Szczegóły publikacji
Opis bibliograficzny
Vehicle classification based on multi-frequency impedance magnetic profiles in distance domain / Zbigniew MARSZAŁEK, Filip HALLO, Dominik RZEPKA, Krzysztof DUDA // IEEE Transactions on Intelligent Transportation Systems ; ISSN 1524-9050 . — 2026 — Publikacja first online, s. 1–10. — Bibliogr. s. 9–10, Abstr. — Publikacja dostępna online od: 2026-06-08
Autorzy (4)
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 170289 |
|---|---|
| Data dodania do BaDAP | 2026-09-29 |
| Tekst źródłowy | URL |
| DOI | 10.1109/TITS.2026.3697936 |
| Rok publikacji | 2026 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Czasopismo/seria | IEEE Transactions on Intelligent Transportation Systems |
Abstract
Vehicle magnetic profile (signature) acquired from inductive loop embedded in a road is a well-established signal for vehicle classification. This paper demonstrates how classification errors can be almost entirely suppressed by the Multi-Frequency Impedance Measurement (MFIM) system along with the proposed application of data fusion and artificial neural network. Two slim and two wide inductive loops operating at different measurement frequencies are employed to acquire the magnetic signatures, extracting not only the reactance but also the resistance component. The influence of vehicle speed is eliminated by scaling the magnetic profile from the time domain into the distance domain. This allows efficient pattern learning application directly on the signature samples, without the necessity of feature extraction. Furthermore, a comprehensive optimization of the classification process is presented, including the impact of signal resolution in distance domain. It is confirmed, by means of extensive tests, that the number of samples, that are the classification features, in distance domain can be reduced significantly up to 40 times, in respect to original 1 cm resolution, and still a high classification accuracy is preserved. Data fusion from different channels, i.e. different inductive loops and measurement frequencies, is investigated using an artificial neural network classifier. Finally, an energy-efficient spiking neural network is employed, and tested.
