In the previous research on driving behavior, based on multi-source sensor data, the traditional method adopts the method of manually designing features, which has a large number of features and involves many professional fields. Most of the research on multi-source time series sensor data mainly focuses on the extraction of time series features, ignoring the connection between feature channels. Therefore, this chapter uses the GRU-FCN (Gated Recurrent Unit-Fully Convolutional Network) neural network model based on channel attention to study driving behavior recognition. A fully convolutional neural network (FCN) neural network that can automatically extract features is used to replace the traditional manual feature extraction method. At the same time, the Gated Recurrent Unit (GRU) is used to act on multi-source sensor data. Experiments show that GRU is even better than the LSTM model in some tasks, which can greatly improve the training efficiency. Finally, in order to pay special attention to the contribution of certain feature channels to classification and improve the accuracy of classification, this chapter introduces a squeeze-and-excitation block (SE block) to adaptively adjust the weight of each feature channel (channel attention), and recalibrate the features to improve the representation ability of the network.