基于CNN特征的宽频段智能测向方法

A broadband intelligent direction finding method based on CNN features

当前电磁环境日益复杂,利用机器学习方法实现快速且精确的宽频段无线电测向逐渐成为研究的热点。使用卷积神经网络基于端到端的方式完成宽频段测向的方法能够在一定程度上解决宽频段相位模糊的问题,但卷积运算后特征维数大大增加,稀疏的特征影响了最后一层全连接前馈神经网络的分类效果。针对这一问题,提出将无线电测向分为特征学习任务和方向预测任务,使用卷积神经网络作为特征提取器,将通过多层卷积运算得到的结果视为二次提取的特征,作为方向预测任务的输入;针对二次提取特征的稀疏性,提出使用主成分分析算法对特征进行降维,并将稀疏性降低后的特征作为后续分类器的输入。此外,针对特征的特点,探索了几种分类模型作为分类器的效果,包括决策树、随机森林、径向基函数神经网络和K-近邻。实验结果表明,使用主成分分析算法对特征进行降维能够提升训练和测试效率;采用K-近邻构成分类器的准确度明显高于原卷积神经网络的准确度;若需要兼顾准确度和测向效率,采用随机森林构成分类器的效果最好。

At present,electromagnetic environment is becoming more and more complex.U-sing machine learning methods to achieve radio direction finding has become a research hots-pot.The existing research work attempts to use convol ution neural network to complete broadband radio direction finding based on end-to-end method.It can solve the problem of wide band phase ambiguity.However,the dimension of the feature after convolution increa-ses greatly,which results in the phenomenon of feature sparse and further,affects the per-formance of the fully connected feedforward neural network to a certain extent.To solve this problem,we divided radio direction finding into feature learning task and direction prediction task,and used convolutian neural network as feature extractor to obtain secondary extractedfeatures,which were the inputs of the direction prediction task.Further,we used the princi-pal component analysis algorithm to reduce the dimension of the features.In addition,we ex-plored the performance of several classification models as the final classifier,including deci-sion tree,random forest,radial basis function neural network and K-nearest neighbor.Theexperimental results showed that using principal component analysis algorithm to reduce thedimension of features could improve the efficiency of training and testing,and the accuracy of K-nearest neighbor classifier was significantly higher than that of the original convolution neuralnetwork.If both accuracy and efficiency were considered,random forest classifier was the best.