AI-Based Vehicle Type Prediction from Magnetic Induction Signal Signatures

Abstract:

Accurate vehicle-type classification is pivotal in intelligent transportation systems (ITS) for traffic monitoring, adaptive signal control, and automated tolling. We present a machine-learning approach that predicts vehicle categories from magnetic induction loop signal signatures, with a specific emphasis on early heavy-vehicle identification to help mitigate road-infrastructure degradation due to overloading. Motivated by Indonesia’s urgent needs in road safety and ODOL (overloading) enforcement, we develop a physics-guided synthetic data generator that simulates loop responses across representative classes under varying speeds, lateral offsets, and noise. We benchmark two models: model A InceptionTime (a strong 1D-CNN baseline) and model B a Physics-Informed TSMixer (PI-TSMixer) that mixes time/channel tokens while injecting physically meaningful cues (distance-domain normalization and axle-pattern hints). On synthetic, stress-tested scenarios, Model B achieves higher macro-F1 and better out-of-distribution robustness than InceptionTime (≈+1.5 pp in-distribution; ≈+4.0 pp OOD), suggesting that lightweight, physics-aware architectures generalize better for loop-based vehicle classification and integrate well with Weigh-in-Motion (WIM) pipelines for Indonesian corridors.

• Periodical: Engineering Headway (Volume 38)
• Pages: 131-140
• DOI: https://doi.org/10.4028/p-Ltn8Cx
• Online since: June 2026
• Authors: Langgeng Asmoro*, Inayatul Inayah, Maman Budiman, Rudy Hermawan, Mitra Djamal
• Keywords: Inductive Loop, Magnetic Induction Signatures, ODOL Enforcement, Vehicle Type Classification

References

[1] J. J. Lamas-Seco, P. M. Iglesias, and F. J. Díaz-Pernas, "Vehicle classification using the discrete Fourier transform with traffic inductive sensors," Sensors, vol. 15, no. 10, p.27201–27214, 2015.

DOI: 10.3390/s151027201

[2] M. Adresi, M. Abedi, W. Dong, M. Yekrangnia, L. Fu, and E. O. Toygar, "A review of different types of weigh-in-motion sensors: State-of-the-art," Measurement, vol. 225, p.114042, 2024.

DOI: 10.1016/j.measurement.2023.114042

[3] Y. Li, A. Y. C. Tok, and S. G. Ritchie, "Deep ensemble neural network approach for FHWA axle-based vehicle classification using advanced single inductive loops," Transp Res Rec, vol. 2676, no. 3, p.1–16, 2022.

DOI: 10.1177/03611981211049142

[4] F. Mocholí-Belenguer, A. Martínez-Millana, A. Mocholí-Salcedo, and V. Milián-Sánchez, "Vehicle modeling for the analysis of the response of detectors based on inductive loops," PLoS One, vol. 14, no. 9, p. e0218631, 2019.

DOI: 10.1371/journal.pone.0218631

[5] S.-A. Chen, C.-L. Li, N. Yoder, S. O. Arik, and T. Pfister, "TSMixer: An all-MLP architecture for time series forecasting," Transactions on Machine Learning Research, 2023.

[6] H. I. Fawaz et al., "InceptionTime: Finding AlexNet for time series classification," Data Min Knowl Discov, vol. 34, no. 6, p.1936–1962, 2020.

DOI: 10.1007/s10618-020-00710-y

[7] K. Zhou, Z. Liu, Y. Qiao, T. Xiang, and C. C. Loy, "Domain generalization: A survey," IEEE Trans Pattern Anal Mach Intell, vol. 45, no. 11, p.4396–4415, 2023.

DOI: 10.1109/tpami.2022.3195549

[8] Y. Shi, X. Ying, and J. Yang, "Deep unsupervised domain adaptation with time series sensor data: A survey," Sensors, vol. 22, no. 15, p.5507, 2022.

DOI: 10.3390/s22155507