Research On Radar Target Classification Recognition Technology Based On Convolutional Neural Network

Radar target classification and recognition is an important research direction in the field of radar information processing,no matter which has a wide range of requirements in both military and civilian fields.The existing classification and recognition methods based on feature extraction of radar target echo mostly use manual extraction of single-dimensional features or fusion features,which mainly rely on subjective factors and do not make use of all the information of target echo.The recognition rate is low in the complex clutter environment and under the condition of low single-to-noise ratio,so it is difficult to meet user needs.In addition,another type of classification and recognition method based on radar imaging has the disadvantages of partial information loss of original target,complex process,special requirements for radar equipment and high cost,which is not conducive to realization.In view of the above problems,this paper uses the complete high-resolution Doppler measured data of the original radar echoes to propose a radar target classification and recognition method based on Convolutional Neural Network(CNN),which is designed on a general-purpose computer platform,and a Python-based target classification and recognition software module has been implemented to realize automatic classification and recognition of human and vehicle targets.The specific works are as follows:First,a one-dimensional CNN model is constructed using the complete high-resolution doppler data,and the network structure and initial parameter configuration are given according to the previous experiments.Then,experiments were carried out on the effects of different activation functions,different optimizers,different network layers and different convolution kernel sizes on the recognition performance of the CNN model.The CNN model was optimized and adjusted by analyzing the effects of various factors on the accuracy and loss of network model training.Simulation results show that,compared with the existing classifiers,the CNN designed in this paper can learn useful features from high resolution doppler data and achieve a high accuracy in the classification and recognition of human and vehicle targets.On the basis of the above research,according to the requirements of practical application scenarios,an engineering implementation scheme of radar target classification and identifier based on CNN is designed.Firstly,the position and function of the radar target classifica-tion and recognition software module in the whole radar system platform are defined,and the design idea of the software framework and the concrete realization process of the radar target classification and recognition module are elaborated in detail.Secondly,the multi-task software framework for radar target classification and recognition based on general computer platform is built,including message data driver,data message sending and receiving process,data serialization and deserialization processing,data classification and recognition processing and output result process,and the online target classification and recognition software module is realized.The simulation verification and the implementation of engineering software show that radar target classification recognizer based on CNN model designed in this paper has complete feasibility and high recognition performance,which can provide support for the practical target classification and recognition function of the subsequent radar equipment applications.