Rufu Qin

Accurate sea wave parameter estimation is essential for marine science and coastal engineering, yet deep learning approaches in this domain remain constrained by the scarcity of large-scale, accurately labeled visual datasets. This paper addresses this bottleneck by employing stereo vision 3D reconstruction to generate labeled datasets from public stereo imagery, deriving instantaneous extreme elevation (Emax), signifcant wave height (Hs), and peak period (Tp) through multi-domain analysis. Building upon this data foundation, three novel deep learning architectures are designed and comparatively evaluated across distinct visual data domains: WPEN-S, a spatial domain CNN model for estimating max from single frames (MRE = 0.24%); WPEN-ST, a spatiotemporal model integrating CNN–LSTM with spatial and temporal attention for predicting Hs and Tp from image sequences (Hs MRE = 11.35%, Tp MRE = 8.89%); and WPEN-TF, a lightweight time-frequency model using dual-branch LSTM and 1D CNN to estimateHs and Tp from pixel-level intensity series (Hs MRE = 20.71%), demonstrating the feasibility of parameter prediction without full image processing. These results validate multi-domain deep learning as a practical and effective strategy for vision-based wave measurement.

Remote sensing data are of considerable significance for monitoring global climate, detecting harmful algae bloom, etc. However, due to sensor failures, cloud cover, and thick aerosols, the collected remote sensing data for various ocean factors, such as Chlorophyll-a (Chl-a) concentration and Sea Surface Temperature (SST), often have a high missing rate, which seriously hinders their applications. Existing data imputation models mostly ignore highly-sparse spatial locations or perform bad when the data missing rate is high due to the lack of available information. Large Language Models (LLMs) possess powerful representation learning capabilities and can effectively capture sequential correlations even with extremely limited information, thereby presenting promising potential for highly-sparse remote sensing data imputation. Therefore, we proposed a novel LLM-based spatio-temporal model for highly-sparse remote sensing data imputation, i.e., LLM4HRS. First, LLM4HRS develops a LLM-based bidirectional temporal representation learning module to learn forward and backward temporal dependencies in data sequences and fuses them together to obtain comprehensive temporal representations. Next, LLM4HRS constructs a LLM-based spatial representation learning module to learn spatial correlations with the learned temporal representation. Finally, a spatio-temporal representation fusion and data imputation module is developed to achieve data imputation. Experiments on the SST and Chl-a remote sensing datasets demonstrate that LLM4HRS significantly outperforms existing data imputation models, with its advantages becoming more pronounced as the masking rate increases. Furthermore, when extended to the remote sensing PAR data in a large region, LLM4HRS still achieves the best performance, further validating its broad applicability for remote sensing data imputation. The code of LLM4HRS is publicly available at https://github.com/ssyuwang/LLM4HRS-master.

Marine sensor networks, essential for obtaining marine environmental data, face new challenges as marine environment monitoring extends into the deep sea and the demand for long-term in-situ time series analysis increases. Traditional centralized computing mode is inadequate for deep-sea monitoring and intelligent marine observation due to its limitations in communication and real-time processing. Moreover, long-term time series analysis models are typically complex and unable to provide in-situ real-time services. To overcome these challenges, this paper integrates edge computing into marine sensor networks and combines it with a lightweight time series forecasting model based on the Transformer structure, MarineEdgeformer. This presented model utilizes a linear attention mechanism and an optimized decoder structure, significantly reducing the model’s computational costs and improving inference speed while maintaining forecasting accuracy. Experimental evaluations using a real-world dataset with a Raspberry Pi as the edge computing node demonstrate that MarineEdgeformer effectively reduces computational costs and achieves reliable long-term forecasting accuracy. These results not only confirm the potential of edge computing in enhancing the real-time processing capabilities, reliability, and intelligence of marine sensor networks but also provide a technical solution for the development of future marine environment monitoring.

Seafloor observatories enable long-term, continuous, real-time, weather-independent, and multidisciplinary scientific observation and research that will promise major breakthroughs in ocean sciences. China has started to establish a seafloor observation system in the East China Sea (ECSSOS), and the purpose of this paper is to describe the remote control for ECSSOS. Technological development in system architecture and remote communication was first performed in this paper. Based on this work, an overall framework for the remote control solution for ECSSOS was developed, adopting a client-server architecture containing three components, that is, a control center, a shore station, and observatory facilities, as well as two strands of information flow, that is, flow of observation data and flow of control commands. To implement the present solution, operational data flow and the core module were further designed and the corresponding remote control subsystem for ECSSOS was developed against all desig... AbstractSeafloor observatories enable long-term, continuous, real-time, weather-independent, and multidisciplinary scientific observation and research that will promise major breakthroughs in ocean sciences. China has started to establish a seafloor observation system in the East China Sea (ECSSOS), and the purpose of this paper is to describe the remote control for ECSSOS. Technological development in system architecture and remote communication was first performed in this paper. Based on this work, an overall framework for the remote control solution for ECSSOS was developed, adopting a client-server architecture containing three components, that is, a control center, a shore station, and observatory facilities, as well as two strands of information flow, that is, flow of observation data and flow of control commands. To implement the present solution, operational data flow and the core module were further designed and the corresponding remote control subsystem for ECSSOS was developed against all desig...