應用語意分割卷積神經網路於建築設計平面機能認知辨識之研究-以住宅平面為例

CCUR > College of Environmental Planning & Bioresources > College of Environmental Design > Department of Architecture and Urban Design & Graduate Institute of Architecture and Urban Planning > project > Item 987654321/48637

Please use this identifier to cite or link to this item: https://irlib.pccu.edu.tw/handle/987654321/48637

Title:	應用語意分割卷積神經網路於建築設計平面機能認知辨識之研究-以住宅平面為例 Applying Semantic Segmentation Convolutional Neural Network to Cognitive Recognition of Architectural Design Functions: a Case Study of Residential Floor Plans
Authors:	溫國忠
Contributors:	建築及都巿設計學系
Keywords:	有深度學習語意分割卷積神經網路建築平面認知 Deep Learning Semantic Segmentation Convolutional Neural Networks Floor Plan Cognition
Date:	2020
Issue Date:	2020-09-17 14:24:07 (UTC+8)
Abstract:	建築設計師可以輕鬆且直觀的僅以純幾何圖形辨識平面圖中的各項資訊，且不需額外加以任何標註，就可以一一指出可能的空間功能名稱。以住宅為例，只看平面圖即可認出何為客廳？何為餐廳？何為浴廁？何為臥室？這樣的指認是如何辦到的？建築設計的專業認知，空間認知能力是最為關鍵且基礎的能力之一，也是建築專業教育中最為重要的一環，無論是空間幾何(如尺度、比例、布局、組構…)或是空間的機能(如功能名稱、形式意義、文化隱函、社會關聯…)都再再反映出建築專業本能的直覺，即建築專業設計的生產操作，尤其平面設計圖則為其所操作最為重要且關鍵的圖面型式之一。本研究的理論包含有深度學習(Deep Learning, DL)、語意分割(Semantic Segmentation, SS)卷積神經網路(Convolutional Neural Networks, CNN)、建築設計平面認知等。整體預期研究項目與成果，第一年為：語意分割深度學習與卷積神經網路相關文獻資料收集整理、建築住宅設計平面實證資料與相關程式系統或網頁搜尋、建構相關集合住宅設計平面辨識之空間物件基礎資料庫、推導建置建築住宅設計平面辨識系統程序模型、辨識系統的語意分割之卷積神經網路架構推導建立、辨識系統程式系統分析與程式編撰、建築住宅設計平面辨識系統平台雛形的建構。第二年為：平面辨識之空間物件與其語意屬性規則的關聯探討、辨識系統平台的修正與擴充、辨識系統平台有效性與準確性的驗證並校估、辨識系統應用效能的檢證與自動化效益評估、建築住宅設計平面辨識系統於設計自動化概念的創意表現等。其結果可供空間專業者研究、教學課題與設計策略參照使用。 Architects can easily and intuitively identify each piece of information in the floor plan with pure geometric figures without any additional labeling, and can point out the names of possible space functions. Taking a house as an example, you can recognize the living room just by looking at the plan. What is a restaurant? What is a bathroom? What is a bedroom? How is this identification done? One of the most critical and basic abilities of architectural design professional cognition, spatial cognition is also the most important part of architectural professional education, whether it is spatial geometry (such as scale, proportion, layout, organization ...) or the function of space (Such as function names, formal meanings, cultural implicit letters, social connections, etc.) all reflect the instinct of the architectural profession, that is, the production operation of the architectural professional design, especially the graphic design drawings, which are the most important and critical diagrams that they operate. One of the face types. The theory of this study includes deep learning (DL), semantic segmentation (SS) convolutional neural networks (CNN), and architectural design plane cognition. The overall expected research projects and results, the first year is: Semantic segmentation deep learning and convolutional neural network related literature data collection and arrangement, building and housing design plane empirical data and related program systems or web search, construction of related collection of house design plane identification Basic database of spatial objects, deduction of program model for construction and design of residential building plane identification system, convolutional neural network architecture for semantic segmentation of identification system, derivation and establishment of identification system program system analysis and program compilation, prototype of building residential design plane identification system platform Construction. The second year is: the exploration of the relationship between the spatial objects of plane identification and their semantic attribute rules, the modification and expansion of the identification system platform, the verification and verification of the effectiveness and accuracy of the identification system platform, the verification and verification of the application performance of the identification system, and the automation benefits Appraisal, design of building and housing design plane identification system in the creative expression of design automation concepts, etc. The results can be used as reference for space professionals' research, teaching topics and design strategies.
Appears in Collections:	[Department of Architecture and Urban Design & Graduate Institute of Architecture and Urban Planning ] project

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	306	View/Open

View Licence

Loading...