Scholarlink

<p>The monitoring equipment installed atop tower cranes plays a crucial role in detecting potential safety hazards around construction workers. However, due to the small size of the targets involved in detection tasks and the occlusions caused by extreme weather conditions such as heavy rain and snow, traditional object detection methods struggle to capture global contextual information, thereby affecting overall detection performance.To address these issues, we propose a novel method named SHM-YOLO for detecting occluded small objects from top-down views of tower cranes. This method is based on the You Only Look Once Version 11n (YOLOv11n). Our approach begins with the construction of a backbone network that incorporates a Haar Wavelet Downsampling (HWD) module, endowing the network with HWD mapping capabilities and lightweight characteristics. Secondly, we design a Neck network that integrates a Separated and Enhancement Attention Module (SEAM), which reduces the impact of occlusions and enables the network to focus more effectively on image features. Finally, we employ Inner-MPDIoU to enhance the network's overall performance. Compared to the YOLOv8n, SHM-YOLO demonstrates a 4.80 percentage point improvement in precision on our custom-built Overlooking Picture (OP) datasets and a 6.03 percentage point improvement on the custom-built Rain-Snow Overlooking Picture (RSOP) datasets. Additionally, the overall parameter count is reduced by 19%, the GFLOPs is decreased by 23%, and the speed satisfies real-time detection requirements. Compared to original YOLOv11n, the detection precision improved by 2.52 percentage points on the OP dataset and by 2.22 percentage points on the RSOP dataset. The results indicate that SHM-YOLO excels at detecting occluded objects under extreme weather conditions, thereby fulfilling the practical requirements of real-world applications.</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

SHM-YOLO: Detection of occluded small objects in top-down views from tower cranes under adverse rain and snow conditions

  • 期刊论文

  • 发表日期：

    2025年10月21日

查看全文
  • Yudong Pang,
  • Zhixing Li,
  • Cong Du,
  • Yanxue Wang,
  • Zhenkun Guo

摘要

The monitoring equipment installed atop tower cranes plays a crucial role in detecting potential safety hazards around construction workers. However, due to the small size of the targets involved in detection tasks and the occlusions caused by extreme weather conditions such as heavy rain and snow, traditional object detection methods struggle to capture global contextual information, thereby affecting overall detection performance.To address these issues, we propose a novel method named SHM-YOLO for detecting occluded small objects from top-down views of tower cranes. This method is based on the You Only Look Once Version 11n (YOLOv11n). Our approach begins with the construction of a backbone network that incorporates a Haar Wavelet Downsampling (HWD) module, endowing the network with HWD mapping capabilities and lightweight characteristics. Secondly, we design a Neck network that integrates a Separated and Enhancement Attention Module (SEAM), which reduces the impact of occlusions and enables the network to focus more effectively on image features. Finally, we employ Inner-MPDIoU to enhance the network's overall performance. Compared to the YOLOv8n, SHM-YOLO demonstrates a 4.80 percentage point improvement in precision on our custom-built Overlooking Picture (OP) datasets and a 6.03 percentage point improvement on the custom-built Rain-Snow Overlooking Picture (RSOP) datasets. Additionally, the overall parameter count is reduced by 19%, the GFLOPs is decreased by 23%, and the speed satisfies real-time detection requirements. Compared to original YOLOv11n, the detection precision improved by 2.52 percentage points on the OP dataset and by 2.22 percentage points on the RSOP dataset. The results indicate that SHM-YOLO excels at detecting occluded objects under extreme weather conditions, thereby fulfilling the practical requirements of real-world applications.