Título: A Deep Learning Model for Heavy Vehicle Classification Based on Silhouette
Autores: Daniel José Cerqueira Brito, Acbal Rucas Andrade Achy, Brenner dos Santos Araujo, Tiago Palma Pagano, Weiner Costa & Mario Sergio Souza de Almeida
Resumo: This paper presents a deep learning model for silhouette-based multiclass heavy vehicle classification using computer vision and machine learning techniques. The objective of this work is to identify heavy vehicle categories from images by their number of axles, as specified by the DNIT guidelines. The model generates reliable and scalable data to support road planning, pavement design, and predictive maintenance of the road network. To achieve this, our proposed methodology combines a pre-trained YOLO model for automated image collection and a ResNet-50 model, fine-tuned via transfer learning. The model’s practical application contributes to highway planning and maintenance by improving the accuracy of pavement load impact prediction. It achieved an Area Under the Precision-Recall Curve (AUC-PRC) of 97% and a validation loss of 0.15, even when faced with a limited and unbalanced dataset.
Palavras-chave: Vehicle classification; machine learning; transfer learning; YOLO; ResNet-50; AUC-PRC; highways.
Páginas: 8
Código DOI: 10.21528/CBIC2025-1184875
Artigo em PDF: CBIC_2025_paper1184875.pdf
Arquivo BibTeX:
CBIC_2025_1184875.bib