Citation

Sun J, Su J, Yan Z, Gao Z, Sun Y, Liu L. Front. Neurosci. 2023; 17: e1243847.

Copyright

(Copyright © 2023, Frontiers Research Foundation)

DOI

10.3389/fnins.2023.1243847

PMID

37638309

PMCID

PMC10448051

Abstract

Efficient and reliable transportation of goods through trucks is crucial for road logistics. However, the overloading of trucks poses serious challenges to road infrastructure and traffic safety. Detecting and preventing truck overloading is of utmost importance for maintaining road conditions and ensuring the safety of both road users and goods transported. This paper introduces a novel method for detecting truck overloading. The method utilizes the improved MMAL-Net for truck model recognition. Vehicle identification involves using frontal and side truck images, while APPM is applied for local segmentation of the side image to recognize individual parts. The proposed method analyzes the captured images to precisely identify the models of trucks passing through automatic weighing stations on the highway. The improved MMAL-Net achieved an accuracy of 95.03% on the competitive benchmark dataset, Stanford Cars, demonstrating its superiority over other established methods. Furthermore, our method also demonstrated outstanding performance on a small-scale dataset. In our experimental evaluation, our method achieved a recognition accuracy of 85% when the training set consisted of 20 sets of photos, and it reached 100% as the training set gradually increased to 50 sets of samples. Through the integration of this recognition system with weight data obtained from weighing stations and license plates information, the method enables real-time assessment of truck overloading. The implementation of the proposed method is of vital importance for multiple aspects related to road traffic safety.

Language: en

Keywords

automatic weighing station; fine-grained visual categorization; MMAL-Net; overload detection; truck model recognition