Study On Feature Extraction And Classification Of Ground Motion Target In Low-resolution Radar

With the development of information technology,radar automatic target recognition has become one of the important development directions of modern radar systems.It plays an important role in the threat classification,tracking,interference and interception of military targets.Review the development history of radar systems,low resolution radar is mainly used for early warning and target detection.At present,high resolution radar is expensive and complex,the radar in active service at home is mainly low resolution radar.Though,it is of great significance to study the preliminary classification and recognition of the target in low resolution radar.This paper focuses on the classification of ground targets in low resolution radar and two technical routes are studied along the target feature extraction and classifier design.The main work and innovation are as follows:1.Aimed to instability of features in a single Coherent Processing Interval(CPI),an objective feature extraction method is proposed based on characteristic probability distribution curve.Firstly,the two basic target feature quantities is extracted,which are relative Radar Cross Section(RCS)and doppler spectrum entropy,using an single Coherent Processing Interval(CPI)of target echo.Then,we combine multiple CPI by sliding window method and calculate characteristic sequence probability density distribution curve.Finally,we extract the most probable target feature by the characteristic probability density distribution curve.A performance test has been carried out on measured radar data.The result shows that the features extracted by our method not only high conservatism but also low computation cost,high real time and convenient for engineering.2.There are differences in the micro-motion of trucks,motorcycles and people.Based on the micro-motion mathematical model,the method of extracting target characteristics using time-frequency distribution is studied.Three feature extraction methods are defined based on the characteristics of time-frequency distribution of different targets.Then the statistical values of each characteristic of radar measured data are analyzed.Support Vector Machine(SVM)is used to simulate the classification performance of the features.The comparison of the average recognition rate based on the measured data shows that the extracted target features can better realize the classification of three kinds of ground radar targets.3.From the perspective of image feature extraction,a target feature extraction method based on time-frequency distribution is proposed.First,transform the time-frequency distribution of target into gray image by transform function.Then,according to the feature extraction method of gray image,the image entropy and the statistical features of the Gray Level Co-occurrence Matrix(GLCM)texture are extracted.Finally,support vector machine classifier is used to classify experiments.The experimental results show that the proposed feature extraction method can effectively classify the three kinds of ground targets and improve the recognition rate of the people.4.From the perspective of radar target classifier design,aimed to difference of characteristic attribute recognition performance in low resolution radar recognition system and the problems of outside target,such as jamming target,false target,etc.Two improved classifiers are proposed.Firstly,the optimization algorithm is used to search the optimal eigen-value and adjust the contribution of different feature attributes to classification recognition.Then the rejection analysis of the test sample set is analyzed by using the rejection algorithm.Finally,the basic classifier is used to classify the output samples.Experiments on measured radar data show the proposed classifier have better classification ability compared with existing methods.5.In order to overcome the limitation of single classifier in radar target classification performance,we propose a multi-classifier fusion method based on Water-Filling Theory.The integration of multi-classifiers can make full use of classification information of different classifiers and avoid the one-sidedness of a single classifier to implement multiple classifiers complementary,improving the classification performance of the system.WFT uses various sub-classifiers to distribute multi-classifier fusion weight in training sample set training.The sample category with high recognition rate is assigned a higher power value coefficient.The classifier with high average recognition rate is also assigned a higher power value coefficient.The experimental analysis of the measured radar data shows that the effectiveness of applying the principle of water flood to multi-classification integrated system.The fusion algorithm can make full use of the prior knowledge of training samples and improve the overall performance of the multi-classifier integrated system.