1 May 2020 in ISPRS Journal of Photogrammetry and Remote Sensing
10 March 2025 in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
The seabed sediment classification plays an im-portant part in marine ecological environment protection and other related fields. To fully explore the application ability of marine geographic information in seabed sediment classification, the paper makes a contribution to overcome the low accuracy and reliability shortcomings of using single data source and traditional classifiers. Based on extracted multi-source features, the scale invariant feature transform (SIFT) - random sample consensus (RANSAC) model is applied to realize feature-level fusion be-tween airborne LiDAR bathymetry (ALB) data and multispectral remote sensing images. Furthermore, a dual branch convolutional neural networks (CNN) classifier is constructed to classify the seabed sediment into five categories (coral reef, sand, gravel, coastal zone, and vegetation). To verify the effectiveness of fused data in seabed sediment classification, experiments were con-ducted using multispectral remote sensing images and ALB data. Experimental results show that the overall classification accuracy and Kappa coefficient of the dual CNN classifier constructed in this paper are 98.2% and 0.977, respectively. In addition, the classification results using multi-source fusion data are higher than those using single-source data, indicating the accuracy and effectiveness of multi-source fusion features for classification. The research results can provide effective technical support for seabed sediment classification.
3 April 2026 in IEEE Transactions on Geoscience and Remote Sensing
Airborne LiDAR bathymetry is particularly suited for acquiring three-dimensional spatial information in shallow water zones. However, its signal undergoes rapid attenuate in the water column, resulting in faint bottom echoes that are challenging to detect and directly limit the maximum bathymetric performance. To achieve accurate extraction of these echoes, a faint bottom echo detection method based on a constrained waveform stacking model is proposed. The model employs distances between water-surface points and their refraction vectors as dual constraints to filter spatially adjacent waveforms. The filtered waveforms are then utilized to construct a pseudo waveform with high signal-to-noise ratio by waveform stacking. On this basis, the accurate faint bottom echo moment is subsequently determined based on a normalized peak significance score. The proposed method is validated using two Mapper-20KU datasets acquired from a coastal zone and an inland reservoir in Qingdao, China. The results demonstrate that the method increases the maximum bathymetric depth by approximately 50% in both regions and significantly improves the point density. Compared with existing topographic data, the root mean square errors (RMSEs) of the derived bathymetric points are 0.22 m and 0.09 m, respectively. Therefore, the proposed method can significantly enhance the maximum bathymetric depth while ensuring the bathymetric accuracy.