HyMambaNet: Efficient Remote Sensing Water Extraction Method Combining State Space Modeling and Multi-Scale Features.

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Artificial Intelligence

Background:

• Accurate water body segmentation is vital for environmental monitoring and resource management.
• Challenges include scale variations, blurred boundaries, and complex backgrounds in remote sensing data.
• Existing methods struggle with small and morphologically complex water features.

Purpose of the Study:

• To develop a robust and scalable deep learning framework for high-precision water body extraction.
• To address limitations of current methods in segmenting diverse and complex water bodies.
• To improve water resource management and ecological monitoring through advanced remote sensing analysis.

Main Methods:

• Proposed HyMambaNet, a hybrid deep learning model combining convolutional local feature extraction with Mamba state space models.
• Incorporated multi-scale and frequency-domain enhancement for improved boundary precision.
• Utilized optimized skip connections to enhance segmentation robustness.

Main Results:

• HyMambaNet significantly outperformed existing Convolutional Neural Network (CNN) and Transformer-based methods.
• Achieved 74.82% IoU and 88.87% F1-score on the LoveHY dataset, surpassing UNet.
• Attained 81.30% IoU and 89.99% F1-score on the LoveDA dataset, outperforming advanced models.

Conclusions:

• HyMambaNet offers an efficient and generalizable solution for water body extraction from remote sensing imagery.
• The model demonstrates superior performance in handling complex hydrological and ecological scenarios.
• Findings support large-scale water resource monitoring and ecological applications.