Small Dim Target Detection Technology Of Long Distance Imaging And Performance Evaluation

In recent years, the small dim target detection over a long distance is becoming more and more popular. In the modern war, the infrared sensor is widely used, but its limitation is also obvious-- a short range. By improving the performance of detection algorithm can achieve the goal that increase the infrared sensor’s work distance. The difficulties of small dim target detection over a long distance lie in: small target area, low signal-to-noise ratio, complex background.On the basis of the full study about aerial infrared small dim target image characteristics, this dissertation did in-depth research about the small dim target detection method based on background suppression, and discussed the performance evaluation method of infrared small dim target detection algorithm. The main research is as follows:1. Add the image data generated section to expand the scope of this dissertation’s data, in order to test the performance of algorithms within a full range of data, but also to overcome the bottlenecks arising from the lack of real data.2. Provide an overview of existing infrared dim target background suppression algorithm, and make a theoretical analysis of the typical infrared small dim target background suppression algorithm(background suppression algorithm based on mathematical morphology, background suppression algorithm based on the high-pass filtering and background suppression algorithm based on wavelet transform, etc.).Make a analysis about the algorithm’s advantages and disadvantages through the experimental method. On this basis, this dissertation improved the background suppression algorithm based on wavelet transform for the application background, in conjunction with preprocessing algorithm based on cubic facet model. Finally, construct a testing process as this algorithm for structure to deal with the aerial infrared small dim target image.3. Introduce the SVR(Support Vector Regression, SVR) theory, and evaluate the four infrared small target detection algorithm performance with SVR. Do research about the dependence curve between background of parameters(weighted contrast, the weighted information entropy and angular second moment and local SNR) and algorithm performance indicators(the probability of detection and false alarm probabilities); and a new method was proposed to predict the dependence surface of any two background parameters and detection probability, false alarm probability. The key is using a higher dimensional feature matrix to train SVR model. Finally, verified its feasibility through experiment and given the application boundary of the method.