透過您的圖書館登入
IP:18.224.62.25
  • 學位論文

以創新機器學習的影像分析 - 以辨識不同海岸垃圾為例

The study of novel machine learning techniques in image classification on different categories of coastal waste

指導教授 : 萬絢

摘要


過去海岸篩檢仰賴以人進行實地探勘來決定海岸是否存有大量人為垃圾,這樣的方法既耗時又費力,若能透過遙測影像獲得圖像資料並以影像辨識技術進行機器學習協助垃圾判釋,即能節省大量時間以及人力資源。 本研究將透過一般譜影像對海岸周圍進行影像辨識,以區分海岸邊人為垃圾和非人為垃圾。研究設計的策略為導入多種紋理資訊以輔助分類成效,在分類器使用支持向量機(Support Vector Machine, SVM)、隨機森林(Random Forest)、類神經網路進行分類,然後產生誤差矩陣並繪製主題圖。 本研究主要研究地區為新北市瑞芳區台2線海岸邊,主要以綠地、岩石、漂流木、人為垃圾4種地貌作為辨識。本研究第一步為獲取海岸一般譜影像,將所產生訓練資料和測試資料藉由資料前處理進行整理,透過組合不同的紋理資訊以及分類器進行比較,最後產生出誤差矩陣以及繪製主題圖來比較兩種分類器辨識準確度。 自動編碼器(Auto-Encoder, AE)將資料進行前處理後輸入至各分類器進行分類辨識,並比較有無AE處理後的誤差矩陣以及主題圖。

並列摘要


In the past, coastal monitoring relied on people to conduct on-site exploration to determine whether there is a large amount of man-made garbage. This method is very time-consuming and labor-intensive. If image data can be obtained through remoted image data with machine learning with image classification technology, which can save a lot of time and human resources. Accordingly, this study the image recognition around the coast will be carried out through general spectrum imagery to distinguish the waste on the coast. The strategy of the study is to import a variety of texture information to assist the classification effect, using Support Vector Machine (SVM), Random Forest (Random Forest) and neural-like network in the classifier. Then generate an error matrix and plot the thematic map. The main research area of this study is the coast of Taiwan Line 2, Ruifang District, New Taipei City, which includes 9 landforms including seawater, roads, green spaces, cars, parking lots, rocks, wave-absorbing blocks, driftwood, and man-made garbage. The first step of this research is to obtain the general spectral image of the coast. The generated training data and test data are organized by data preprocessing, and then compared by combining different texture information and classifiers, and finally an error matrix is generated and a theme map is drawn. Auto-encoder(AE) is used to examine to preprocessing progress of the input image data through those machine learning classifiers. The difference with/without AE is compared with error matrix and thematic.

參考文獻


周榮祥(2004)。以資料探勘技術分析股市個人投資績效之研究。世新大學資訊管理學系碩士論文。
陳佑耕(2014)。以遙測影像的水稻田影像判釋人工蜂群聚類演算法之研究。嶺東科技大學資訊管理碩士班碩士論文。
謝旻修(2018)。以遙測影像資料分析植生覆蓋與台灣南部山崩發生之關聯。國立成功大學資源工程研究所碩士論文。
邱盛鴻(2017)。地方審計人員應用GIS情形與影響因素之研究。東海大學公共事務碩士專班論文。
方寶國(2018)。地理資訊系統在環保局空污稽查管制工作的應用─以南投縣為例。國立中興大學環境工程學系在職專班碩士學位論文。

延伸閱讀