Font Size: a A A

Research On The Application Of Deep Learning Network To Radar Target Recognition Technique

Posted on:2017-05-13Degree:MasterType:Thesis
Country:ChinaCandidate:J Y XiaFull Text:PDF
GTID:2428330569498757Subject:Information and Communication Engineering
Abstract/Summary:PDF Full Text Request
Target recognition based on radar is crucial to space target recognition.The manners of extracting features in target classification of the space environments and threat scenarios is complex and dynamic,which requires that space target recognition technology can automatically adapt to these complex and diverse needs.Existing radar target recognition methods,however,usually extract features based on a priori knowledge,and then build a feature recognition algorithm based on these hand-craft features.It is well known that the performance of the feature extraction is greatly dependent on the prior knowledge,which makes it difficult to meet the requirements of the actual combat environment of change and complexity.In order to solve the above problems in radar target recognition,this thesis firstly analyzes the characteristics of specific recognition scene extraction in radar target signal sequence data,designs adaptive feature learning algorithms from radar target data,extracts discriminative features,and builds the classifiers based on these features.Its main contributions are summarized as follows.1.Based on the experiments on three airplane targets HRRP data and three orbit targets RCS data,we verify the learning ability of RBM network on extracting separable features,obtaining significant separability of the learning results.2.In view of the defects of traditional radar target recognition process and feature extraction method,we construct the RBM model to unify the procedures of feature extraction and target classification.The experimental results show that the recognition accuracy is significantly superior to the traditional recognition methods and the deep learning methods.3.In view of small and noise data in radar target recognition,we propose a fuzzy RBM model,and validate its superiority over general RBM network.In specific,our algorithm improves the recognition rate on this task from 67% to 90%.Moreover,the recognition performance of the proposed fuzzy RBM network in the 5-20 dB noise environment is also conducted,and the results show its robustness to the noise environment when compared with traditional RBM model.4.We construct a recognition process based on the Fuzzy discriminative RBM model,and validate its effectiveness in the presence of multiple targets automatic classification and supervised classification on fifteen types of space orbit targets group original radar echo signal data.These research results not only have the innovation in the algorithm theory,but also provide some guidelines for practical applications,which has important guiding significance to the development of radar target recognition.
Keywords/Search Tags:Deep learning, Restricted Boltzmann Machine(RBM), Feature extracting, Target recognition, Fuzzy function
PDF Full Text Request
Related items