Radar has a wide range of applications in both military and civil fields,and the location and detection environment of radar are complex and changeable.Therefore,clutter suppression and target detection in clutter environment have very important research value.Traditional constant false alarm rate(CFAR)detection adapts to the decision threshold according to the clutter statistical distribution.A CFAR detection method can not suit to a variety of clutter environment.For these cases,regards the target detection problem as a dichotomy model of target and clutter,proposes a target detection algorithm based on extreme learning machine(ELM)by combining machine learning.Finally the algorithm is improved to improve the generalization performance of the algorithm.The main content of the article is summarized as follows:1.Making data sets.Standard data sets play an important role in the performance of machine learning algorithms.The working system of the radar is pulse Doppler radar.The training data set of the classifier is constructed by extracting sample features from radar echo data.Firstly,Range-Doppler spectrum is generated by processing radar echo data,then,different methods are used to calculate the range dimension SNR,Doppler dimension SNR and twodimensional SNR of the target and clutter points in the range Doppler spectrum as the sample features,and finally generate the sample data set according to a certain proportion of target and clutter.2.ELM has the advantages of fast computation speed and good generalization performance.The feasibility of using the algorithm in target detection is analyzed,and the principle and process of the algorithm are introduced in detail.A target detection algorithm based on ELM is proposed.Under the same detection background,some sample data sets were selected as training data,and the rest were used as test data to detect algorithm performance.The radar data in the same time period are used as the training set and the test set to detect the algorithm performance.The experimental results show that ELM detects more target points and fewer clutter points by mistake than the traditional constant false alarm.The experimental results show that ELM has better generalization performance than traditional constant false alarm detection.3.According to the detection surroundings of radar is complicated and unpredictable,the distribution of training data and test data is usually uneven.Aiming at this situation,this paper introduces the domain adaptation method in machine learning,and the target detection algorithm of domain adaptation extreme learning machine(DAELM)is proposed.When radar data of different time periods are used as training set and test set respectively,DAELM has better detection effect than ELM.The detection effect of DAELM algorithm is better than ELM algorithm.Further,the distribution difference between source domain and target domain data is reduced by local linear embedding method,and the new data set with reduced distribution difference is input into DAELM.At the same time,the target detection algorithm based on local first embedding is proposed by using covariance regularization term as constraint in the output layer.Finally,the ROC curves of each algorithm are drawn,and the results show that the accuracy of LLE-CDAELM is higher than that of DAELM and ELM.For data sets of different clutter background types,LLE-CDAELM algorithm also has the best robustness,so it can improve the accuracy of classification in the case of uneven sample distribution. |