Laser Ablation Image Segmentation using Vision-Language Models

doi:10.1117/12.3086681

Laser Ablation Image Segmentation using Vision-Language Models

2026-01-01

0

PRE

AI

DOI：10.1117/12.3086681

Original

Original request for help

Save

Abstract

En 中文

Laser-material processing produces complex surface morphologies, including craters, melt displacement, splatter, and vapor recondensation. These observable features encode useful information about laser parameters and material response. We introduce a hybrid segmentation pipeline using Meta AI's Segment Anything (SAM) and Contrastive Language-Image Pre-Training (CLIP) models to generate binary masks on 1,000+ ablation images acquired with a 1064 nm laser over seven orders of magnitude in pulse duration and fluence. Our approach yields feature masks with 90% mean intersection-over-union across four physically relevant feature categories, enabling rapid extraction of quantitative descriptors (e.g., crater diameter, splatter radius) for process-parameter mapping. Sensitivity analyses on resolution, sampling, and runtime demonstrate the method's robustness and pave the way toward real-time, in-situ optimization of laser processing variables.

Keywords:

vision-language models

laser ablation

image segmentation

laser-material interaction

Laser Ablation Image Segmentation using Vision-Language Models

Abstract

Journal

Organization

Cited Papers