Research On Radar Signal Sorting And Recognition Method Based On Machine Learning

With the rapid development of electronics and information technology,electronic reconnaissance has become a key factor affecting the direction of war.Electronic reconnaissance undertakes very important tasks in the process of electronic warfare,such as intercepting enemy radiation source information,battlefield reconnaissance,etc.,all require the cooperation of electronic reconnaissance.It is more important to recognize the different operating modes of the new system phased array radar that are produced on the battlefield.This paper mainly studies the problem of radar pulse sorting and recognition based on machine learning.This article goes deep into each working module of phased array radar,and generates pulse descriptors for different working modes intercepted by the electronic reconnaissance system based on the workflow of each module.The clustering algorithm preprocesses the generated data set to complete the radar pulse dilution and preprocessing.Finally,through further processing combined with neural network-based self-encoder and classifier to complete the signal classification.First,based on the work flow of phased array radar,it simulates its antenna module,wave position arrangement module,resource scheduling module and other modules.Complete the phase dimension array radar in different working modes(search while tracking,search and tracking)pulse descriptor character dimension model,as the basis of subsequent pulse sorting and recognition.Next,the clustering algorithm was used to dilute and preprocess the complex pulse stream for different working modes of the phased array radar and the pulse data set of the traditional radar.The article introduces the basic principles of K-means clustering method and cluster density method based on neighborhood density grid,and discusses the choice of clustering algorithm and the selection of feature parameters through experiments.Finally,the characteristics of pulse width and angle of arrival are selected to perform single-part radar pulse extraction using neighborhood density grid clustering algorithm.Finally,the clustering results are further analyzed,the pulse sequences are sorted according to time,and the clustering results are window-divided by means of windowing.Each dimension of the pulses in each window is averaged or normalized to form features vector.The sparse autoencoder based on neural network is used for data dimensionality reduction and further feature extraction.This paper first optimizes the various parameters of the entire network,then comparatively analyzes the classification effects of Softmax classifier and SVM classifier,and finally selects Softmax classifier as the final classifier.The experimental verification can achieve a recognition rate of 94%.