High-accuracy seabed sediment classification using multi-beam acoustic backscatter data

The seabed sediment classification map has been widely applied in marine science and engineering for providing high-resolution and high-accuracy sediment type. However, some factors degrade the resolution and identification of sediment type, which include calculation of initial incidence angle of beam, backscattering intensity extraction method and coordinate calculation for the echo samples, echo intensity correction, wavelet-BP neural network (WBPNN) improvement and so on. To generate a highly accurate seabed sediment map, a five-step sediment classification method using multi-beam acoustic backscatter data was proposed. First, initial incidence angle of beam was calculated accurately taking into account vessel attitude and beam assigned angle, and echo samples were extracted in the way of grid within the beam footprint, meanwhile, echo samples coordinate was calculated according to the geometrical relationship between echo samples and pointing echo sample in beam, footprint. Second, samples intensities were corrected based on Time Varying Gain (TVG), Angle Varying Gain (AVG). Third, a critical step was proposed for determining the image sample width and preferred characteristic parameters based on the relevant principles. In addition, an improving the wavelet-BP neural network with self-adaptive learning rate and momentum factor, and the network parameters initialization was introduced. Ultimately, the selected image samples were trained to classify the seabed sediment map of a selected test area in Jiaozhou Bay. The experimental results indicated that sample identification accuracy and image identification accuracy reached 93.3% and 98.8% respectively, the proposed method can significantly improve the resolution of the seabed sediment map and identification accuracy of sediment classification.