To address the issue of insufficient feature information extraction caused by neural network architecture in current encrypted traffic recognition research, this paper proposes a multi-feature fusion encrypted traffic recognition method based on attention mechanism. The proposed method focuses on the hierarchical structure characteristics of encrypted traffic and designs two parallel network branches for feature extraction. Branch one uses residual neural network(ResNet) to extract the original features of traffic, while branch two uses an Inception-CNN composed of irregular-sized convolution kernels to extract statistical features of traffic for characterization and compensate for the information loss caused by traffic cropping. In addition, this paper converts the statistical features from the existing grayscale image to the RGBA image format as input to help the model more effectively extract features. The features extracted by the two branches are merged into a new feature vector and input into the channel attention module for weighting to enhance the representation ability of traffic features. The experimental results show that the proposed model performs better than existing typical encrypted traffic classification methods, with significantly improved accuracy, recall rate, and F1-score, among which the comprehensive performance metric F1-score is increased by an average of 6% compared to existing methods.