Acoustic emission and electromagnetic radiation of coal-rock effective and interference signal identification utilizing generative adversarial learning and image feature mining

Acoustic Emission (AE) and Electromagnetic Radiation (EMR) are playing an increasingly important role in the field of coal and rock dynamic disaster early warning due to their accurate response to the evolution process. However, blasting, drilling, and other coal mine technical activities are easily to produce interference signals, which seriously affect the credibility of early warning information. Moreover, unbalanced samples and complex characteristic characterization cannot achieve accurate identification. This paper presents a novel identification method for effective and interference signal of AE and EMR based on generative adversarial learning and image feature mining. First, Kalman filter is applied to AE and EMR monitoring signals to remove noise and retain key features. The Wasserstein Generative Adversarial Network, then, resolves the imbalance between the sample numbers of effective and various types of interference signals to ensure generalization of the identification. The effective and interference signal samples are further converted graphically by Symmetrized Dot Pattern, and intuitive different distribution characteristics are obtained. Finally, the EfficientNet model accurately identified typical effective and six interference signals collected downhole. The practical case of a coal mine in Liaoning Province shows that the proposed method is feasible and effective, and can provide a basis for reliable early warning of coal and rock dynamic disasters.