Intelligent recognition of coal and gas outburst precursor and potential hazards using YOLOv8 neural network

Abstract

Fluctuations in gas emission or concentration at working face represent primary indicators of impending coal and gas outbursts, making them essential for monitoring processes. However, the direct use of original signals for predictive purposes may cause false warnings due to the inability to recognize valuable potential information. This can result in decreased prediction accuracy. Additionally, gas emission or concentration is affected by potential hazards such as sensor calibration, borehole spraying, and air duct damage, leading to complexity of their behavior and recognition process. This paper proposes an intelligent recognition method leveraging YOLOv8 neural network to discern coal and gas outburst precursors and potential hazards. By employing continuous wavelet transform (CWT) on the gas concentration signal processed by critical slowing down (CSD) method, a two-dimensional time–frequency representation is generated. This representation is then fed into YOLOv8 model to recognize the outburst precursor characteristics and potential hazards. The research results show that YOLOv8-based intelligent recognition adeptly identifies the relevant precursor characteristics and potential hazards, enhancing both coal mine safety protocols and the accuracy of early outburst warning mechanisms.