Aiming at the complex maneuver modes of hypersonic vehicle which are difficult to classify in practical applications, this paper proposes a Sequence-to-Sequence (Seq2Seq) multi-label classification model based on Dual channels Convolutional Neural Network and attention LSTM (DCNN-LSTM) encoder and LSTM decoder. Trajectory data is represented by the multi-label sequence, local and global spatial features of the sequence are extracted using the dual channels convolution structure, and temporal features are extracted using the attention LSTM layer of the encoder. The LSTM decoder is used to generate the multi-label sequence corresponding to the trajectory data. The experimental findings demonstrate that DCNN-LSTM outperforms other deep learning encoder networks in terms of classification performance. In addition, the optimal network structure parameters are determined by analyzing the impact of various parameters on the classification performance. The accuracy of label classification is tested to be more than 85%, which meets the requirements of hypersonic vehicle trajectory classification with complex maneuver modes.