Seafloor habitat mapping using multibeam bathymetric and backscatter intensity multi-features SVM classification framework

Abstract

Seafloor habitat mapping plays an important role in marine environmental protection and marine database management. Multibeam echo-sounding system (MBES) has unique advantages in detecting seafloor habitats with its high-precision and full-coverage capability. To further explore the mapping technology of seafloor habitat, this paper builds multi-features support vector machine (SVM) classification framework using multibeam bathymetry and backscatter intensity. First, a training sample robustness evaluation method is proposed based on the Jeffries-Matusita (J-M) index, improving the quality of priori samples input classifier. Second, an SVM classification method based on an Askey-Wilson polynomial kernel function is proposed to optimize the mapping function of kernel function structure and related parameters to the high-dimensional acoustic feature space. Combining multi-beam bathymetry and backscattering intensity together with their derived features, the proposed framework is implemented for automatically classification of five sediment types in the shallow water of southern Wellington, New Zealand. The analysis results show that, compared with the 3 commonly used kernel functions of SVM, the proposed method is more suitable for multibeam seafloor habitat mapping, and that the overall classification accuracy and Kappa coefficient have reached 90.02% and 0.87, respectively. This also highlights the greater potential of multi-beam high-precision detection technology in the prediction of seafloor habitat types.