|
|
Litton7：Litton視覺景觀分類深度學習模型
Litton7: A Deep Learning Model for the Visual Landscape Classification of Litton
何立智、李沁築、邱浩修
Li-Chih Ho, Chin-Chu Lee, Hao-Hsiu Chiu
|
【摘要】
視覺景觀分類(visual landscape classification)是以視覺特徵進行景觀資源歸類，良好的分類系統可以讓後續的規劃設計順利進行，並且讓資源管理更有效率。Litton自1968年開始在美國林務局進行了一系列的研究，建立起視覺景觀分類與評估方法，其分類架構具有相當的代表性。本研究試圖以深度學習進行Litton視覺景觀分類系統的人工智慧模型訓練，目的在降低視覺景觀資源調查的人力需求，同時增加判斷標準一致性。訓練方法上使用深度學習中的遷移學習(transfer learning)進行模型訓練，結果顯示模型實際使用精確度(precision)達80%，是可實際運用於實務的分類模型，該模型命名為Litton7(https://github.com/lichihho/Litton7.git)，未來模型可朝多類別訓練改進，使其更符合人類對環境分類的習慣。
Visual Landscape Classification (VLC) is the categorization of landscape resources based on visual features. A good classification system can facilitate subsequent planning and design and make increase resource management efficiency more efficient. Litton has conducted a series of studies in the U.S. Forest Service since from 1968, to establish establishing a visual landscape classification and evaluation method, and its classification system is quite representative. This study attempts to train the an artificial intelligence model of of Litton’s visual landscape classificationVLC system with using deep learning. The use of , with the deep learning aims to of reduceing the manpower requirements of associated with visual landscape resource surveyance and as well asand to increaseing the consistency of judgment standards. The training method uses transfer learning to train the model, and the. The results show that indicate a model the accuracy of the model reaches up to 80%, which is a classification model that can be indicating that the model can be practically applied in the field. This model, named Litton7 (https://github.com/lichihho/Litton7.git), has the potential for future improvements by incorporating multi-class training, making it more amenable to environment classifications.
In the future, the model can be improved to encompass multi-class training, so that it can be more in line with themaking it more amenable to human habit of classifying the environment classification. Litton7 can be obtained from the following website: (https://github.com/lichihho/Litton7.git).
|
【關鍵字】
人工智慧、遷移學習、景觀規劃、景觀資源、視覺景觀
Artificial intelligent, Transfer learning, Landscape planning, Landscape resource, Visual landscape
|
|
|