or not in the same background. To solve these problems, the research on the method to evaluate the algorithms is necessary to be carried out.In this paper, algorithms applicable for the detection of small target in Gaussian noise are researched, and the performances of these algorithms are also evaluated.First of all, the signal-to-noise ratio (SNR) algorithm and the general signal-to-noise ratio algorithm are presented based on the binary hypothesis and maximum likelihood ratio test, and take the probability of false alarm and detection as the performance index.Secondly, the multi-band SNR and multi-band GSNR algorithms are researched based on the detection algorithms in single band.Finally, the framework evaluating the algorithms is built up. Sets of testing images are gained by emulating the imaging of the small target. By comparing the performance of the algorithms applicable for the detection of small target, the results of the analysis are presented, and the optimum design method is also presented.