Research On Target Recognition Based On Radar Echo

Target recognition technology has achieved remarkable results in many fields.The primary task of target recognition is to obtain target information.Compared with other sensors such as cameras,radar has obvious advantages in obtaining target information.These advantages can be applied not only to the civilian industry,but also to the military field.Therefore,target recognition based on radar echoes is of great significance.Based on the processing of radar echo signals and the optimization of target recognition methods,the following research works have been carried out:1)For radar targets,select the target's amplitude as the recognition feature,generalize the amplitude of the target's instantaneous moment to the amplitude sequence during the complete movement of the target and use the amplitude sequence as the feature for target recognition.LFM continuous wave radar was used to measure the four types of targets in the actual scene: motorcycle,car,drone and pedestrian.A two-dimensional fast Fourier transform and a two-dimensional constant false alarm rate detection method are used to extract the amplitude sequence of the target.2)Based on the extracted radar target amplitude sequence characteristics,stacked denoising sparse autoencoder are used for recognition and the effects of two noise addition methods of denoising autoencoder on the recognition effect are compared.The use of extreme learning machines to replace the Soft Max classification layer of traditional neural networks,to some extent,solves the problem of deep network training speed reduction as the structure deepens.3)For the imbalance of sample data in target recognition,a weighted regularized extreme learning machine based on the sample's own attributes is proposed.By combining the stacked denoising sparse autoencoder and the weighted regularized extreme learning machine,the advantages of both are fully utilized to improve the recognition effect of the entire network.The experimental results show that this method has better recognition performance than the stacked denoising sparse autoencoder and the traditional extreme learning machine.Figure 33;Table 8;Reference 56...