Scholarlink

<p>Ensuring the safe operation of substations, vital nodes in power systems, is critical for grid stability and preventing economic losses and injuries. This study introduces YOLO-IASE, an advanced deep-learning model designed for robust safety inspection in complex substation environments, focusing on flame and safety helmet detection. The model integrates an Intelligent Adaptive Scene Enhancement (IASE) technique, which dynamically enhances real-time training data to improve adaptability and generalization beyond traditional augmentation methods. A novel CCBS module incorporating Coordinate Convolution layers enhances spatial awareness and target localization accuracy during feature extraction. The proposed <i>δ</i>-EIoU loss function also reduces sensitivity to extreme aspect ratios, boosting detection performance for irregularly shaped objects like flames. Experimental results demonstrate that YOLO-IASE outperforms baseline models such as YOLOv5s, achieving a recall rate of 0.868 and a 12.1% improvement in flame detection recall. With superior accuracy, robustness, and real-time processing capabilities, YOLO-IASE sets a new standard for multi-target detection in challenging substation settings, offering a reliable solution for enhancing safety monitoring in critical infrastructure.</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

Enhanced YOLO-IASE for robust safety inspection in complex substation environments

  • 期刊论文

  • 发表日期：

    2025年6月5日

查看全文
  • Jiang Junjie,
  • Zhang Yongqi,
  • Xiaowei Zhou,
  • Khalil AL-Bukhaiti,
  • Wan Anping,
  • Xiaomin Cheng,
  • Xiaosheng Ji

摘要

Ensuring the safe operation of substations, vital nodes in power systems, is critical for grid stability and preventing economic losses and injuries. This study introduces YOLO-IASE, an advanced deep-learning model designed for robust safety inspection in complex substation environments, focusing on flame and safety helmet detection. The model integrates an Intelligent Adaptive Scene Enhancement (IASE) technique, which dynamically enhances real-time training data to improve adaptability and generalization beyond traditional augmentation methods. A novel CCBS module incorporating Coordinate Convolution layers enhances spatial awareness and target localization accuracy during feature extraction. The proposed δ-EIoU loss function also reduces sensitivity to extreme aspect ratios, boosting detection performance for irregularly shaped objects like flames. Experimental results demonstrate that YOLO-IASE outperforms baseline models such as YOLOv5s, achieving a recall rate of 0.868 and a 12.1% improvement in flame detection recall. With superior accuracy, robustness, and real-time processing capabilities, YOLO-IASE sets a new standard for multi-target detection in challenging substation settings, offering a reliable solution for enhancing safety monitoring in critical infrastructure.