Detection of coal fire by deep learning using ground penetrating radar. (30th September 2022)

• Record Type: Journal Article
• Authors: Gao, Rongxiang; Zhu, Hongqing; Liao, Qi; Qu, Baolin; Hu, Lintao; Wang, Haoran
• Is Part Of: Measurement. Volume 201(2022)
• Publication Date: 2022-09-30
• Subjects: Coal fire -- Physical model -- Ground-penetrating radar -- Deep learning -- Object detection; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.111585
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A deep learning-based method of detecting coal fire using GPR is proposed. GPR images of coal fire are obtained based on the physical model of coal fire. The detection target of GPR and the corresponding signal in GPR images are defined. YOLOv5 is the best algorithm in terms of speed and performance of detection. Abstract: Coal fire seriously endangers coal resources. Accurate detection of its combustion range is the basis of disaster control. In this paper, a deep learning-based method of recognizing coal fire using ground-penetrating radar (GPR) is proposed, which improves the accuracy and speed of delineating coal fire areas. The self-built coal fire physical model is scanned by the GPR, and the radar images are obtained. The test results are compared with GPR images to summarize the spatial evolution law of coal fire areas and interpret the signal characteristics of the coal fire in radar images. The signal characteristics include combustion cavity, combustion surface, and underground combustion collapse surface. Comparing different algorithms, the results show that the YOLOv5l has the highest detection accuracy, which meets the need for detection of the coal fire. The proposed method lays the foundation for the detection of the combustion range in the coal fire areas.