基于深度学习的海洋色散信道信号处理研究

Signal Processing for Marine Dispersive Channels Based on Deep Learning

海洋非均匀介质构成的海底信道中,电磁波色散现象会产生明显的频域谐波特征。然而,传统谱线检测方法通常只关注单一谱线的识别而未能对谐波特征进行全局分析,且谱线能量的分散导致传统检测方法的性能显著下降。为了克服这些挑战,提出了一种基于先验霍夫增强的卷积神经网络检测方法,利用霍夫变换对谱线进行预定位,并借助深度学习的全局视野与强大特征学习能力,实现了对色散现象特征的准确检测,显著提升了检测性能。

In marine channels formed by non-uniform media,electromagnetic wave dispersion results in distinct harmonic characteristics in the frequency domain.However,traditional spectral line detection methods typically focus on identifying individual spectral lines without a global analysis of harmonic characteristics,and the dispersion of spectral line energy leads to significant performance degradation of these traditional methods.To overcome these challenges,a convolutional neural network(CNN)detection method based on prior Hough enhancement is proposed.This method utilizes the Hough transform for spectral line pre-localization,and leverages the global perspective and powerful feature learning capabilities of deep learning to achieve accurate detection of dispersion characteristics,significantly improving detection performance.