在精细化海洋观测领域,遥感卫星海浪方向谱的获取存在精度低、累积误差大等问题,设计了一种基于Transformer神经网络的海浪方向谱估计方法,通过引入稀疏自注意力机制以及物理引导模块等,实现海浪方向谱的高精度表达;海上试验测试结果表明,新提出的海浪方向谱估计方法与传统海浪方向谱估计方法相比,文章提出的模型估计方向谱更符合理论谱形,伪峰较少,平滑性和一致性方面表现较好,并且对数据噪声的容忍度更高.同时,设计了一种基于人工智能芯片的漂流波浪浮标系统,通过 AI 软硬件系统对不同任务进行动态调度分配,实现浮标低功耗运行,有效提升了波浪测量性能及海上工作时间,是遥感卫星高精度获取大范围、长时序全球性海洋观测数据的一种有效补充.
Research on Key Technologies of Drifting Wave Buoy for Ocean Wave Remote Sensing Satellite Calibration
In the field of refined ocean observation,the acquisition of direction spectrum of ocean waves by remote sensing satellites has some problems such as low accuracy and large cumulative errors.In this paper,a method for estimating the direction spectrum of ocean waves based on Transformer neural networks is designed,which introduces sparse self-attention mechanism and physical guidance module to achieve high-precision expression of ocean wave directional spectrum.The results of sea trial testing show that the newly proposed method for estimating the direction spectrum of ocean waves is more in line with the theoretical spectrum,with fewer pseudo peaks,better smoothness and consistency,and a higher tolerance for data noise compared to traditional methods.At the same time,a drift wave buoy system based on artificial intelligence chips is also designed,which dynamically schedules and allocates different tasks through AI software and hardware systems,achieving low-power operation of the buoy,effectively improving wave measurement performance and sea working time.It is an effec-tive supplement to high-precision acquisition of large-scale,long-term global ocean observation data by remote sensing satellites.
direction spectrumdrifting buoyocean observationremote sensing calibration
国家科技期刊平台
NSTL
万方数据
