YOLO-WRC-based UAV detection method for spontaneous combustion in open-pit coal seams
WANG Biao
TAO Qing
LIU Wei
LI Fanhuan
Abstract：UAVs have significant advantages over traditional measurement and remote-sensing technologies in monitoring open-pit mining areas.At present,existing UAV-based detection methods for spontaneous combustion in open-pit coal seams mainly suffer from the lack of corresponding detection models capable of identifying high-temperature points,low recognition accuracy for small-size and multi-scale high-temperature points,and confusion between exhaust-pipe high temperatures of excavators and spontaneous combustion high-temperature points on coal seams.To address these problems,a YOLO-WRC-based detection method for spontaneous combustion in open-pit coal seams using UAV imagery was proposed.Wavelet Transform Convolution(WTConv)was integrated into the backbone network to focus on richer feature information;a Reparameterized Generalized Feature Pyramid Network(RepGFPN)was used to reconstruct the neck network,enhance the ability of feature extraction and fusion and the recognition accuracy of easily confused high temperature points;a Concentrated Layerwise Localization Attention Head(CLLAHead)was introduced to coordinate feature and semantic information across different levels,focusing on the identification of micro high-temperature points;and the PIoUv2 loss function was adopted to improve the model's regression performance for multi-scale abnormal high-temperature points.The experimental results showed that ① the accuracy,recall,and mAP@0.5 of YOLO-WRC reached 88.2％,90.1％,and 95.4％,respectively,which were 1.3％,2.2％,and 3.2％higher than those of the original YOLOv8n model.② The recall and mAP@0.5 of YOLO-WRC were superior to mainstream models such as SSD,Faster-RCNN,YOLOv5,and YOLOv10n,demonstrating high robustness and adaptability in identifying abnormal high-temperature points.③ YOLO-WRC yielded higher confidence for detection targets and identified targets missed by YOLOv8n,exhibiting stronger recognition capability for easily confused and small-sized targets.
Keywords：spontaneous combustion of open-pit coal seamsUAV inspectionYOLOv8nYOLO-WRCWavelet Transform ConvolutionReparameterized Generalized Feature Pyramid NetworkConcentrated Layerwise Localization Attention HeadPIoUv2 loss function
Publication Date：2025-11-30
Online Publishing Date：2025-12-25(First online date of this platform, not the publication date of the document)
Pages：10( 87-95,111 )