Scholarlink

<p>Visual inspection of industrial surface defect is critical for assessing structural integrity, yet the minute defects on industrial structure surface pose a significant challenge for visual inspection methods. The typical difficulties in inspecting the minute defects for industrial structure are the differential size defects with low background contrast, the training data with unbalanced distribution, and the detection systems with high real-time performance. We propose a novel detection approach, which integrates the attention-guided detection strategies, the lightweight design strategies, and the separate adaptive data augmentation strategy. This approach effectively detected minute defects for complex industrial structure surface, while simultaneously reducing the model’s computational complexity and inference latency. Experimental results demonstrate that the proposed approach outperforms mainstream detection algorithms in key metrics. This integrated approach addresses the critical bottleneck of detection accuracy and inference speed for minute defects, providing a universal and effective solution for industrial surface defect inspection.</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 novel attention-guided inspection approach for minute industrial surface defect based on separate adaptive data augmentation

  • 期刊论文

  • 发表日期：

    2026年7月22日

查看全文
  • Jinghui Wang,
  • Lin Zhang,
  • Zhen Wang

摘要

Visual inspection of industrial surface defect is critical for assessing structural integrity, yet the minute defects on industrial structure surface pose a significant challenge for visual inspection methods. The typical difficulties in inspecting the minute defects for industrial structure are the differential size defects with low background contrast, the training data with unbalanced distribution, and the detection systems with high real-time performance. We propose a novel detection approach, which integrates the attention-guided detection strategies, the lightweight design strategies, and the separate adaptive data augmentation strategy. This approach effectively detected minute defects for complex industrial structure surface, while simultaneously reducing the model’s computational complexity and inference latency. Experimental results demonstrate that the proposed approach outperforms mainstream detection algorithms in key metrics. This integrated approach addresses the critical bottleneck of detection accuracy and inference speed for minute defects, providing a universal and effective solution for industrial surface defect inspection.