環保專案成果報告查詢系統

專案計畫編號
經費年度	111
計畫經費	6,800 千元
預算科目	固定污染源管制
專案開始日期	2022/07/11
專案結束日期	2023/05/31
專案主持人	彭保仁
主辦單位	空保處
承辦人	戴鴻勳
執行單位	工業技術研究院量測技術發展中心
專案分類	環境監測及檢測、大氣空氣
中文關鍵字	人工智慧、影像辨識技術、煙霧不透光率辨識
英文關鍵字	Artificial Intelligence, Image recognition technology, Smoke opacity identification
協同主持人
共同主持人
計畫聯絡人	彭保仁
計畫聯絡電話	03-57#32251
計畫聯絡信箱	bjpong@itri.org.tw

中文摘要	本計畫運用人工智慧影像辨識技術完成煙霧不透光率辨識，達成計畫成果摘要說明如下。在智慧判煙系統整合及建立雲端服務模組之整體架構雛型上，完成易拆解之智慧影像判煙系統一套，系統之黑白煙不透光率辨識平均不透光率已穩定至 5%~7%之間，黑煙最大誤差可至 15%以下，白煙最大誤差仍大於 15%。在智慧判煙方法草案上，參考國內外方法(例如目測判煙訓練手冊，method 9， ASTM 7520-2016)與專家建議制定出“排放管道排放粒狀污染物之不透光率拍照判定方法”草案。在建立智慧影像判煙辨識模組校驗(校正calibration)方法及其軟體設備驗證程序上，完成室內校正環境建置，包含高照度光源，復刻灰階標準板一套(與 BSI 2742M煙霧灰階標準板的4個等級對應之不透光率最大誤差小於3 %)。完成室內校正程序草案一份，量測品保方案含不確定度評估一份，設備建議清單一份。完成判煙辨識模組驗證7件次，通過驗證之設備達3件。完成智慧影像判煙辨識軟體驗證程序草案一份。台南及桃園判煙班實地考察，建置完成不透光率平均誤差< 7 %的黑煙及白煙標準影像圖庫合計50張。戶外驗證程序草案上，完成戶外驗證程序草案一份，執行戶外模擬驗證實驗，黑煙與白煙各5次。完成影像判煙系統之黑煙模擬驗證一套，驗證結果優於目測判煙的標準。在推廣智慧影像判煙辨識系統說明宣導及其他事項上，完成配合環保署推廣智慧影像判煙辨識系統辦理3場次宣導說明會議合計 126人參與。配合環保署完成每月之計畫工作檢討會議累計15場次。社會經濟可能產生的相關影響上，透過智慧影像判煙系統，未來可做為工廠端自主管理的監控系統，協助達成減碳與零碳的長期目標，減少空氣污染造成的社會與經濟成本。經濟效益分析上，智慧影像判煙系統的效益上，提升時間效率66 %，縮小誤差率 1/3，突破煙流漂移不定，突破照度不足的限制。判煙任務之效益上，減少目測判煙主觀判定的影響。人員培訓之效益上，可降低投入人力，進而降低訓練成本。自動化作業之效益，固定式系統可利用遠端操作，減少人員外派任務之成本，並提高整體監控效益。
中文關鍵字	人工智慧、影像辨識技術、煙霧不透光率辨識

中文摘要

本計畫運用人工智慧影像辨識技術完成煙霧不透光率辨識，達成計畫成果摘要說明如下。在智慧判煙系統整合及建立雲端服務模組之整體架構雛型上，完成易拆解之智慧影像判煙系統一套，系統之黑白煙不透光率辨識平均不透光率已穩定至 5%~7%之間，黑煙最大誤差可至 15%以下，白煙最大誤差仍大於 15%。在智慧判煙方法草案上，參考國內外方法(例如目測判煙訓練手冊，method 9， ASTM 7520-2016)與專家建議制定出“排放管道排放粒狀污染物之不透光率拍照判定方法”草案。在建立智慧影像判煙辨識模組校驗(校正calibration)方法及其軟體設備驗證程序上，完成室內校正環境建置，包含高照度光源，復刻灰階標準板一套(與 BSI 2742M煙霧灰階標準板的4個等級對應之不透光率最大誤差小於3 %)。完成室內校正程序草案一份，量測品保方案含不確定度評估一份，設備建議清單一份。完成判煙辨識模組驗證7件次，通過驗證之設備達3件。完成智慧影像判煙辨識軟體驗證程序草案一份。台南及桃園判煙班實地考察，建置完成不透光率平均誤差< 7 %的黑煙及白煙標準影像圖庫合計50張。戶外驗證程序草案上，完成戶外驗證程序草案一份，執行戶外模擬驗證實驗，黑煙與白煙各5次。完成影像判煙系統之黑煙模擬驗證一套，驗證結果優於目測判煙的標準。在推廣智慧影像判煙辨識系統說明宣導及其他事項上，完成配合環保署推廣智慧影像判煙辨識系統辦理3場次宣導說明會議合計 126人參與。配合環保署完成每月之計畫工作檢討會議累計15場次。社會經濟可能產生的相關影響上，透過智慧影像判煙系統，未來可做為工廠端自主管理的監控系統，協助達成減碳與零碳的長期目標，減少空氣污染造成的社會與經濟成本。經濟效益分析上，智慧影像判煙系統的效益上，提升時間效率66 %，縮小誤差率 1/3，突破煙流漂移不定，突破照度不足的限制。判煙任務之效益上，減少目測判煙主觀判定的影響。人員培訓之效益上，可降低投入人力，進而降低訓練成本。自動化作業之效益，固定式系統可利用遠端操作，減少人員外派任務之成本，並提高整體監控效益。

中文關鍵字

人工智慧、影像辨識技術、煙霧不透光率辨識

類型	檔名	檔案大小	說明	最新版公開日期
期末報告	成果報告.pdf	12MB		2023/7/31 上午 12:00:00

摘要	This project uses artificial intelligence image recognition technology to complete smoke opacity identification, and achieves the project results are as follows: An integration of the intelligent smoke detection system and an overall architecture prototype of the cloud service module has been completed, which includes an intelligent image recognition system. Regarding the opacity recognition of the system, the average opacity recognition rate for black and white smoke has stabilized between 5 % to 7 %, with the maximum error for black smoke being less than 15 %, while the maximum error for white smoke remains above 15 %. The draft of the intelligent smoke recognition method, which is ”A draft of photographic determination method of opacity of particulate pollutants discharged from chimney“ has been formulated, referencing domestic and foreign methods (such as visual smoke recognition training manuals, method 9, and ASTM 7520-2016), as well as expert suggestions. In the process of establishing the verification method for the intelligent smoke image recognition module and its software and hardware verification program, we completed the construction of the verification environment, including a high-intensity light source, a set of replica standard grayscale boards, which’s maximum error is less than 3 % for the four opacity levels corresponding to that of BSI 2742M smoke grayscale standard board. An indoor calibration document for the intelligent image smoke detection and recognition module was completed, including a draft of the indoor calibration procedure, a quality assurance plan with uncertainty assessment, and a recommended equipment list. Seven rounds of equipment verification for image capture and processing were carried out, and three pieces of equipment passed the verification. The software verification program for the intelligent smoke recognition system has been established. Field inspections were conducted at the Taoyuan and Tainan smoke detection centers, a library of 50 standard image data of black smoke and white smoke was established with an average opacity error of less than 7 %. The draft outdoor verification procedure has been completed. A filed test of outdoor verification experiment was conducted with four trials each for black smoke and white smoke. A set of black smoke simulation verification for the smoke detection system was completed, and the verification results exceeded the standards of visual smoke detection. We have completed three promotion and explanation session with 126 attendees to support the Environmental Protection Administration's promotion of the intelligent image-based smoke recognition system and other related matters. Additionally, we have assisted the Administration in conducting 15 regular project work review meetings. Complete the benefit of intelligent smoke identification task, benefit of personnel training and analysis of automated operation analysis for economic benefit analysis by this project. In terms of possible social and economic impacts, the smart image smoke detection system can be used as a factory-side self-managed monitoring system in the future to help achieve the long-term goals of carbon reduction and zero carbon, and reduce the social and economic costs caused by air pollution. In terms of economic benefit analysis, in terms of benefits of the smart image smoke detection system, the time efficiency is increased by 66 %, the error rate is reduced by 1/3, the drift of the smoke stream is uncertain, and the limitation of insufficient illumination is broken. In terms of the effectiveness of the image smoke recognition system, the influence of subjective judgments of visual inspection can be reduced. In terms of the benefit of personnel training, it can reduce the input of manpower, thereby reducing the training cost. The benefits of automated operations, fixed systems can use remote operations, reduce the cost of personnel dispatched tasks, and improve the overall monitoring efficiency.
英文關鍵字	Artificial Intelligence, Image recognition technology, Smoke opacity identification

摘要

This project uses artificial intelligence image recognition technology to complete smoke opacity identification, and achieves the project results are as follows: An integration of the intelligent smoke detection system and an overall architecture prototype of the cloud service module has been completed, which includes an intelligent image recognition system. Regarding the opacity recognition of the system, the average opacity recognition rate for black and white smoke has stabilized between 5 % to 7 %, with the maximum error for black smoke being less than 15 %, while the maximum error for white smoke remains above 15 %. The draft of the intelligent smoke recognition method, which is ”A draft of photographic determination method of opacity of particulate pollutants discharged from chimney“ has been formulated, referencing domestic and foreign methods (such as visual smoke recognition training manuals, method 9, and ASTM 7520-2016), as well as expert suggestions. In the process of establishing the verification method for the intelligent smoke image recognition module and its software and hardware verification program, we completed the construction of the verification environment, including a high-intensity light source, a set of replica standard grayscale boards, which’s maximum error is less than 3 % for the four opacity levels corresponding to that of BSI 2742M smoke grayscale standard board. An indoor calibration document for the intelligent image smoke detection and recognition module was completed, including a draft of the indoor calibration procedure, a quality assurance plan with uncertainty assessment, and a recommended equipment list. Seven rounds of equipment verification for image capture and processing were carried out, and three pieces of equipment passed the verification. The software verification program for the intelligent smoke recognition system has been established. Field inspections were conducted at the Taoyuan and Tainan smoke detection centers, a library of 50 standard image data of black smoke and white smoke was established with an average opacity error of less than 7 %. The draft outdoor verification procedure has been completed. A filed test of outdoor verification experiment was conducted with four trials each for black smoke and white smoke. A set of black smoke simulation verification for the smoke detection system was completed, and the verification results exceeded the standards of visual smoke detection. We have completed three promotion and explanation session with 126 attendees to support the Environmental Protection Administration's promotion of the intelligent image-based smoke recognition system and other related matters. Additionally, we have assisted the Administration in conducting 15 regular project work review meetings. Complete the benefit of intelligent smoke identification task, benefit of personnel training and analysis of automated operation analysis for economic benefit analysis by this project. In terms of possible social and economic impacts, the smart image smoke detection system can be used as a factory-side self-managed monitoring system in the future to help achieve the long-term goals of carbon reduction and zero carbon, and reduce the social and economic costs caused by air pollution. In terms of economic benefit analysis, in terms of benefits of the smart image smoke detection system, the time efficiency is increased by 66 %, the error rate is reduced by 1/3, the drift of the smoke stream is uncertain, and the limitation of insufficient illumination is broken. In terms of the effectiveness of the image smoke recognition system, the influence of subjective judgments of visual inspection can be reduced. In terms of the benefit of personnel training, it can reduce the input of manpower, thereby reducing the training cost. The benefits of automated operations, fixed systems can use remote operations, reduce the cost of personnel dispatched tasks, and improve the overall monitoring efficiency.

英文關鍵字

Artificial Intelligence, Image recognition technology, Smoke opacity identification