This project continues the achievements of 2022 innovation project, continuously optimizing its smart recognition technology and conducting on-site field certification. It establishes an Image Smart Recognition Verification Information Management Testing Platform, which automatically downloads CCTV image data to the server for image recognition using the smart interpretation system. It automatically clips abnormal images for audit personnel to manually reevaluate. The system performs initial screening to select potentially abnormal events and clips event videos, which are then reevaluated by audit personnel, rather than entirely replacing manual interpretation. In the current year, a five-month evaluation of image interpretation verification has been conducted for 18 subsidized institutions regarding "refrigerant leaks" and "abnormal vehicle exits at main gates." The evaluation is conducted using both smart recognition interpretation and manual interpretation in parallel to verify performance. The detection rate (recall rate) through the confusion matrix has averaged 95.15%. The average savings in video length are 98.97% for main gate cameras and 94.33% for refrigerant cameras. The results show an improvement in interpretation quality and a reduction in interpretation time, indicating that automated operations assist audit certification groups in reevaluation, enhancing interpretation efficiency.

Video recognition、Information management platform、Electronic Waste Recycling