Clutter Suppression In Radar Based On Neural Networks

In recent years,more and more unmanned aerial vehicles(UAVs)are abused for illegal activities or even terrorist attacks.Besides,the frequency of bird strikes is increasing year by year.UAVs and bird are all non-cooperative low-altitude slow small targets(LSSTs).Flying at a low altitude,LSSTs are likely to in the shade of terrain.Apart from that,slow speed and small intensity make it hard to detect and track LSSTs.This is a troubling problem in the radar field.This paper mainly focusing on the LSSTs detection in the urban area or airport.Target detection and clutter suppression can be considered as binary classification of targets and clutter.Therefore,this paper constructs appropriate classifiers based on neural networks to classify targets and clutter.The prime work of this paper is as follows:Firstly,we built the data set through experiments.After analyzing the signal processing flow of radar,we got the information in the output of signal processing and selected the input feature vectors.Then,the input feature vectors were processed by standardization.Secondly,we designed two classifiers based on support vector machines and multi-layer neural networks respectively.After training the two classifiers,we compared the performance of them with the classifier based on logistic regression.Finally,considering that the data set is class-imbalanced,we improved the classifiers with cost-sensitive learning.We analyzed the effect of class-imbalance on classifiers based on support vector machine and BP neural network.Then,we constructed cost-sensitive SVM and cost-sensitive BP neural network classifiers.The same data set is used to compare the performance of the classifiers before and after importing the cost sensitive learning,which proves that the cost sensitive classifier can effectively improve the classification accuracy of the minor but more important class.