文献类型：期刊文献

英文题名：Research on Embedded AI Vision Terminals and Service Optimization Applications for Smart Libraries

作者：Guo, Yingxin[1]; Xiao, Kaile[1]; Wang, Jing[1]; Chen, Jingxia[1]

第一作者：Guo, Yingxin

机构：[1] College of Applied Science and Technology, Beijing Union University, Beijing, China

第一机构：北京联合大学应用科技学院

年份：2025

起止页码：68-73

外文期刊名：5th International Conference on Electron Devices and Applications, ICEDA 2025

收录：EI(收录号：20261520499920)

语种：英文

外文关键词：Behavioral research - Computer vision - Digital libraries - Edge detection - Embedded systems - Environmental protection - Human computer interaction - Learning systems - Three dimensional computer graphics - Three term control systems - Transfer learning

摘要：To address challenges in traditional library services such as real-time data gaps, delayed demand perception, and static environment control, this study develops an embedded AI vision terminal and a service optimization system for smart libraries. Based on NVIDIA Jetson Nano hardware and industrial-grade image sensors, the system builds a library-specific facial expression dataset (svstu2024). By integrating the lightweight YOLOv8 algorithm with the PAD 3D emotion model and applying transfer learning, channel pruning, and FP16 quantization, it achieves 89.7% facial expression recognition accuracy with 100ms inference latency on the edge device, enabling real-time emotional state perception. Building on this, a closed-loop 'perception-decision-execution' optimization system is implemented. Using PID-based control logic, it maps real-time PAD emotion metrics into precise commands for LED lighting and HVAC systems, while providing personalized resource recommendations and dynamic service adjustments. A/B testing (n=482) showed that the experimental group significantly outperformed the control group in key metrics - including dwell time, borrowing conversion rate, and service satisfaction - with environmental control adaptability reaching 92.1%. This research establishes a technical paradigm for deep edge-intelligence integration in smart libraries, enabling precise service optimization while ensuring real-time processing and privacy protection, offering a practical solution for library intelligence transformation. ? 2025 IEEE.