Classification of Marine Sediment in the Northern Slope of the South China Sea Based on Improved U-Net and K-Means Clustering Analysis

• Published in: Remote sensing (Basel, Switzerland) Switzerland), 2023-07, Vol.15 (14), p.3576
• Main Authors: Zhou, Qingjie, Li, Xishuang, Liu, Lejun, Wang, Jingqiang, Zhang, Linqing, Liu, Baohua
• Format: Article
• Language: English
• Subjects: Acoustics; Algorithms; Artificial neural networks; Bathymetry; Classification; Classification (sedimentation); Cluster analysis; Clustering; Coding; Environmental engineering; Fractals; Geological sampling; Geology; Grain size; Gravel; improved U-Net; Information management; Investigations; Marine ecology; Marine engineering; Marine resources; marine sediment classification; Marine sediments; multibeam bathymetry; Neural networks; Northern Slope of the South China Sea; Ocean bottom; Ocean floor; Particle size; Remote sensing; Resource exploitation; Sand; Sediments; Sediments (Geology); Semantics; Silt; Software; Sonar; sub-bottom profile; Topography; Vector quantization
• ISSN: 2072-4292; 2072-4292
• DOI: 10.3390/rs15143576

The classification of marine sediment based on acoustic data is crucial for various applications such as marine resource exploitation, marine engineering construction, and marine ecological environment maintenance. It serves as a valuable alternative to limited geological sampling. However, the accuracy of sediment classification is limited due to constraints in acoustic data detection methods, data quality, and classification techniques. To address this issue, this study proposes an automatic classification method for marine sediment using an improved U-convolutional neural network and K-means clustering algorithm. In the coding part, a spatial pyramid pool layer is introduced to fuse low-dimensional feature data of different scales with the features of each level of the corresponding coding layer. This fusion method enhances the accuracy of the constructed relationship between the physical property parameters of the seabed bottom. The K-means clustering algorithm is optimized through selecting the point at the density center as the initial clustering center during the initial clustering center selection stage. This approach solves the sensitivity problem of the initial clustering center of K-means, improves the edge extraction effect of sediment types, and enhances the classification accuracy of sediment types. To validate the proposed method, an application test is conducted in the Northern Slope area of the South China Sea. The mean grain size of sediments in the study area is predicted using the improved U-Net neural network and the seafloor reflection intensity of the sub-bottom profile. Compared to the standard U-Net network results, the mean grain size prediction results show an increase of 4.9% and 2.8%, respectively. The sediment with the predicted mean grain size is then classified using the K-means clustering algorithm, resulting in the division of five sediment types: gravelly sand, sand, silty sand, sandy silt, and clayey silt. These classifications align well with the South China Sea sediment type map. The findings of this study not only provide an important supplement to existing marine sediment classification methods but also contribute significantly to understanding the sedimentary environment and processes in the Northern Slope of the South China Sea.