基于深度学习自编码器的X波段雷达二维波浪场实时重构方法研究

Research on Real-time Reconstruction Method of Two-dimensional Wave Field of X-band Radar Based on Deep Learning Autoencoder

X波段波浪雷达是一种具有广阔应用前景的海洋环境检测技术。目前,利用X波段波浪雷达提取波浪参数的方法已较为成熟,但尚无法重构和反演实时波浪场。该文提出了一种基于卷积自编码器的深度学习方法,实现了对X波段波浪雷达图像的反演,可以得到二维实时波浪场。该方法对5.5 m≤HS≤8.5 m,3.5√H_(S)≤T_(P)≤4.8√H_(S)(H_(S)为有义波高,T_(P)为谱峰周期)的JONSWAP谱波浪场进行反演的误差在7%以内,对含有噪声干扰的雷达图像和部分区域遮挡的雷达图像具有良好的恢复能力,展现出该方法对X波段波浪雷达图像反演二维波浪场的重要意义。

X-band wave radar is a marine environment detection technology with broad application prospects.At present,the extraction of wave parameters using X-band wave radar is relatively mature,but it is not able to reconstruct and invert the wave field.In this paper,a deep learning method based on convolutional autoencoder is proposed to invert X-band wave radar images and obtain two-dimensional wave fields.The error of this method in inverting of JONSWAP spectral wave field with S 5.5 m≤H_(S)≤8.5 m,3.5√H_(S)≤T_(P)≤4.8√H_(S)(H_(S)is significant wave height and T_(P)is spectral peak period)using this method is within 7%,and this method has good recovery ability for radar images containing noise interference or partial occlusion.This method demonstrates the important significance of inverting two-dimensional wave fields from X-band wave radar images.