A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Improved YOLOv5 for Pavement Patch Detection Using Deformable Convolution and Bidirectional Feature Pyramid Network
This paper introduces an advanced version of the YOLOv5 model specifically designed for pavement patches detection, incorporating Deformable Convolutional Networks (DCNs) and Bidirectional Feature Pyramid Networks (BiFPN). The primary innovations of this work include the integration of DCNs into the YOLOv5 backbone, enhancing the model's ability to adapt to geometric variations and irregular shapes of pavement patches, and the incorporation of BiFPN to improve multi-scale feature fusion, which is crucial for detecting patches of varying sizes. Extensive experimentation on a diverse dataset of pavement patches reveals substantial improvements in detection performance metrics, including precision, recall, and mean Average Precision (mAP), compared to the standard YOLOv5 model. Detailed ablation studies and error analysis shed light on the specific contributions of DCNs and BiFPN to these performance gains. The enhanced model demonstrates significant potential for advancing automated pavement inspection systems, leading to more efficient and reliable road maintenance. Future research will focus on implementing real-time capabilities, expanding datasets, exploring additional cutting-edge techniques, and developing integrated systems for automated pavement inspection and maintenance.
Improved YOLOv5 for Pavement Patch Detection Using Deformable Convolution and Bidirectional Feature Pyramid Network
This paper introduces an advanced version of the YOLOv5 model specifically designed for pavement patches detection, incorporating Deformable Convolutional Networks (DCNs) and Bidirectional Feature Pyramid Networks (BiFPN). The primary innovations of this work include the integration of DCNs into the YOLOv5 backbone, enhancing the model's ability to adapt to geometric variations and irregular shapes of pavement patches, and the incorporation of BiFPN to improve multi-scale feature fusion, which is crucial for detecting patches of varying sizes. Extensive experimentation on a diverse dataset of pavement patches reveals substantial improvements in detection performance metrics, including precision, recall, and mean Average Precision (mAP), compared to the standard YOLOv5 model. Detailed ablation studies and error analysis shed light on the specific contributions of DCNs and BiFPN to these performance gains. The enhanced model demonstrates significant potential for advancing automated pavement inspection systems, leading to more efficient and reliable road maintenance. Future research will focus on implementing real-time capabilities, expanding datasets, exploring additional cutting-edge techniques, and developing integrated systems for automated pavement inspection and maintenance.
Improved YOLOv5 for Pavement Patch Detection Using Deformable Convolution and Bidirectional Feature Pyramid Network
Wu, Xiaojing (author) / Chai, Xinzhuo (author) / Dai, Fei (author) / Wang, Shuai (author) / Dong, Yueyu (author) / Huang, Bi (author)
2024-11-05
4070852 byte
Conference paper
Electronic Resource
English
Feature Pyramid and Hierarchical Boosting Network for Pavement Crack Detection
| ArXiv | 2019
| Taylor & Francis Verlag | 2022
Pavement crack detection and small surface element fitting detection method based on YOLOv5
| European Patent Office | 2023