TITLE:
Interactive Identification of Seismic Faults with the Segment Anything Model
AUTHORS:
Mengchao Zhao, Zhonghua Ma
KEYWORDS:
SAM, Earthquake Faults, Interactive Identification
JOURNAL NAME:
Journal of Applied Mathematics and Physics,
Vol.13 No.12,
December
12,
2025
ABSTRACT: Accurate identification of seismic faults is critical to seismology, geological modeling, and disaster assessment. Traditional manual interpretation entails heavy reliance on domain expertise and suffers from inefficiency and subjectivity. Although deep learning-based segmentation has advanced the field, these methods remain limited by a dependence on scarce high-quality labeled data and a lack of generalization flexibility. The Segment Anything Model (SAM), a powerful vision foundation model, offers a breakthrough with its robust zero-shot transfer capabilities and prompt-driven mechanism. This study investigates the application of SAM to seismic fault interpretation. We propose an interactive prompting strategy tailored to the linear and discontinuous nature of faults, enabling users to guide the model in accurately segmenting complex fault structures via sparse prompts, such as points or bounding boxes. Experimental results indicate that this approach significantly lowers the technical barrier and reduces the time required for interpretation. By enhancing both flexibility and efficiency without compromising accuracy, this method presents a promising new solution for intelligent seismic interpretation.