Research On Infrared Images Preprocessing And Automatic Recognition Of Aerial Targets

Infrared imaging guidance has the virtue of high sensitivity, stronganti-interference, long detection distance, all-weather work, etc. It can satisfy thecombat requirements of complex battlefield environment. So it has widespreadapplication in the missile guidance system. With infrared imaging technology and imageprocessing equipment development, infrared imaging guidance technology has becomeone of the important developing directions of precision guidance technology.The image de-noising, segmentation and target recognition of infrared imagingguidance system are deeply studied in this dissertation. The main works are listed asfollows:(1)Based on the particular analysis of the noise in infrared image, and according tothe characteristics of the infrared image with salt&pepper noise and Gaussian noise,we use different method to de-noising. In order to eliminate the salt&pepper noise ininfrared image, extreme value of the median filter and multilevel median filteringmethod is combined. Based on the existing de-noising method of NSCT transform, weusing the idea of mean filter proposed a new adaptive Bayesian threshold, used toremove the infrared image of Gaussian noise. Experiments show that the two methodscan effectively remove the noise, and protect the edge information.(2)Aiming at the goal of infrared image segmentation, a new thresholdsegmentation method is proposed, which combined with the clustering and OTSUmethod. The experimental results show that the method improves the accuracy of thethreshold, and achieved the better segmentation effect.Due to the introduction of clustering information increased the calculation andreduce the real-time of segmentation algorithm, so this dissertation proposed theimproved PSO algorithm to optimize the segmentation algorithm. The experimentshowed that use the improved PSO algorithm optimize the segmentation algorithm getthe expected effect, and improve the real-time of segmentation algorithm.(3)Based on the research of the multilevel invariant features of Radon transform,used Fourier descriptor to improve the shortcomings of rotation invariance. Weproposed the Radon-Fourier descriptor to identification the air target. Theexperimental results demonstrate that this method performs better in identifying airtargets even with the interference. At the same time, this method has the robustness ofnoise.