【摘要】 近年来，将深度学习应用于调制识别领域是个热门方向，但为了提高识别精度，不断复杂化的网络结构给硬件设备带来巨大压力，提出将MobileNetV2网络应用于调制识别的方法。首先生成11种调制信号的数据集，再利用MobileNetV2网络进行调制识别模型的训练，最后通过全连接层进行11种调制识别的分类输出。实验表明，MobileNetV2的识别率达到95%以上，相较于实验对比的2种卷积网络提高5%左右，且网络参数总量大大降低，训练时间也有所控制，降低了对硬件设备的需求。此方法对后续轻量化深度学习网络在调制识别中的应用有研究价值与意义。更多还原

【Abstract】 In recent years, the application of deep learning to the field of modulation recognition is a popular direction, but the constantly complex network structure puts great pressure on the hardware equipment for raising the recognition accuracy.This paper proposes a method using MobileNetV2 network in modulation recognition.Firstly, the dataset of 11 kinds of modulation signals is generated, then MobileNetV2 network is used to train the modulation recognition model, and finally the classification output of 11 kinds of modulation recognition is performed through the fully connected layer.The experiment shows that the recognition rate of MobileNetV2 reaches more than 95%,which is about 5% higher than that of the two convolutional networks in the experimental comparison, and the number of network parameters is greatly reduced, as well as the training time is controlled, which reduces the requirements to hardware devices.This method has research value and significance for the subsequent application of lightweight deep learning networks to modulation recognition.更多还原