Scholarlink

<p>Precise and rapid detection of overheat-prone areas in cable oil-filled terminals using unmanned aerial vehicle (UAV) imagery is a prerequisite for intelligent defect diagnosis. However, detecting these areas remains challenging due to substantial variations in target sizes, low identification accuracy, and inadequate efficiency. Therefore, we develop the OTH-YOLO model based on the YOLOv11n model. The Ghost bottleneck is integrated into the feature extraction network to reduce complexity. Subsequently, a CSmix fusion attention mechanism is inserted between the backbone and the neck networks to enhance the accuracy. Finally, the model’s bounding box loss function is refined by integrating the complete intersection over union (CIoU)-normalized Wasserstein distance (NWD) loss, strengthening the model’s ability to identify small targets. Compared to the original YOLOv11 model, the OTH-YOLO achieved a 1.3% improvement in the mean average precision calculated when the intersection over union (IoU) threshold was 0.5(mAP@0.5), a 1.8% improvement in recall, a 0.1% improvement in precision, a 0.2 decrease in giga floating-point operations (GFLOPs), and a 4-frames per second (FPS) increase, maintaining the original model size, indicating higher accuracy and efficiency compared with conventional YOLO algorithms. Consequently, the model delivers superior performance in detecting overheat-prone areas of cable oil-filled terminals in UAV-acquired infrared imagery. The source code can be found at <a href="https://gitee.com/yao-shunyu1/real-time-detection-and-analysis-of-overheat-prone-areas-in-cable-oil-filled-terminals-from-uav-infrared-images.git">https://gitee.com/yao-shunyu1/real-time-detection-and-analysis-of-overheat-prone-areas-in-cable-oil-filled-terminals-from-uav-infrared-images.git</a>.</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

Real-Time Detection and Analysis of Overheat-Prone Areas in Cable Oil-Filled Terminals from UAV Infrared Images

  • 期刊论文

  • 发表日期：

    2026年6月2日

查看全文
  • Shunyu Yao

摘要

Precise and rapid detection of overheat-prone areas in cable oil-filled terminals using unmanned aerial vehicle (UAV) imagery is a prerequisite for intelligent defect diagnosis. However, detecting these areas remains challenging due to substantial variations in target sizes, low identification accuracy, and inadequate efficiency. Therefore, we develop the OTH-YOLO model based on the YOLOv11n model. The Ghost bottleneck is integrated into the feature extraction network to reduce complexity. Subsequently, a CSmix fusion attention mechanism is inserted between the backbone and the neck networks to enhance the accuracy. Finally, the model’s bounding box loss function is refined by integrating the complete intersection over union (CIoU)-normalized Wasserstein distance (NWD) loss, strengthening the model’s ability to identify small targets. Compared to the original YOLOv11 model, the OTH-YOLO achieved a 1.3% improvement in the mean average precision calculated when the intersection over union (IoU) threshold was 0.5(mAP@0.5), a 1.8% improvement in recall, a 0.1% improvement in precision, a 0.2 decrease in giga floating-point operations (GFLOPs), and a 4-frames per second (FPS) increase, maintaining the original model size, indicating higher accuracy and efficiency compared with conventional YOLO algorithms. Consequently, the model delivers superior performance in detecting overheat-prone areas of cable oil-filled terminals in UAV-acquired infrared imagery. The source code can be found at https://gitee.com/yao-shunyu1/real-time-detection-and-analysis-of-overheat-prone-areas-in-cable-oil-filled-terminals-from-uav-infrared-images.git.