Ship Target Recognition Method Based On Radar High Resolution Range Profile

The Radar High Resolution Range Image(HRRP)is the vector sum of the projection of the scattered point echoes of a target on the radar line of sight acquired with a broadband radar signal,which contains a large amount of information about the target.Compared to other radar data,HRRP has the advantage of being more convenient to acquire and process,so HRRP-based target recognition methods have become one of the highly valued research directions in the field of automatic radar target recognition.Traditional HRRP target recognition is limited by researchers’ experience in feature selection and difficulties in classifier design.With the rapid development of artificial intelligence,deep learning-based recognition methods can automatically extract deep features of targets,reducing the complexity of manually designing features in traditional target recognition,and therefore deep neural network-based target recognition methods have become one of the current research hotspots.The main contents of this paper are as follows:I.The basic theory of high resolution range profile is introduced,the imaging principle of high-resolution distance image is analyzed,the causes of orientation sensitivity,translation sensitivity and magnitude sensitivity problems of high-resolution distance image are analyzed and the corresponding processing methods are proposed,and the HRRP target recognition method based on support vector machine,the HRRP target recognition method based on KNN and the deep learning-based HRRP target recognition method based on support vector machine,HRRP target recognition method based on KNN and HRRP target recognition method based on deep learning.Second,an HRRP target recognition method based on multi-head CNN and bidirectional LSTM is proposed.Aiming at the advantage that convolutional neural networks can extract deep features of target data and have certain translation invariant features,and the advantage of recurrent neural networks in processing sequence features,multi-head CNN are used to extract features from pre-processed HRRP data,and multiple CNN with parallel operations are used to extract local features of HRRP data,and the sequence feature data composed of them are fed into a bidirectional LSTM network.And considering the influence of noise,the attention mechanism is added to automatically find the location of the target region and assign larger weights,the proposed model has higher recognition accuracy compared with the conventional method.Third,a new Transformer encoder-based modelling strategy is proposed for ship recognition based on the Transformer model’s excellent ability to maintain serial data relevance and its ability to capture global features.The method first uses the encoder part to model the HRRP data,then uses position encoding to record the position information between different sequences,and finally uses the self-attentive mechanism to model the whole HRRP sequence.The experimental data shows that the proposed model achieves95.98% recognition accuracy,which is better than other models and has excellent recognition performance.Fourth,firstly,the feature extraction and classifier design methods for ship targets in one-dimensional distance image were studied,including structural domain features and transform domain features,and the classifiers used SVM,KNN,classification tree,plain Bayes and LDA.Finally,a test platform was developed using C++ language for experimental validation,and the correct classification rate of the developed test platform for ship targets was verified through experimental data.