Mine wireless signal detection based on self-normalization neural network

Aiming at the issues of declining detection performance and poor robustness of traditional Convolutional Neural Networks (CNN) with complex mine channel environments, we propose a mine wireless signal detection network model based on Self-normalization Convolutional Neural Networks (SCNN). The design of the SCNN model builds on traditional CNN by applying the activation function Scaling Exponential Linear Units (SELU) from the Self-normalized Neural Network (SNN) into the convolutional layers of CNN. This utilization of SELU's self-normalization properties ensures network stability and improves signal detection performance. Additionally, the SCNN further regularizes training through dual implementation of AlphaDropout mechanism, thereby enhancing generalization ability. The simulation results demonstrate that compared to other signal detection algorithms, the proposed signal detection network exhibits superior signal detection performance and robustness, achieving at least one order of magnitude reduction in bit error rate at an 8dB Signal-to-Noise Ratio (SNR).