LDLK-U-Mamba: An Efficient and Highly Accurate Method for 3D Rock Pore Segmentation

Area of Science:

• Geoscience and Computational Imaging
• Focuses on 3D rock pore segmentation for geological and petroleum exploration applications.

Background:

• Accurate 3D rock pore segmentation is vital for oil and gas resource exploration.
• Existing methods struggle with capturing 3D spatial relationships and exhibit high computational costs.
• Limitations hinder efficiency in resource-constrained environments.

Purpose of the Study:

• To introduce a novel, lightweight method for precise and efficient 3D rock pore segmentation.
• To address the limitations of existing segmentation techniques in capturing spatial context and computational demands.

Main Methods:

• Proposed LDLK-U-Mamba, a Mamba architecture-based model for 3D rock pore segmentation.
• Incorporated a Lightweight Dynamic Large Kernel (LDLK) module for global contextual information.
• Utilized an InceptionDSConv3d module for multi-scale feature fusion and a Basic Residual Depthwise Separable Block (BasicResDWSBlock) with SE module for efficiency.

Main Results:

• LDLK-U-Mamba demonstrated superior performance compared to mainstream segmentation methods.
• The method effectively captures the intricate 3D spatial relationships of rock pores.
• Achieved precise and efficient segmentation, validated through extensive experiments.

Conclusions:

• LDLK-U-Mamba offers an effective solution for 3D rock pore segmentation.
• The proposed architecture successfully balances accuracy and computational efficiency.
• This method holds significant promise for geological and petroleum exploration applications.