With the ongoing advancement and development of AI technology, the use of large models in the field of defect classification in research and development will continue to broaden and deepen. The crucial question we need to ponder is how to unleash the true potential of these large AI models to assist R&D teams in accelerating innovation, given the challenges of limited sample data, complex sample defects, and the absence of regular patterns.