本研究以指紋特徵比對系統為驗證案例，探討結合 MIAT 方法論系統設計與大型語言模型的系統設計和高階合成方法。MIAT 方法論著重在系統的階層式模組化的架構設計，以及演算法的離散事件建模，最後合成可維護、可擴展的系統程式碼。在方法論的最後階段，我們引進大型語言模型生成每個離散事件狀態所驅動所需要的程式碼。為了驗證這個方法，我們設計了指紋特徵比對系統的階層模組，包括旋轉排除模組、線段方向排除模組及特徵比對模組。實驗部分，基於FVC2004 指紋資料集進行系統性能測試，透過不同比對演算法如Minutia 與FLANN 的比較，驗證了系統在準確性與運行效率上的表現。實驗結果顯示在模組化設計下，各模組的獨立性和可靠性得到了充分驗證。大型語言模型在程式碼生成中的應用，提升了開發效率並降低人為錯誤的風險。此研究不僅對於指紋辨識技術的進一步發展具有啟發性，也提供了一種創新的系統設計方法，適用於其他高複雜度的軟體開發領域。;This research uses a fingerprint feature matching system as a validation case to explore system design and high-level synthesis methods that combine MIAT methodology with Large Language Models (LLMs). The MIAT methodology focuses on hierarchical modular architecture design and discrete event modeling of algorithms, ultimately synthesizing maintainable and scalable system code. In the final stage of the methodology, we introduce LLMs to generate the necessary code driven by each discrete event state. To validate this approach, we designed hierarchical modules for the fingerprint feature matching system, including rotation elimination module, segment direction elimination module, and feature matching module. For experimentation, we conducted system performance testing based on the FVC2004 fingerprint dataset, comparing different matching algorithms such as Minutia and FLANN to validate the system′s performance in terms of accuracy and operational efficiency. The experimental results demonstrated that under the modular design, the independence and reliability of each module were thoroughly validated. The application of LLMs in code generation improved development efficiency and reduced the risk of human error. This research not only provides insights for further development of fingerprint recognition technology but also offers an innovative system design method applicable to other complex software development domains.
