An expectation-maximization algorithm for fitting a Finite Weibull Mixture Model on sidescan sonar image histograms

Abstract

In this paper, we model the stochasticity of sidescan sonar images using a Finite Weibull Mixture Model (FWMM). To determine the optimal mixing proportions, shape, and scale parameters of the constituent components, we derive the specific Expectation-Maximization (EM) update equations tailored for Weibull mixtures and demonstrate the model's efficacy in fitting empirical seabed data that was collected using a commercial-grade sidescan sonar system. Our results show that the FWMM effectively captures the inherent variability of seafloor textures, demonstrating that the model remains valid across varying levels of seabed complexity.