應用混合智慧於海洋生物多樣性資源之管理研究

Abstract

人類對近海資源的利用，由傳統漁業發展至蓬勃的藍色經濟模式，尤其在沿海休閒活動方面特別顯著。然而，當代如人口增長、都市化和氣候變化等環境衝擊已造成潮間帶生態系統顯著的生態影響，需要從科學、社會、經濟、法律等，多元方法進行海岸生態資源的永續管理。在日益增加的旅遊壓力下，小琉球和三貂角馬崗的海岸，面臨了經濟利益與環境保護之間的衝突。因應的海洋保育法尚在進行立法過程，現實存在管理上的窘境，也凸顯了社區自發參與保育倡議的迫切性。本研究探究如何運用AI與人之間的協力合作下，善用人工智慧視覺識別技術在海洋資源管理中的應用，重點關注綠蠵龜保育和通過結合公民科學和人工智慧協同的混合智慧（HI）框架，促進公民參與，希望達到教育遊客改變遊憩行為模式。在AI辨識綠蠵龜研究中，發現影像多樣化的小型訓練集，即可以訓練出高效能的YOLOv5s模型，提供未來運用AI調查野生動物的資料管理需求。三貂角混合智慧研究中，在AI的識別輔助下，不僅能促成公民科學參與者對科學數據的集體調查，且能利用機器學習結果來衡量與增進公民科學調查數據的品質，因而能評估調查協議策略的優缺點，且產生一個交互的、持續的、迭代增強混合智慧學習環境，在同時兼顧聯合國永續發展指標中的「良好工作與經濟發展」、「負責任的消費與生產」與「維護海洋資源」，對於未來海洋保護區的設定和管理提供社會與科技（Socio-technology）綜合模式、確保在海岸生物多樣性資源保護的效能和永續經營。

The utilization of littoral zone resources by human populations has evolved from traditional fisheries to a burgeoning blue economy paradigm, notably in coastal recreation. However, contemporary challenges such as population growth, urbanization, and climate change have precipitated notable ecological impacts on intertidal ecosystems, necessitating multifaceted approaches to ensure sustainable management. Studies in Little Liuqiu and Cape Santiago illustrate this dynamic, showcasing the tension between economic imperatives and environmental stewardship in the face of increasing tourist pressures. Despite legislative efforts, regulatory gaps persist, highlighting the importance of community-led conservation initiatives. This study aims to explore the integration of visual recognition AI technologies in marine resource management, focusing on green sea turtle conservation and citizen participation through a hybrid intelligence (HI) framework combining citizen science (CS) and AI-assisted learning methodologies. Through this experimentation, the streamlined YOLOv5s model consistently eclipsed its more complex counterparts in performance. The Santiago HI initiative not only streamlines the collective gathering and AI-assisted analysis of critical data but also utilizes machine-learning outputs to assess data quality, informing subsequent data collection and refinement strategies. This process fosters a mutual and continuous HI learning environment through collective and iterative enhancement. Our HI model plays a crucial role in promoting community engagement and public participation in CS efforts, developing the skills needed to document changes in rocky intertidal biodiversity. These efforts are essential for guiding the design and socio-technology governance of future Marine Protected Areas (MPAs), ensuring their effectiveness and sustainability in alignment with SDGs 8, 12, and 14 in marine conservation.