Scholarlink

The accumulation of Saharan sand on roadways presents significant transportation hazards, emphasizing the need for reliable detection techniques to reduce risks. This research explores and compares the performance of various deep learning models, such as Convolutional Neural Networks (CNN), U-net, and ResNet50, for identifying sand deposits on road surfaces using satellite images. We conducted a thorough training and evaluation of each model using a labelled dataset, measuring their effectiveness through key metrics like accuracy, precision, recall, and computational efficiency. CNNs are known for their ability to capture spatial patterns, while U-net’s architecture is optimized for pixel-level segmentation, making it well-suited for detecting sand with high precision. ResNet50, leveraging its deep residual learning approach, addresses the vanishing gradient problem in deep networks, potentially boosting detection performance. Our comparative analysis offers valuable insights into the advantages and draw-backs of each model, providing guidance for the design of automated systems aimed at improving road safety in desert environments. This research promotes the application of deep learning techniques in the field of environmental monitoring particularly in enhancing the real-time detection of natural hazards affecting infrastructure.

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

Deep Learning Approaches for Detecting Saharan Sand on Roads: A Comparative Study of CNN, U-net, and ResNet50

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Rabiaa Cheikh Maoulainine,
  • Younes Chihab,
  • Redouan Korchiyne,
  • Loubna Salaheddine,
  • Rachid Dahmani,
  • Soufiane Hajbi,
  • Zineb Squalli Houssaini

摘要

The accumulation of Saharan sand on roadways presents significant transportation hazards, emphasizing the need for reliable detection techniques to reduce risks. This research explores and compares the performance of various deep learning models, such as Convolutional Neural Networks (CNN), U-net, and ResNet50, for identifying sand deposits on road surfaces using satellite images. We conducted a thorough training and evaluation of each model using a labelled dataset, measuring their effectiveness through key metrics like accuracy, precision, recall, and computational efficiency. CNNs are known for their ability to capture spatial patterns, while U-net’s architecture is optimized for pixel-level segmentation, making it well-suited for detecting sand with high precision. ResNet50, leveraging its deep residual learning approach, addresses the vanishing gradient problem in deep networks, potentially boosting detection performance. Our comparative analysis offers valuable insights into the advantages and draw-backs of each model, providing guidance for the design of automated systems aimed at improving road safety in desert environments. This research promotes the application of deep learning techniques in the field of environmental monitoring particularly in enhancing the real-time detection of natural hazards affecting infrastructure.