Work

Research on Fine Water Body Extraction From SAR Images Based on Superpixel Segmentation

Year: 2025

Type: article

Abstract: The water body extraction technology based on superpixel segmentation in SAR images faces challenges such as insufficient precision in extracting fine water boundaries under speckle noise and complex scattering conditions. To address these issues, this paper proposes an enhanced fine water body extraction method based on superpixel segmentation for SAR images. In the superpixel segmentation phase, a simple linear iterative clustering (SLIC) superpixel segmentation algorithm based on eightdirection convolution (referred to as EDC-SLIC algorithm) is introduced. This algorithm constructs three pseudo-channels using eight-direction convolution to replace the three color channels of traditional color images and employs logarithmic difference measurement in the color distance calculation part of the SLIC algorithm, thereby adapting it to the segmentation requirements of SAR images. In the water body information extraction phase, a multi-feature weighted Otsu water body information extraction algorithm integrating superpixels (referred to as MFW-Otsu algorithm) is proposed. This algorithm integrates local mean and variance into a new feature image through weighting, enabling more accurate representation of texture changes in the image and enhancing the algorithm's ability to process complex image structures. The experimental results demonstrate that the EDC-SLIC algorithm and MFW-Otsu algorithm exhibit significant advantages in accuracy, robustness, and practicality. Furthermore, the integration of superpixels effectively improves the algorithm's adaptability to complex scenes, enhances detail processing capabilities, reduces misclassification phenomena, and improves the accuracy of water body information extraction.

Source: IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing

DOI: 10.1109/jstars.2025.3615837

Authors: Fengcheng Guo, Xiaoxiao Ma, Ning Sun, Lianpeng Zhang, Wensong Liu

Institution: Jiangsu Normal University

Language: English

Topic: Remote-Sensing Image Classification

Subfield: Media Technology

Field: Engineering

Domain: Physical Sciences

SDG: Clean water and sanitation

Open Access status: gold

Estimated APC paid: $1,800