創新機器學習裝袋算法在影像物件上辨識與估算以東北角海岸廢棄物為例

海岸垃圾目前造成台灣一個嚴重的問題在於上游崩塌漂流木和下游海岸人造垃圾，導致原本的海岸地貌加劇改變。過去海岸地貌調查作業都需要動用大量的時間與人力來進行。然而隨著科技的日新月異，現今地理資訊系統（Geographic Information System, GIS）的蓬勃發展以及遙測技術（Remote Sensing, RS）能在短時間內擷取大範圍量化資訊、獲取資訊受限制條件少的特性，有越來越多的實際應用取代了過往的傳統調查。本實驗藉由遙測技術結合影像分類的應用，以多元分類（Multiclass classification）及集成學習（Ensemble learning）的模型訓練方式進行實作，比較不同演算法所得出的分類成效差異，並與實際地貌進行差異化分析，應用資料視覺化模組繪製出主題圖以展示不同物件於地表之分布情形，供後續各物件之間的估算作業使用。由於裝袋算法是整體學習的一種代表性方法，本研究以決策樹（Decision Tree）和裝袋算法（Bootstrap aggregating or Bagging）的比較，再與人工檢核地貌的成果進行比較，製作出誤差矩陣表，藉此評估不同的演算法之分類成效，以找出應用於該種地徵資料之最佳的分類演算法。

關鍵字

海洋垃圾；監督式學習；資料視覺化；裝袋算法

並列摘要

In recent years, coastal waste is currently causing a serious problem in Taiwan, which is produced by the landslide of driftwood in upstream and the man-made garbage in downstream. This has led to aggravated changes in the original coastal landscape. In the past, coastal geomorphological survey operations required a lot of time and manpower to carry out. However, with the rapid development of science and technology, the current geographic information system and remote sensing technology can capture a large range of quantitative information in a short period of time. It provides an access to information that is subject to few restrictions. More and more practical applications on remote sensing have replaced the traditional investigations in the past. This study uses remote sensing technology with the application of image classification to implement the model training methods of Multiclass classification by using Ensemble learning. It compares the differences in classification results obtained by different algorithms and compares them with the actual landforms. Differential analysis, using the data visualization module to draw a thematic map to display the distribution of different targets on the ground. The usage in subsequent estimation operations is done between each target. The bootstrap aggregating algorithm is one of the representative methods in Ensemble learning. Thus, the decision tree and the bootstrap aggregating are compared, and then the results on the inspection of the landcover in this study. An error matrix is produced to evaluate the classification effectiveness of different algorithms to find out the best classification algorithm applied to the observation data.

並列關鍵字

coastal waste ； supervised learning ； vector data ； data visualization ； bootstrap aggregating

參考文獻

Fan-Jun Kuo & Hsiang-Wen Huang (2014). Strategy for mitigation of marine debris: Analysis of sources and composition of marine debris in northern Taiwan. Marine Pollution Bulletin, 83(1), 70-78. https://doi.org/10.1016/j.marpolbul.2014.04.019

Leo Breiman (1996). Bagging predictors. Machine Learning, 24(2), 123–140. https://doi.org/10.1007/bf00058655

Piovan, S. E. (2020). Geographic Information Systems. In Springer Geography, 119-170. https://doi.org/10.1007/978-3-030-42439-8_6

Hartigan, J. A. & Wong, M. A. (1979). Algorithm AS 136: A K-Means Clustering Algorithm. Applied Statistics, 28(1), pp.100. https://doi.org/10.2307/2346830

Zhang, J., Liu, W., & Gruenwald, L. (2011). A Successive Decision Tree Approach to Mining Remotely Sensed Image Data. In Data Warehousing and Mining. IGI Global, 2978–2992. https://doi.org/10.4018/978-1-59904-951-9.ch190

國際替代計量

創新機器學習裝袋算法在影像物件上辨識與估算以東北角海岸廢棄物為例

全文下載

主題瀏覽