Font Size: a A A

Low-resolution Radar, Ground Target Recognition

Posted on:2010-03-25Degree:MasterType:Thesis
Country:ChinaCandidate:L B WenFull Text:PDF
GTID:2208360275998812Subject:Signal and Information Processing
Abstract/Summary:PDF Full Text Request
Target identification has important practical significance and application of value in the current radar system, and it has received intensive attention at home and abroad in recent years.Radar target identification of two major tasks: First, feature extraction, which is a key technology of target identification; followed by the target classification, it has a direct impact on the target recognition rate. Through a variety of signal processing methods, this dissertation explores ways of ground target identification of the low-resolution radar by closely combining multidisciplinary knowledge with specific engineering application.The research work of this dissertation is divided into the radar data preprocessing, feature extraction and target classifier. The radar data preprocessing mainly analyses the radar signal form, and introduces the moving target detection based on tow-dimensional FFT and some of the clutter suppression measures, moreover, through the constant false alarm rate processing and target merger we obtain the number, distance and spectrum information of the targets, Finally, a brief introduction of the Picket-fence effect, which is brought about by discrete Fourier transform, and its suppression methods. The feature extraction methods based on RCS, spectrum entropy estimation, short-time Fourier transform and instantaneous frequency are researched in the main sections of the target feature extraction. At the same time, the stability and reliability of the truck, motorcycle and person features are discussed in different SNR, at different range, different processing time as well as different coherent time, and finally the advantages and disadvantages of above four features are compared by the distance divisibility measure which is a method of pattern evaluation. Through the neighbor classifier and improved neighbor classifier, support vector machine (SVM), the classification combined with real radar data of trucks and motorcycles, persons is the major section of the target identification. And the results of different processing time and different coherent time are analyzed. The average recognition rate of the targets achieves 88% by using improved neighbor classifier, and achieves 90% by using SVM classifier.
Keywords/Search Tags:feature extraction, target classification, RCS, short-time Fourier transform, instantaneous frequency, nearest neighbor classifier
PDF Full Text Request
Related items