Adaptive signal recognition in mines based on deep learning.

To address the challenges of low recognition accuracy and high system complexity arising from the coexistence of multiple wireless communication technologies and severe signal interference in the complex wireless environment of coal mines, this paper proposes a deep learning-based adaptive signal recognition method. By incorporating grouped residual convolution and channel shuffling techniques, the proposed method significantly reduces the number of model parameters (37% fewer than the original WaveNet) while utilizing dilated causal convolution to capture long-range dependencies in the signal, thereby enhancing the model's ability to discriminate multipath interference features. The introduction of a dynamic channel attention mechanism facilitates adaptive adjustment of feature weights, emphasizing key features while suppressing noise interference, thereby improving recognition accuracy. Experimental results demonstrate that the Group Residual Shuffle Attention WaveNet achieves average recognition rates of 93.2% and 94.5% on the public dataset (RML2016.10a) and a simulated dataset, respectively, outperforming other methods (such as CTDNN) by more than 1.5% in recognition accuracy, while improving inference speed by over 14%. The proposed method performs well on general datasets and effectively adapts to complex signal recognition tasks in mine environments, providing an efficient and reliable solution for intelligent mine communication.