整合YOLO與AI Agents之風水評估及房地產服務系統

Abstract

在華人文化 中，『風水』經常被視為購屋決策的重要依據，購屋者會透過風水師的專業諮詢或自行參考網路資訊，利用房屋平面圖判斷格局的優劣。此風水評估流程多仰賴人工判斷，耗時、不穩定且成本較高。為此本研究提出一套使用YOLO影像辨識技術的自動化風水評估方法，整合了布雷森漢姆直線演算法與形態學影像處理技術，能自動偵測房屋平面圖中的風水沖煞格局，例如「門對門」或「大門對廚房」等。同時；本研究將風水評估與房地產領域為導向結合，開發了一套結合AI助理（AI Agents）架構的聊天機器人，並導入LINE平台做為企業與消費者之間的溝通橋樑。該聊天機器人整合LangChain框架與GPT-4o模型，可呼叫不同類型之AI助理工具執行不同任務。整體而言，相較於過去研究之系統須由人工標記物件，且缺乏格局暢通驗證，本研究改善了過去研究的不足，完成全自動檢測風水沖煞之系統，辨識成功率F1 score達0.89。此外，AI 助理的房屋查詢功能在實驗中達到 99.47% 的正確率，為房地產產業提供了具體且實用的AI技術應用案例。

In Chinese cultural contexts, feng shui is often regarded as a pivotal criterion in residential decision-making. Home buyers usually consult professional feng-shui practitioners or refer to online resources and use floor plans to judge whether a layout is auspicious or inauspicious. This evaluation process remains largely manual, which makes it time consuming, inconsistent, and costly. To address these limitations, the present study proposes an automated feng-shui assessment framework that synthesizes YOLO-based object detection, the Bresenham line algorithm, and morphological image processing. The system automatically identifies unfavorable layouts on floor plans, such as a door directly facing another door or the main entrance aligned with the kitchen. The research further links feng-shui assessment to real-estate services by developing a chatbot built on an AI agents architecture and deploying it on the LINE platform to facilitate communication between companies and consumers. Leveraging the LangChain framework and the GPT-4o model, the chatbot can invoke various AI tools to accomplish diverse tasks. Compared with earlier approaches that required manual annotation and lacked validation of spatial continuity, the proposed method achieves fully automated detection of feng-shui conflicts and attains an F1 score of 0.89. The AI agent’s real-estate retrieval function achieved 99.47 percent accuracy in experimental evaluations, demonstrating the practical utility of AI in the real-estate domain.