基于软阈值函数的深度学习磁信号去噪算法

Deep Learning Magnetic Signal Denoising Algorithm Applying Soft Threshold Function

针对采用固定阈值函数的深度学习信号处理方法在处理实测磁信号时学习特征的能力差、去噪效果差的问题,提出基于自适应参数软阈值函数(APSTF)的去噪算法。该算法将软阈值函数作为激活函数引入神经网络框架,用于剔除特征图中的噪声特征,增强神经网络的学习能力,从而实现信号处理效果的提高;将U-Net网络中的ReLU激活函数替换为自适应参数软阈值函数,增强了实测噪声下神经网络的学习能力,显著的提高了去噪后磁信号的信噪比。区别于以往采用传统软阈值函数的去噪算法,利用ECA模块来自适应选取阈值,取消了对噪声分布已知的假设前提,能更好地适应实测噪声并将其去除。

We propose a denoising algorithm based on an adaptive parameter soft threshold function to solve the problem of the neural network with weak learning ability when process the measured magnetic signal.To remove noise in feature map,we use the adaptive parameter soft threshold function in neural network framework.In this paper,we use an adaptive parameter soft threshold function in U-Net to enhance the learning ability of the U-Net.The ECA module is used in adaptive parameter soft threshold function to select automatically threshold,compare with the traditional denoising algorithm based on th traditional soft threshold function,it is different.The algorithm is proposed by us can adapt better to the measued noise and remove them.