基于深度学习的星载SAR工作模式鉴别

针对传统星载合成孔径雷达（SAR）工作模式反演方法在识别准确率和时效性上存在局限性的问题，根据SAR信号的特点，提出基于一维卷积神经网络的星载SAR工作模式识别模型.该模型以星载SAR信号脉冲峰值幅度作为输入，利用卷积神经网络的自主学习和模式识别能力，避免了传统方法的人为影响因素，能够学习原始信号更具有代表性的特征，最终实现星载SAR工作模式的有效识别.在设计一维卷积神经网络结构时，参考了现有性能较优的卷积神经网络，根据网络训练过程中准确率和损失值的反馈，调整设置了较优的参数以训练得到具有良好识别性能的模型.基于仿真数据的对比实验表明，该模型相较于传统反演方法具有更高的识别准确率，同时对于主旁瓣信号和不同侦收条件均具有较优的鲁棒性和抗噪性.

Operating modes identification of spaceborne SAR based on deep learning

An operating modes identification of spaceborne synthetic aperture radar (SAR) model based on one-dimensional convolutional neural network was proposed according to the timing characteristics of SAR signals, in order to solve the limitation of the recognition accuracy and the timeliness of traditional spaceborne SAR operating modes inversion methods. The impulse peak amplitude of the SAR signal was taken as input, more subtle and representative features of the original signal were learned by using adaptive learning and pattern recognition ability of the convolutional neural network, the human interference factors of traditional methods were avoided, and the effective identification of the operating modes of spaceborne SAR was finally realized. The one-dimensional convolutional neural network structure was designed referring to the existing convolutional neural network with good performance, and the better parameters were adjusted and set to train a model with good recognition performance according to the feedback of the accuracy and the loss value in the training process of the network. Contrast experiments based on simulation data demonstrate that the model has higher recognition accuracy than traditional spaceborne SAR operating modes inversion methods and has excellent robustness and noise anti-noise ability under different types of signals and different detection conditions.