Scholarlink

<p>The primary innovation of this work is a task-driven, system-level architecture tailored for UAV-based transmission line inspection in complex environments. Rather than merely adopting existing concepts, the proposed framework introduces a coordinated and synergistic redesign of attention mechanisms, multi-scale feature fusion, and loss formulation–specifically targeting the challenges inherent in aerial imagery, such as small target size, frequent occlusion, and complex backgrounds . The resulting AIR-YOLO framework features: (1) an Adaptive Interactive Feature Integration (AIFI) module, which supersedes the standard SPPF to enable enhanced cross-scale semantic interaction via multi-head self-attention; (2) a lightweight ASF-Small neck that incorporates a high-resolution P2 feature layer with sequential scale fusion, preserving fine-grained details of small defects; (3) a hybrid Spatial-and-Channel SEAM attention head embedded in the detection head to bolster robustness under partial occlusion through adaptive feature recalibration; and (4) a PowerIoU (PIoU) loss, specially designed for elongated structures to improve localization precision. This holistic design achieves an exceptional balance between detection accuracy, computational efficiency, and model complexity, rendering it highly suitable for real-time deployment on edge devices in power line inspection scenarios.</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

AIR-YOLO: Enhanced small-object detection for transmission line inspection in complex backgrounds

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Lihui Lu,
  • Xinyuan Wu,
  • Sihao Zhang,
  • Jianguo Liu,
  • Junlu Jiang,
  • Yuke Gu

摘要

The primary innovation of this work is a task-driven, system-level architecture tailored for UAV-based transmission line inspection in complex environments. Rather than merely adopting existing concepts, the proposed framework introduces a coordinated and synergistic redesign of attention mechanisms, multi-scale feature fusion, and loss formulation–specifically targeting the challenges inherent in aerial imagery, such as small target size, frequent occlusion, and complex backgrounds . The resulting AIR-YOLO framework features: (1) an Adaptive Interactive Feature Integration (AIFI) module, which supersedes the standard SPPF to enable enhanced cross-scale semantic interaction via multi-head self-attention; (2) a lightweight ASF-Small neck that incorporates a high-resolution P2 feature layer with sequential scale fusion, preserving fine-grained details of small defects; (3) a hybrid Spatial-and-Channel SEAM attention head embedded in the detection head to bolster robustness under partial occlusion through adaptive feature recalibration; and (4) a PowerIoU (PIoU) loss, specially designed for elongated structures to improve localization precision. This holistic design achieves an exceptional balance between detection accuracy, computational efficiency, and model complexity, rendering it highly suitable for real-time deployment on edge devices in power line inspection scenarios.