Scholarlink

<p>The cost of anticorrosion maintenance is considered the most substantial expense associated with steel construction. Preventive maintenance of steel structures primarily depends on visual inspections, but the results are often inconsistent due to the inspector’s experience. To overcome this problem, this study proposed a three-pronged approach comprising distinct methodologies for automated image analysis. The proposed RustNet architecture uses a feature encoder, feature decoder, and Rust module for image analysis. The encoder applies convolution and down-sampling to generate feature maps, while the decoder, with three stacked blocks, reconstructs the image. The Rust module employs a multilayer attention network (MAN) to extract rust-specific features. Furthermore, gradient edge detection and linked segment analysis are employed to precisely quantify individual rust areas. The proposed model is validated on a large dataset of images captured under different light conditions; the validation&#xa0;results demonstrates promising performance in rust growth detection and quantification.</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

RustNet: an automated approach for rust detection and quantification in steel structures using feature encoding and multilayer attention networks

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • Touqeer Abbas,
  • Muhammad Mubeen,
  • Abdul Razzaq,
  • Umair Javed,
  • Muneeb Raza,
  • Mohammad Tabish,
  • Khalid M. Alotaibi,
  • Ghulam Yasin,
  • Hui Li

摘要

The cost of anticorrosion maintenance is considered the most substantial expense associated with steel construction. Preventive maintenance of steel structures primarily depends on visual inspections, but the results are often inconsistent due to the inspector’s experience. To overcome this problem, this study proposed a three-pronged approach comprising distinct methodologies for automated image analysis. The proposed RustNet architecture uses a feature encoder, feature decoder, and Rust module for image analysis. The encoder applies convolution and down-sampling to generate feature maps, while the decoder, with three stacked blocks, reconstructs the image. The Rust module employs a multilayer attention network (MAN) to extract rust-specific features. Furthermore, gradient edge detection and linked segment analysis are employed to precisely quantify individual rust areas. The proposed model is validated on a large dataset of images captured under different light conditions; the validation results demonstrates promising performance in rust growth detection and quantification.