Research on Robust Terrain Tracking Data Processing Technology Between Frames for Underwater Multibeam Bathymetry Systems

This paper proposes an optimized filtering strategy for attitude data and terrain tracking data related to beam acquisition in multibeam bathymetry systems, addressing key challenges in marine geospatial data processing within complex underwater environments to enhance the accuracy and robustness of underwater terrain mapping. The framework first employs Kalman filtering to fuse GNSS and Heave data, correcting vessel attitude errors and significantly reducing measurement inaccuracies caused by vessel dynamics. Subsequently, a Support Vector Machine (SVM) One-Class classification method is utilized for outlier detection, accurately identifying clustered anomalies in multibeam data. An adaptive weighting function based on M-estimation theory is then developed to dynamically adjust data point weights, further suppress systematic noise and signal non-uniformity. This approach minimizes reliance on repeated measurements or extensive subsequent data, improving computational efficiency and algorithmic stability. Experimental results demonstrate superior performance in challenging marine environments, achieving a root mean square error (RMSE) of 46.35, a mean absolute error (MAE) of 13.56, and a discrete point count of 35,370. Compared to Fast Fourier Transform (FFT) and moving average filtering methods, the proposed method significantly enhances data smoothness and consistency. Validation through simulations and real-world sea trials confirms its excellent robustness and real-time accuracy across diverse terrain conditions, providing an efficient solution for seabed terrain mapping, marine engineering, and underwater infrastructure inspection. By integrating machine learning-driven outlier detection with multi-stage data optimization strategies, this study offers an innovative approach to multibeam bathymetric data processing, with significant theoretical and practical implications. Future research will focus on optimizing real-time processing algorithms to address the demands of larger-scale and more dynamically complex marine datasets.