圖書典藏及影音數位平台

集水區水源保育範圍長期以來之違規開發、超限利用及颱風豪雨造成坡地崩塌及土石滑動等土砂災害，常直接或間接影響水源水質或造成嚴重水庫淤積，威脅水資源之運用與永續，目前集水區內土地之管理，採各權責單位定時巡查有無違規利用情形，以維護集水區之生態保育，惟集水區水源保育範圍廣大，為能加強管理，本計畫將建置水源保育智慧影像管理平台，發展智慧影像判釋技術，進行影像資料彙整及違規開發資料判讀，協助水源保育與土地利用管理。 另本計畫使用無人載具攝影機和後端的影像數據處理，配合疑似土地違規利用案例或集水區水源、水質水量關注議題，取影像資料後透過影像數據處理及分析，作為後續相關案件輔助參考。

This project collected the management of abnormal land use and inspection mechanism for various agencies, and proposed their advantages and disadvantages according to the application method of image interpretation. The collected data showed that only China monitored land use with satellite imagery to understand land changes or abnormal development practices. Other countries in the world mainly focused on the basic data survey and research of remote sensing technology and land information systems. The project collected environmentally sensitive maps and built on the integrated information system "Water Sources Conservation Intelligent Image Management Platform". This study also developed an integrated information system combining information management and photogrammetry technologies such as MIS, GIS and GPS telemetry. Moreover, the project completed eight unmanned aerial vehicle (UAV) photogrammetric exploration missions to designated areas. The mosaic area of Orthophoto images was 870 hectares (8.7 square kilometers) and the average ground resolution (GSD) was 5.31 cm/pixel. The relevant data statistics of the water quality and water conservation areas inspection system in the past three years were completed, and the cases of New Taipei City and Taoyuan City were counted. Among them, 26 were made into a case report. In addition, 9-satellite image map was inquired to judge the difference to explain the image differences before and after the variation of the abnormal land use areas. This project also evaluated the normalized difference vegetation index (NDVI) using hyperspectral remote sensing data for the upper reaches of the Xindian River catchment area from the FORMOSAT-2 satellite with 8-m spatial resolution. The satellite spectral values were converted to monitoring data according to the transformation relationship. With the calculated NDVI, the vegetation location can be determined. In this study, the threshold of the NDVI for green and non-green coverage was 0.05. This value was further used to automatically generate images with a NDVI greater or less than the threshold. The green parts of the image were vegetation areas and the dark gray parts were places where the landscape changed. Recent developments in computing power have led to the expansion of deep learning technology and application of neural networks, especially in terms of image recognition based on convolutional neural networks. This project also interpreted the variation point of abnormal land use areas by the method of convolutional neural network. This study presented the interpretation of smart images in water conservation and land use management, and the planning of spectral images in water resources management. UAV was used to take images, and Altizure network cloud computing was used to establish 3D modeling of the environment around the exploration mission. The study also explored the feasibility of virtual reality VR, augmented real-world AR, alternative real-world SR and hybrid real-world MR in water conservation and land use management, and evaluates the software, hardware and funds required. The study results showed that intelligent image interpretation technology combined with image data acquisition and interpretation can help water resources protection and land use management in catchment areas.