環保專案成果報告查詢系統

專案計畫編號
經費年度	112
計畫經費	3,809 千元
預算科目	環境衛生管理-溫室氣體減量及環境清潔管理
專案開始日期	2023/06/07
專案結束日期	2023/12/20
專案主持人	吳秉耕
主辦單位	氣候署調適韌性組
承辦人	吳霆修
執行單位	環輿科技股份有限公司
專案分類	環境衛生、環境資訊系統
中文關鍵字	海岸廢棄物調查、空拍、無人機、正射影像、AI影像辨識
英文關鍵字	Coastal Waste Survey, Aerial Photography, Unmanned Aerial Vehicle(UAV), Orthophotos, AI Image Recognition
協同主持人
共同主持人
計畫聯絡人	諸百興
計畫聯絡電話	02-27788500#333
計畫聯絡信箱	jeffrey79913@gi-tech.com.tw

中文摘要	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10趟次空拍作業，共進行9次海岸髒亂通報，清理總量為456.438噸；正射影像分析結果廢棄物面積約占3.42%，較去年計畫成果廢棄物面積約占3.85%有略為下降，廢棄物種類部分由於拍攝地點多位於西南部，養殖漁業活動較為盛行，因此在廢棄物佔比較高的種類多為竹排/竹桿(廢棄蚵棚及蚵棚浮具)，佔廢棄物種類比例80%以上；蒐集36式海岸空拍影像素材，並將成果提供機器影像辨識模型進行學習，判釋準確率與人員手動框選比對已達78%；AI影像辨識功能開發建置，針對連續拍攝且相互覆蓋的海岸環境測拍照片，於系統上傳後由系統工具自動完成幾何校正進行影像拼接，產出整段海岸線廢棄物分布的狀況，並透過此影像進行AI辨識廢棄物，可減少單趟任務人力檢視廢棄物通報作業2小時、影像拼接及人工廢棄物數化作業5天；蒐整臨海19縣市地方環保局執行海岸空拍作業情況；舉辦2場次地方試辦作業，分別於新北市白沙灣及桃園市白玉海濱；雲端物聯網及智慧城市參賽協助，獲得「雲端物聯網創新獎」優良應用獎並配合AI影像辨識的所需要的成果規範，調整「海岸環境空中稽(巡)查」作業程序。
中文關鍵字	海岸廢棄物調查、空拍、無人機、正射影像、AI影像辨識

中文摘要

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10趟次空拍作業，共進行9次海岸髒亂通報，清理總量為456.438噸；正射影像分析結果廢棄物面積約占3.42%，較去年計畫成果廢棄物面積約占3.85%有略為下降，廢棄物種類部分由於拍攝地點多位於西南部，養殖漁業活動較為盛行，因此在廢棄物佔比較高的種類多為竹排/竹桿(廢棄蚵棚及蚵棚浮具)，佔廢棄物種類比例80%以上；蒐集36式海岸空拍影像素材，並將成果提供機器影像辨識模型進行學習，判釋準確率與人員手動框選比對已達78%；AI影像辨識功能開發建置，針對連續拍攝且相互覆蓋的海岸環境測拍照片，於系統上傳後由系統工具自動完成幾何校正進行影像拼接，產出整段海岸線廢棄物分布的狀況，並透過此影像進行AI辨識廢棄物，可減少單趟任務人力檢視廢棄物通報作業2小時、影像拼接及人工廢棄物數化作業5天；蒐整臨海19縣市地方環保局執行海岸空拍作業情況；舉辦2場次地方試辦作業，分別於新北市白沙灣及桃園市白玉海濱；雲端物聯網及智慧城市參賽協助，獲得「雲端物聯網創新獎」優良應用獎並配合AI影像辨識的所需要的成果規範，調整「海岸環境空中稽(巡)查」作業程序。

中文關鍵字

海岸廢棄物調查、空拍、無人機、正射影像、AI影像辨識

類型	檔名	檔案大小	說明	最新版公開日期
期末報告	112年海岸環境空中稽(巡)查計畫_成果報告_公開v3.pdf	33MB		2024/2/29 上午 12:00:00

摘要	The Executive Yuan launched the "Respect the Sea" policy in 2020, aiming to establish and define coastal cleanliness. It focuses on five aspects: "cleaning, reducing, disposal, transparency, and education." The coordination among various government departments and local authorities is crucial to ensure that every inch of the coastline in Taiwan is responsibly cleaned, managed, and maintained. Currently, environmental inspections of coasts and beaches by relevant authorities and international environmental organizations rely heavily on field visits by inspection personnel, with rapid screening investigations as the primary method. This project emphasizes the integration of environmental inspection management with emerging technology – Unmanned Aerial Vehicles (UAVs) – to develop an intelligent management mechanism with operational efficiency and real-time inspection capabilities. The project conducted 10 aerial survey operations, responding to a total of 9 reports of coastal pollution, with a cleanup volume of 456.438 tons. The ortho-image analysis results showed that the waste area accounts for approximately 3.42%, a slight decrease from the 3.85% reported in the previous year's project. Due to the locations predominantly in the southwest, where aquaculture activities are prevalent, the waste types with higher proportions are mainly bamboo rafts/bamboo poles (waste oyster racks and oyster rack floats), comprising over 80% of the waste types. The project collected 36 sets of coastal aerial image materials, providing them to machine image recognition models for learning. The accuracy of interpretation compared to manual selection by personnel reached 78%. The development of AI image recognition functionality involved continuous shooting and mutually overlapping photos of coastal environments. The system automatically performed geometric correction and image stitching upon upload, presenting the waste distribution along the entire coastline. Using this image, AI recognition of waste can reduce the manual inspection time for a single mission by 2 hours, and the image stitching and manual waste digitization operation by 5 days. The project compiled information on coastal aerial survey operations conducted by local environmental protection bureaus in 19 coastal counties and cities. Two local trial operations were held at White Sands Bay in New Taipei City and Baiyu Beach in Taoyuan City. Assistance from cloud IoT and smart city initiatives led to winning the "Cloud IoT Innovation Award" Excellent Application Award. The project also adjusted the operational procedures for "Coastal Environmental Aerial Inspection" based on the specifications required for AI image recognition.
英文關鍵字	Coastal Waste Survey, Aerial Photography, Unmanned Aerial Vehicle(UAV), Orthophotos, AI Image Recognition

摘要

The Executive Yuan launched the "Respect the Sea" policy in 2020, aiming to establish and define coastal cleanliness. It focuses on five aspects: "cleaning, reducing, disposal, transparency, and education." The coordination among various government departments and local authorities is crucial to ensure that every inch of the coastline in Taiwan is responsibly cleaned, managed, and maintained. Currently, environmental inspections of coasts and beaches by relevant authorities and international environmental organizations rely heavily on field visits by inspection personnel, with rapid screening investigations as the primary method. This project emphasizes the integration of environmental inspection management with emerging technology – Unmanned Aerial Vehicles (UAVs) – to develop an intelligent management mechanism with operational efficiency and real-time inspection capabilities. The project conducted 10 aerial survey operations, responding to a total of 9 reports of coastal pollution, with a cleanup volume of 456.438 tons. The ortho-image analysis results showed that the waste area accounts for approximately 3.42%, a slight decrease from the 3.85% reported in the previous year's project. Due to the locations predominantly in the southwest, where aquaculture activities are prevalent, the waste types with higher proportions are mainly bamboo rafts/bamboo poles (waste oyster racks and oyster rack floats), comprising over 80% of the waste types. The project collected 36 sets of coastal aerial image materials, providing them to machine image recognition models for learning. The accuracy of interpretation compared to manual selection by personnel reached 78%. The development of AI image recognition functionality involved continuous shooting and mutually overlapping photos of coastal environments. The system automatically performed geometric correction and image stitching upon upload, presenting the waste distribution along the entire coastline. Using this image, AI recognition of waste can reduce the manual inspection time for a single mission by 2 hours, and the image stitching and manual waste digitization operation by 5 days. The project compiled information on coastal aerial survey operations conducted by local environmental protection bureaus in 19 coastal counties and cities. Two local trial operations were held at White Sands Bay in New Taipei City and Baiyu Beach in Taoyuan City. Assistance from cloud IoT and smart city initiatives led to winning the "Cloud IoT Innovation Award" Excellent Application Award. The project also adjusted the operational procedures for "Coastal Environmental Aerial Inspection" based on the specifications required for AI image recognition.

英文關鍵字

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