Automatic Water-Land Segmentation Algorithm via Weighted Mixture Model With Dual Spatial Constraints

Water-land segmentation faces significant challenges due to land-water interface heterogeneity. Existing methods based on spatially constrained statistical models often yield unsatisfactory segmentation results. To address this limitation, this paper proposes an automatic water-land segmentation algorithm based on a Weighted Gaussian Mixture Model with Dual Spatial Constraints (DSC-WGMM). First, a Weighted Gaussian Mixture Model (WGMM) is established to characterize the complex statistical distributions of heterogeneous regions. The dual weights of the WGMM effectively leverage pixel spatial correlations. Next, Markov Random Fields(MRF)are constructed on both subclass and class weights to introduce dual spatial constraints into the WGMM, thereby further reducing the influence of heterogeneous pixels. Finally, model parameters are estimated using the Expectation-Maximization (EM) method, while the optimal smoothing coefficient is determined through a gradient optimization method. This combination enables fast and adaptive water-land segmentation. Numerous experiments on high-resolution remote sensing images demonstrate that the proposed algorithm accurately fits complex histograms, reduces heterogeneous pixel interference, and achieves precise water-land segmentation. The proposed algorithm consistently outperforms comparative methods. It demonstrates a significant improvement in accuracy and adaptivity of water-land segmentation, offering an effective and robust technical method for waterline extraction in complex scenes.