摘要
高分辨率的海洋流速场对提高数值模拟和预测精度具有重要意义。以海洋环流模型HYCOM的1/12°分辨率模拟海洋流速场为基础,重建超分辨率海洋流速场。但是,直接以HYCOM结果重建高分辨率海洋流速场存在困难,同时传统插值难以准确恢复杂海洋流速场的细节。为解决这一难题,提出一种基于SobelUnet的超分辨率重建方法。该模型包含初始态模块、多语义融合模块和残差Sobel注意力模块3部分。其中,初始态模块通过分析多年海洋流速场数据成功捕捉了关键海陆细节,多语义融合模块合成不同语义级别的特征来弥补信息损失,残差Sobel注意力模块利用Sobel卷积和注意力机制增强了海洋流速场边缘。实验结果表明,相比其他参考模型,该方法在PSNR、SSIM、MSE和MAE等多个指标上有所提高。SobelUnet模型能够高效地重建海洋流速场数据,并恢复更丰富的细节信息。
High-resolution ocean current velocity fields are significant for improving the accuracy of numerical simulation and prediction.In this study,the ocean current velocity fields simulated by the Hybrid Coordinate Ocean Model(HYCOM)with a resolution of 1/12°are utilized to construct a super-resolution field.However,directly reconstructing high-resolution ocean currents from HYCOM results is challenging,and conventional interpolation methods fail to accurately recover complex flow details.To address this problem,a super-resolution reconstruction method based on SobelUnet is proposed.This model consists of three modules:the initial state module,the multi-semantic fusion module,and the residual Sobel attention module.Specifically,the initial state module successfully captures critical land-sea details by analyzing multi-year ocean current data;the multi-semantic fusion module synthesizes features at different semantic levels to compensate for information loss;the residual Sobel attention module enhances ocean current edges by employing Sobel convolutions and attention mechanisms.Experimental results demonstrate that compared with other reference models,the proposed method improves PSNR,SSIM,MSE and MAE.It illustrates that the SobelUnet model can efficiently reconstruct ocean current data and recover richer detail information.
作者
杜俊坤
王明清
黄占鳌
李孝杰
DU Junkun;WANG Mingqing;HUANG Zhan'ao;LI Xiaojie(School of Computer Science,Chengdu University of Information Technology,Chengdu 610225,China;Ninecosmos Science and Technology Ltd.,Wuxi 214072,China)
出处
《软件导刊》
2025年第7期185-193,共9页
Software Guide
基金
成都信息工程大学科技创新能力提升计划项目(KYTD202330)。
