Scholarlink

<p>Roads are an essentiasl part of both urban and rural infrastructure and the backbone of transportation. These roads experience varied degrees of damage because of variations in vehicle loads and construction lifespans. On the other hand, problems with road damage become more noticeable as roads get bigger. Hence, the proposed study introduces an innovative DL model for classifying multiple road damages effectively. Initially, the images collected from the raw database are preprocessed by performing denoising using the Adaptive Lee filtering (Adapt-LF) technique. Then, an innovative Hybridized CSwin Transformer-enabled You Only Look Once Version 8 (HyCSwinT-YOLOv8) technique is introduced to detect and classify various road damages for different countries. In addition to this, the parameters present in the proposed model are optimized using the Tent Chaotic Al-Biruni Earth optimization (TCh-AEO) technique that effectively minimizes network complexities during the training process. Finally, various road damages like potholes, longitudinal cracks (LC), transverse cracks (TC), and alligator cracks (AC) are classified using the proposed model. Performance indicators like Accuracy, Kappa coefficient (KC), false discovery rate (FDR), Intersection of union (IoU), mean average precision (mAP), G-mean, and computation time (CT) are scrutinized and associated with other studies. The overall accuracy of 99.55%, KC of 98.76%, mAP of 95.6%, IoU of 98.86%, CT of 92.96%, and FDR of 0.765 are obtained by the developed framework, outperforming existing methods by 2–5%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A robust optimized Swin Transformer-enabled You Only Look Once version 8 model for road damage classification in various countries

  • 期刊论文

  • 发表日期：

    2025年8月30日

查看全文
  • G. Gomathi,
  • G. Niranjana

摘要

Roads are an essentiasl part of both urban and rural infrastructure and the backbone of transportation. These roads experience varied degrees of damage because of variations in vehicle loads and construction lifespans. On the other hand, problems with road damage become more noticeable as roads get bigger. Hence, the proposed study introduces an innovative DL model for classifying multiple road damages effectively. Initially, the images collected from the raw database are preprocessed by performing denoising using the Adaptive Lee filtering (Adapt-LF) technique. Then, an innovative Hybridized CSwin Transformer-enabled You Only Look Once Version 8 (HyCSwinT-YOLOv8) technique is introduced to detect and classify various road damages for different countries. In addition to this, the parameters present in the proposed model are optimized using the Tent Chaotic Al-Biruni Earth optimization (TCh-AEO) technique that effectively minimizes network complexities during the training process. Finally, various road damages like potholes, longitudinal cracks (LC), transverse cracks (TC), and alligator cracks (AC) are classified using the proposed model. Performance indicators like Accuracy, Kappa coefficient (KC), false discovery rate (FDR), Intersection of union (IoU), mean average precision (mAP), G-mean, and computation time (CT) are scrutinized and associated with other studies. The overall accuracy of 99.55%, KC of 98.76%, mAP of 95.6%, IoU of 98.86%, CT of 92.96%, and FDR of 0.765 are obtained by the developed framework, outperforming existing methods by 2–5%.