Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Asphalt Pavement Crack Identification based on Two-Stage Training and Multi-Branch Model Integrating Multiple Attention
Due to the high labeling cost, there are few available labeled data in the transportation field. At the same time, the characteristics of asphalt pavement cracks are not obvious and the similarity of different categories is high, which makes the task of asphalt pavement crack identification and classification more difficult. In order to realize the rapid and accurate identification of cracks in large-scale asphalt pavement data, a two-stage training method of self-supervised pre-training and supervised fine-tuning and a multi-branch network integrating multiple attention are proposed in this paper. Experiments show that our method can significantly improve the accuracy of asphalt pavement crack classification.
Asphalt Pavement Crack Identification based on Two-Stage Training and Multi-Branch Model Integrating Multiple Attention
Due to the high labeling cost, there are few available labeled data in the transportation field. At the same time, the characteristics of asphalt pavement cracks are not obvious and the similarity of different categories is high, which makes the task of asphalt pavement crack identification and classification more difficult. In order to realize the rapid and accurate identification of cracks in large-scale asphalt pavement data, a two-stage training method of self-supervised pre-training and supervised fine-tuning and a multi-branch network integrating multiple attention are proposed in this paper. Experiments show that our method can significantly improve the accuracy of asphalt pavement crack classification.
Asphalt Pavement Crack Identification based on Two-Stage Training and Multi-Branch Model Integrating Multiple Attention
Wang, Yue (Autor:in) / Mo, Jiadi (Autor:in) / Yu, Zhi (Autor:in) / Wang, Yangyang (Autor:in) / Yan, Shoujing (Autor:in) / Cao, Xiaochun (Autor:in)
27.05.2022
635964 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch