語意分割技術於景觀評估之運用

Application of Semantic Segmentation Technology to Landscape Assessment

心理物理模式（psychophysical paradigm）是景觀評估中的一種方法，該方法著重於了解實質環境的物理特徵對於心理感受的影響。在進行評估時，物理特徵常使用人工的方式描繪影像而取得，然近年來人工智慧技術的發展中，語意分割（semantic segmentation）可由電腦自動化擷取上述之影像特徵，但由於目前語意分割的模型相當多樣，何種模型適用於景觀領域、有何限制，需要進一步的討論與比較。因此，本研究取用目前在語意分割技術中不同的模型25個，比較模型於景觀偏好、自然度與開放度等三項景觀評估指標之表現，結果顯示ADE20K之語意分割模型是目前較適合景觀領域所使用之模型，而Cityscapes模型對於街道景觀的分析效果不錯，然該模型無法分析水體、山岳等非都市地區的景觀元素，其使用較受限制。

The psychophysical paradigm is a method in landscape assessment that focuses on understanding the impact of physical features on psychological perceptions. In the assessment, physical features are often obtained by manually depicting images. With the development of artificial intelligence technology, semantic segmentation can automatically capture physical features, but the current semantic segmentation models are pretty diverse. What models are suitable for the landscape domain, and what are the limitations? Further discussions and comparisons are needed. Therefore, in this study, 25 different models of the semantic segmentation technique were used to compare the performance of three landscape evaluation indicators: landscape preference, naturalness, and openness. The results showed that the ADE20K semantic segmentation model is more suitable for the landscape domain than other models. By contrast, the Cityscapes model is ideal for streetscape analysis. However, it cannot analyze landscape elements in non-urban areas, such as water bodies and mountains, so its use is more limited than that of the ADE20K model.