113

3,990 千元

推動環境管理工作-環境衛生

2024/02/22

2024/11/30

余政剛

環管署環境衛生組

廖銘仁

環輿科技股份有限公司

環境監測及檢測

AI影像辨識、海岸廢棄物調查、空拍、無人機

Coastal Waste Survey, Aerial Photography, Aerial Vehicle(UAV), AI Image Recognition

林韋志

吳士愷

02-27788500#326

shihkaiwu@gi-tech.com.tw

行政院於109年推動「向海致敬」政策，目標為確立及界定海岸清潔，從「清理、減量、去化、透明、教育」5個面向著手，統籌各部會、縣市權管單位妥適分工，使全臺每一吋海岸均有人負責清潔、管理與維護。目前各權管單位，以及國際間相關環保團體，執行海岸/海灘環境稽查作業，仍仰賴稽查人員實地走訪，快篩調查為主要調查方式。本計畫著重於如何將環境稽(巡)查管理與新興科技－無人飛行載具(UAV)結合，研擬具備執行效率與即時稽查的智慧管理機制。 本計畫執行15趟次空拍作業，共進行22次海岸髒亂通報，清理總量為582.057噸；蒐集國外5篇研究無人飛行載具及海洋垃圾辨識之學術論文式，並持續以相關研究為基礎精進本計畫海岸廢棄物自動辨識(AI)功能，準確率與專人框選比對已達80%；AI影像辨識功能開發建置，完成56,00張過去海岸空拍影像標註，並將影像中的海岸廢棄物區分為17種類，共標註出36,626張廢棄物圖片，並將數量較多的漂流木、竹木、蚵棚浮具開發種類辨識功能；另精進現有廢棄物辨識圈選功能，使標註結果更符合廢棄物形狀，減少標註廢棄物面積計算誤差，並更新相關功能；擴充「海岸清潔維護空中稽(巡)查管理系統」，開發權限管理功能、空拍作業執行成果儀表板、廢棄物熱區分析功能、海岸權責單位圖層匯入及廢棄物種類樣態圖鑑開發；舉辦4場次地方試辦作業，分別於新北市白沙灣、桃園市白玉海濱、嘉義縣八掌溪、臺南市青草崙；並配合系統功能開發，調整「海岸環境空中稽(巡)查」作業程序。

AI影像辨識、海岸廢棄物調查、空拍、無人機

成果報告定稿.pdf

16MB

2025/1/30 上午 12:00:00

In 2020, the Executive Yuan initiated the "Salute to the Sea" policy with the goal of establishing and defining coastal cleanliness through the five aspects of "cleaning, reduction, disposal, transparency, and education." The coordination of various government departments and local authorities was organized to ensure that responsible individuals oversee the cleaning, man-agement, and maintenance of every inch of Taiwan's coastline. Currently, both government agencies and international environmental organizations re-ly on on-site visits by inspection personnel as the primary method for coastal/beach environmental inspections, with rapid screening investigations being the main investigative approach. This project focuses on integrating environmental inspection management with emerging technology – Un-manned Aerial Vehicles (UAVs) – to develop an intelligent management mechanism capable of efficient execution and real-time inspections. This project conducted 15 aerial photography operations and made 22 coastal litter reports, with a total cleanup volume of 582.049 tons. We col-lected 5 international academic papers on UAV and marine debris identifica-tion research, and continued to improve the automatic coastal waste identifi-cation (AI) functionality based on related research, achieving an accuracy rate of 80% compared to manual selection. For AI image recognition devel-opment, we completed the annotation of 56,00 previous coastal aerial imag-es, categorizing coastal waste into 17 types and annotating 36,626 waste images. We developed specific identification functions for the most common items: driftwood, bamboo/wood, and oyster rack floats. Additionally, we improved the existing waste identification selection function to better match waste shapes, reducing errors in waste area calculations, and updated related features. We expanded the 'Coastal Cleaning and Maintenance Aerial Inspec-tion Management System' by developing access control functions, aerial op-eration performance dashboards, waste hotspot analysis capabilities, coastal jurisdiction unit layer import, and waste type pattern atlas. The project compiled the execution of coastal aerial operations by 19 local environmental protection bureaus, organized four local trial operations in Baishawan, New Taipei City, Baiyu Beach, Taoyuan City, Bajhang River, Chiayi Country, and Qingcaolun, Tainan City. We also adjusted the 'Coastal Environment Aerial Inspection' operation procedures in accordance with system function development.

Coastal Waste Survey, Aerial Photography, Aerial Vehicle(UAV), AI Image Recognition