110年度補助應回收廢棄物回收處理創新及研究發展計畫-應用AI影像辨識系統代替稽核認證之人工查驗試驗計畫-以廢電子電器為例
| 中文摘要 | 目前以人工判讀方式稽查異常事件有兩項缺點，其一需要長時間觀看監視器影片，其二容易因人為因素發生漏抓異常事件之情況，所以希望藉由本計畫開發AI智慧影像判讀系統來提高判讀準確率以及判讀效率，減少人力負擔及縮短人工判讀時間。根據上述所提出欲改善之項目，本次系統的異常事件主要針對：載貨車輛於非稽核認證時段有無異常進出的情形、物料貯存區於違規時段時間有無人、卡車以及堆高機移動之情形以及冷媒抽取作業時是否有冷媒逸散之情形。為此本計畫開發將AI智慧影像判讀系統分成大門進出口及物料貯存區AI智慧影像判讀以及冷媒AI智慧影像判讀，對上述異常事件進行處理。本系統係由中華電信千里眼平台撈取處理廠上傳之CCTV監視器畫面作為測試樣本，在系統運行過程中若有異常事件發生時，將會自動判讀並記錄異常時間與當前的監視器畫面，再將判讀結果及異常畫面彙整並儲存。未來若處理廠對於判讀結果有所異議，則稽核人員無須再觀看完整之監視器畫面，只需透過系統篩選所指定之異常事件，便可直接調閱進行複判，達到減少人力負擔及縮短人工判讀時間之目的。本系統優勢在於上述兩套系統皆能在現行處理場之硬體規格下執行，無須額外升級硬體設備即可協助稽核人員提升判讀效率。經實際驗證後，該系統平均能縮減高達97%的影片長度，大幅減少稽核人員判讀所需花費的時間，降低長時間觀看影片的疲勞，增加稽核人員判讀效率。 | ||
|---|---|---|---|
| 中文關鍵字 | 冷媒偵測、電子圍籬、智慧判讀 | ||
基本資訊
| 專案計畫編號 | EPA-110-XA03 | 經費年度 | 110 | 計畫經費 | 4380 千元 |
|---|---|---|---|---|---|
| 專案開始日期 | 2021/01/01 | 專案結束日期 | 2021/11/30 | 專案主持人 | 高立人 |
| 主辦單位 | 回收基管會 | 承辦人 | 蕭英琪 | 執行單位 | 國立臺北科技大學 |
成果下載
| 類型 | 檔名 | 檔案大小 | 說明 |
|---|---|---|---|
| 期末報告 | 應用AI影像辨識系統代替稽核認證之人工 查驗試驗計畫-以廢電子電器為例.pdf | 5MB |
A Test System Plan base on AI Image Recognition for Automatic Audit Certification-Take Waste Electronic Appliances As an Example
| 英文摘要 | At present, there are the following shortcoming in the inspection of non-conformance events by manual interpretation. It takes a long time to watch the monitor and it is easy to miss the non-conformance event due to human factors. Therefore, it is hoped that through this project, the AI intelligent image interpretation system will be developed to improve interpretation accuracy and interpretation efficiency, reduce manpower burden and shorten manual interpretation time. According to the above-mentioned items to be improved, the non-conformance events of this system are mainly aimed at whether there is a non-permitted entry and exit of truck during the non-audit certification period; whether people, trucks, and forklifts move in the material storage area during the non-permitted period, and whether the refrigerant escapes during the refrigerant extraction operation. To fulfill these requirements, this project develops an AI smart image interpretation system, which is divided into gate entrance and material storage areas AI smart image interpretation, as well as refrigerant AI smart image interpretation, to deal with the above-mentioned abnormal events. In this project, the CCTV monitor screens of the treatment plant obtained from Chunghwa Telecom’s clairvoyance (中華電信千里眼) are used as system input. When a suspected non-conformance event is detected, the time of occurrence and the current monitor screen will be automatically interpreted and recorded, and then the interpreted results and non-conformance screens are collected and stored. In the future, if the treatment plant disagrees with the interpretation results, the auditors will no longer need to watch the complete monitor screen. They only need to screen the specified non-conformance events through the system, and then they can directly access the shortened video sequence for re-judgment to reduce labor burden and reach the purpose of shortening manual interpreting time. The advantage of this system is that the above two systems can be executed under the hardware specifications of the current treatment plant, to help the auditors improve the efficiency of interpretation without the need of upgrading the hardware equipment. After verification, the proposed system can reduce the length of the video by 97% and greatly reduce the time spent in interpretation by the auditors. It can effectively improve the concentration of watching videos and increase the efficiency of interpretation by auditors. | ||
|---|---|---|---|
| 英文關鍵字 | Refrigerant Detection, Electronic Fence, Intelligent Interpretation | ||