Probabilistic Neural Networks Application for Vehicle Classification

Abstract

Federal, state, and local agencies use vehicle classification data for planning, design, and conducting safety and operational evaluation of highway facilities. In conformity with federal reporting requirements, most states use the scheme which classifies vehicles based on their axle configurations; primarily the number of axles and the length of axle spacings. However, the scheme is prone to errors resulting from imprecise demarcation of class thresholds. To improve classification, the problem is hereby viewed as a pattern recognition problem in which statistical techniques such as probabilistic neural networks (PNN) can be used to assign vehicles to their correct classes. In this research, the network was trained and applied to field data composed of individual vehicle's axle spacing and number of axles per vehicle. The PNN reduced the error rate by 3.3% compared to an existing classification algorithm. The error rate was further reduced by 6.5% when the individual vehicle's gross weight was added as a classification variable. These results confirm the promise of neural networks in axle classification but the technique still requires additional field validation as well as exploration of additional variables to improve categorization of vehicles into the F or other schemes.