使用深度學習進行路樹生長評估

Street Tree Growth Assessment by Using a Deep Approach

許子文(Tzu-Wen Hsu)

指導教授 ： 陳履恒

國立暨南國際大學/科技學院/資訊工程學系/碩士(2022年)

https://doi.org/10.6837/ncnu202200119

摘要

要成為智慧城市很重要的要素之一是能夠管理都市綠地智慧化，將都市打造成智慧環境，不僅僅能大幅地增加都市品質，還能顯著地促進碳固存(carbon sequestration)，因此都市綠地智慧化是成為智慧城市的重要指標之一。本研究提出一個自動化監控系統，其系統結合AI技術以及即時攝影機，首先，我們利用先前本研究室開發出的Google Street View自動擷取街景系統以及深度學習技術取得我們的路樹影像資料庫，除了上述方法，我們還利用Selenium套件即時地抓取靜止式網路街景攝影機影像擴充資料庫。之後我們利用TensorFlow所開發的物件偵測API自動地框出樹的位置。我們使用U-Net並觀察路樹的分割結果，除此之外，我們還採取Progressive refinement策略來擴張訓練資料且增加分割結果準確率。實驗結果顯示我們的系統可以成功地像素級deep-detect路樹覆蓋率，換句話說，我們的系統可被用來即時為智慧城市監測碳固存。最後我們將樹木占比率算出，並展示了我們系統能以更智慧的方式監測和管理城市綠地的潛力。

關鍵字

智慧城市 ； 深度學習 ； 樹木生長評估 ； 物件分割 ； 即時攝影機

並列摘要

The intelligent urban green areas are one of the key infrastructures of a smart city. It does not only greatly enhance the livability and sustainability of a city, but also significantly contributes to the carbon sequestration from the atmosphere. Therefore, the development of "intelligent urban green areas" becomes an important indicator of smart cities. In this research, we propose an automatic monitoring system, which assesses the growth of street trees by integrating Artificial intelligence technology and live webcam images. First, we developed a street tree image database by utilizing the deep learning technologies and the image resources of Google Street View. Then we used the Selenium to capture the static live street webcams in real time, and use the object detection API of TensorFlow to automatically box the bounding areas of trees. We used the U-Net to segment the area of street trees. Moreover, we adopted a semi-automatic dataset expansion strategy to expand training data and increase the accuracy of the segmentation results. The experimental results showed that our system can successfully deep-detect the street trees' pixel coverage which can be used as an important real time indicator of carbon sequestration in smart cities. Finally, we calculate the ratio of trees to show the potential of our system for monitoring and managing urban green areas in a more intelligent manner.

並列關鍵字

Smart City ； Deep Learning ； Tree Growth Assessment ； Object Segmentation ； Live Camera

參考文獻

[1] Eremia, Mircea, Toma, Lucian, Mihai, Sanduleac, "The Smart City Concept in the 21st Century.," Procedia Engineering, pp. 181:12-19, 12 2017.

[2] Sánchez-Corcuera, R., Nuñez-Marcos, A., Sesma-Solance, J., Bilbao-Jayo, A., Mulero, R., Zulaika, U., Azkune, G., Almeida, A., "Smart cities survey: Technologies, application domains and challenges for the cities of the future," International Journal of Distributed Sensor Networks, 10 6 2019.

[3] K. Su, J. Li and H. Fu, "Smart city and the applications," in 2011 International Conference on Electronics, Communications and Control (ICECC), 2011.

[4] Harrison, C., Donnelly, IA., "A theory of smart cities," in Proceedings of the 55th annual meeting of the ISSS, 2011.

[5] I. Jonckheere, S. Fleck, K. Nackaerts, B. Muys, P. Coppin, M. Weiss and F. Baret, "Review of methods for in situ leaf area index determination: Part I. Theories, sensors and hemispherical photography.," Agricultural and Forest Meteorology., pp. 121(1):19-35, 1 2004.