基于域适应神经网络的调制方式分类方法

针对深度学习进行调制方式识别领域测试样本与训练样本存在分布差异的问题，提出了基于域适应神经网络的调制识别方法。首先采用VGG16深度卷积神经网络提取信号小波变换后系数图像特征；然后利用自编码器对高维特征进行降维处理；再计算训练样本特征与测试样本特征之间的CORAL损失；最后联合优化分类损失和CORAL损失使模型达到最优。通过仿真实验证明，在信号类别存在差异或信道环境存在差异的条件下，引入域适应技术可提高待测信号识别准确率5%以上。

A Modulation Recognition Method Based on Domain Adaptive Neural Network

In order to solve the problem of the different distribution between the testing samples and training samples, this paper proposes a modulation recognition method based on the domain adaptive neural network. Firstly, VGG 16 deep convoluted neural network is utilized for extracting the features of wavelet transform images.Then the high dimensional features are reduced by using the auto encoder, and the CORAL losses between training samples and testing samples are calculated.Finally, the optimal classification loss and the CORAL loss are combined to optimize the model.The simulation results show that under condition of different signal categories or different channel environments, the recognition accuracy of signals tested can be improved more than 5% by introducing domain adaptation technology.