智慧影像判讀系統效能優化及應用驗證計畫-以廢電子電器及資訊物品受補貼機構為例
| 中文摘要 | 環保署對於廢棄家電與資訊設備之回收拆解處理訂定有相關作業規範，現行對於處理作業過程是否發生有違反所訂定規範之情形係透過人工抽判監看現場攝錄影像畫面進行判斷；然而以人工判讀方式稽查異常事件有許多缺點，例如長時間觀看監視器影片容易導致稽核人員疲憊，又或是判讀標準易因人而異；因此本計畫藉由開發AI智慧影像判讀系統，輔助稽核人員判讀，減少人力負擔並提升稽核認證作業品質。 本計畫針對載貨車輛於非稽核認證時段有無異常進出以及冷媒抽取作業是否發生有冷媒逸散之情形為監測標的；在冷媒逸散智慧影像判讀部份，本計畫優化去（110）年度所提出之演算法雛型，使優化後之判讀模型於冷媒散逸偵測具有更佳的辨識效能與泛用性以及更低的誤判率；本計畫以14家受補助單位進行為期一個月的冷媒散逸偵測實地導入測試，演算法之平均片長節省率與召回率分別可達92%以及95%。在大門異常進出偵測部份，本計畫與各受補貼機構進行協調，針對未符需求之鏡頭進行角度、位置之調整，並重新繪製判讀區域，針對18家受補助單位進行為期一個月之實地導入測試，經驗證所提出演算法之平均片長節省率可達到99%；經實地場域測試，系統確實可達到減少人力負擔、提升稽核認證作業效能的目的。計畫同時制訂CCTV智慧判讀查核作業配套措施，以利本計畫與現有稽核認證作業規範整併，為系統日後推展落地預做最佳準備。 | ||
|---|---|---|---|
| 中文關鍵字 | 冷媒散逸偵測、CCTV監控判讀、異常進出偵測 | ||
基本資訊
| 專案計畫編號 | EPA-111-XB05 | 經費年度 | 111 | 計畫經費 | 1980 千元 |
|---|---|---|---|---|---|
| 專案開始日期 | 2022/01/01 | 專案結束日期 | 2022/11/30 | 專案主持人 | 高立人 |
| 主辦單位 | 回收基管會 | 承辦人 | 劉妍玉 | 執行單位 | 國立台北科技大學 |
成果下載
| 類型 | 檔名 | 檔案大小 | 說明 |
|---|---|---|---|
| 期末報告 | EPA-111-XB05-公開用V1.1.pdf | 16MB |
Intelligent Image Interpretation System Performance Optimization and Application Verification Project - Taking Waste Electronic Appliances and Information Items Subsidized Organizations as an Example
| 英文摘要 | The Environmental Protection Agency has set operating rules for the recycling and dismantling of waste household appliances and waste information equipment. Currently, whether there is a violation of the set rules during the processing operation is judged by manual monitoring of the video footage. However, there are many drawbacks with the current procedure. For example, watching the monitor video for a long time can lead to fatigue of the auditors, or the interpretation standard is easy to vary from person to person. Therefore, we develop an AI-based intelligent image interpretation system in this project to improve the accuracy and efficiency of the interpretation, reduce the burden on manpower, shorten the time of manual interpretation, and improve the quality of audit certification operations. This project is aimed at monitoring whether there is abnormal entry and exit of cargo vehicles during non-audit and certificated period, and whether the refrigerant leaks during the refrigerant extraction operation. In the part of refrigerant leakage detection, we optimized the prototype algorithm proposed last year, and the optimized model has better recognition performance and generalizability, as well as lower false positive rate. Through a one-month field test at 14 subsidized institutions, the average film length saving rate of the proposed algorithm reached 92% and the recall rate can reach 95%. Besides, we coordinated with the subsidized institutions to adjust the angle and position of the cameras that did not meet the requirements and redrew the interpretation area for abnormal entry and exit of cargo vehicles detection. We also conducted a one-month field test for 18 subsidized units and verified that the average film length saving rate of the proposed algorithm can reach 99%. The system developed in this project can really achieve the purpose of reducing manpower burden and improving the efficiency of audit and certification operations. Moreover, we formulate supporting measures for the CCTV intelligent reading inspection operation, so as to facilitate the integration of this project with the existing audit certification operation standard, and to make the best preparation for the future implementation of the system. | ||
|---|---|---|---|
| 英文關鍵字 | Refrigerant Leakage Detection, CCTV Surveillance and Interpretation , Unusual Entry and Exit Detection | ||