DFTrans: Dual Frequency Temporal Attention Mechanism-Based Transportation Mode Detection

Area of Science:

• Computer Science
• Transportation Engineering
• Data Science

Background:

• Diversified transportation leads to vast daily travel data, crucial for transportation mode detection.
• Existing transportation mode detection methods face challenges in accuracy and robustness.
• Accurate detection of transportation modes has applications in various fields.

Purpose of the Study:

• To present a novel transportation mode detection algorithm, DFTrans.
• To improve the accuracy and robustness of transportation mode detection.
• To effectively utilize traffic data for identifying various modes of transport.

Main Methods:

• DFTrans algorithm utilizes Temporal Block and Attention Block.
• Discrete wavelet transforms decompose traffic sequences into low- and high-frequency components.
• A two-channel encoder captures temporal and spatial correlations; CNN inductive bias is used for high frequencies, and Attention Block for low frequencies.

Main Results:

• DFTrans achieved macro F1 scores of 86.34% on the SHL dataset and 87.64% on the HTC dataset.
• The model accurately identifies eight transportation modes: stationary, walking, running, cycling, bus, car, underground, and train.
• DFTrans demonstrates superior performance compared to existing baseline algorithms.

Conclusions:

• The proposed DFTrans algorithm offers improved accuracy and robustness in transportation mode detection.
• The integration of Temporal and Attention Blocks effectively captures essential features from traffic data.
• DFTrans provides a significant advancement in analyzing and classifying diverse transportation modes.