MobileNet Compression and Edge Computing Strategy for Low-Latency Monitoring

Abstract

This paper designs and implements a lightweight real-time monitoring system for IoT wearable devices by integrating MobileNet with edge computing. The system aims to improve sensing efficiency and response speed across multiple usage scenarios. It adopts the MobileNet model, built with depthwise separable convolutions, as the base network architecture. A dynamic channel pruning strategy is introduced to reduce model size and computational complexity, making it suitable for resource-constrained end devices. At the same time, the system incorporates an entropy-based edge-assisted inference mechanism to enable intelligent task allocation between local and edge nodes. This approach significantly improves overall energy efficiency and real-time processing capability. Experimental evaluations were conducted across several typical wearable scenarios. The results show that the proposed system maintains high accuracy while achieving effective latency control and energy efficiency. It successfully meets the demands of real-time monitoring tasks for low power consumption and fast system response.