一种改进的三通道毫米波雷达目标识别方法OA

To address the problems of single feature and low recognition rate in single-channel range-Doppler(RD)spectra used in convolutional neural network(CNN)target recognition,a three-channel CNN-based radar target recognition method is proposed and a complete multi-target classification pipeline is established in this paper.First,a two-dimensional fast Fourier transform algo-rithm and a vector mean cancellation algorithm are jointly employed to process the radar echo signals to achieve the original RD spectral data.Subsequently,three feature channels,namely for the original target range information,the RD spectrum after elimi-nating multi-target interference,and the RD spectrum after processing by constant false alarm rate detection,are extracted on this basis,effectively enhancing the target feature dimensions.Finally,the three-channel feature maps are fed into the improved light-weight S-MobileNet network to accomplish radar target feature learning and classification decision.Comparative experiments on re-al-world datasets show that the proposed three-input channel method can significantly improve the accuracy and robustness of target recognition in complex scenarios.Meanwhile,the improved network further reduces the model complexity and computational cost while maintaining a lightweight structure,making it valuable for engineering applications such as automotive radar.