| 摘要: | 臺灣海峽由於其特殊地理位置及環境背景，造就其複雜的自然生態及豐富的資源。因臺灣西部主要為沙質海岸，在輸送與沉積作用下使海峽內呈現平緩的地形及廣大的海砂分布區，深度一般不超過100公尺，在臺灣灘區域更淺於20公尺。近期，此區域內蘊藏的豐富海砂資源也引來了不少非法業者的覬覦，導致在臺灣海峽的範圍內經常發生非法越界盜採海砂的事件，尤其以臺灣灘、金門和馬祖列島這三個區域最為嚴重。而盜採海砂對於臺灣而言，除了破壞生態之外，影響最大的便是海岸線倒退所造成的國土流失問題。
為了解決這項問題，本研究的主要目標在於利用衛星遙測中的合成孔徑雷達 (Synthetic Aperture Radar, SAR) 技術，並監測特定區域內不明船隻在時空上的熱點。星載合成孔徑雷達具有穿透性強、解析度高以及全天候成像等優點，特別適合做為海面觀測工具。以歐洲太空總署所提供的 Sentinel-1 SAR 影像作為資料源，並選取三幅涵蓋臺灣海峽的圖框，以觀察近年來非法採砂案件較多的主要目標區域。研究中使用2017年至2021年的影像以分析熱點隨時間的變化。針對抽砂船以及運砂船開發出一套偵測系統，其流程主要由四個步驟所組成，分別是影像前處理、陸地遮罩、預篩選以及判識與分類。實驗的結果顯示這套船隻偵測系統具有自動偵測特定大小目標之能力。該系統在十五張隨機選擇的影像上進行了測試，並且以人工判讀的結果進行驗證，驗證結果F1-score達到平均0.88的準確度。至於熱點時空分析，結果顯示出在這三個目標區域中，分別呈現了不同的熱點分布情況，偵測數量主要在2019年和2020年左右達到峰值。由此可以得到結論，Sentinel-1 SAR 影像具有作為船舶探測工具的能力，而本研究所提出的偵測系統則可以有效的監測非法越界盜採海砂事件的分佈情形，有利於提升後續執法效能。
;With the unique geological setting, Taiwan Strait was formed by shallow bathymetry and gentle topography composed of sandy substrate types. The depth of this area seldom exceeds 100 m, and it is shallower than 20 m in the Taiwan Shoal area. Recently, the abundant sea sand resources contained in this area have also been coveted by many illegal businesses. Therefore, in recent years, there have been frequent cases of illegal cross-border sea sand mining around the Taiwan Strait, especially in Taiwan Shoal, Kinmen Islands, and Matsu Islands. Apart from destroying marine ecology, the greatest problem of illegal sand mining is the consequential retreat of the neighboring coastline.
To address this problem, the objective of this research aims to take advantage of Synthetic Aperture Radar (SAR) technology in satellite remote sensing and to monitor the spatial-temporal hot spots of unidentified vessels in specific areas. Spaceborne SAR instruments have the advantages of superior penetration, high resolution, and independence from sunlight, making them a great tool for ocean object detection. This research uses Sentinel-1 SAR imagery as the data source. Three image frames covering the Taiwan strait are chosen to observe three main target areas with a higher number of illegal sand mining cases in recent years. Images from 2017 to 2021 are obtained to analyze the temporal variation of the hot spot. A detection system has been developed for sand carrier and sand dredger detection with SAR imagery. The workflow of this detection system is mainly composed of four steps, including preprocessing, land masking, prescreening, and discrimination & classification. Our preliminary results show that the developed algorithm can automatically detect targets over a specific size. The detection system was tested on 15 randomly selected images and the results were validated with manually identified ground truth data. The validation result achieved a detection accuracy, in terms of the F1-score, at 0.88 on average. The spatial-temporal hot spot analysis shows different movements of hot spots between these three target areas while the detection number peaks around 2019 and 2020. It is concluded that Sentinel-1 SAR images could serve as a tool for ship detection, and the detection system is able to monitor the distribution of unidentified ships in specific area, which is conducive to improving the efficiency of follow-up law enforcement. |
