To address the shortcomings of the existing smartphone user authentication methods, this paper proposes a continuous identity authentication method based on adaptive weight feature fusion. A convolution neural network is designed to extract and fuse the deep features of user behavior information data obtained from the built-in sensors (accelerometer, gyroscope, magnetometer) of mobile phones. In the network, three kinds of sensor features are extracted from the three sub-network flows respectively, and weighted fusion is performed in the feature fusion layer. The weight of each feature is adaptively assigned according to the contribution of different features in the network learning process. After feature selection for fused features, one-class support vector machine is used for user classification and authentication. The experimental results show that this method achieves an equal error rate of 1.20% for different user authentication. Compared with other existing authentication methods, the proposed method demonstrates better authentication accuracy.