With the rapid development of the Internet of Things and smart devices, edge computing, as a distributed computing architecture, has been widely used in many scenarios with high real-time requirements and limited power consumption. However, traditional AI models are difficult to meet the needs of edge devices due to their huge computing requirements and high power consumption. To this end, lightweight technology for AI models has emerged, which can effectively reduce the computing burden of the model, improve the computing efficiency of edge devices and reduce power consumption. This paper studies the accelerated application of AI model lightweight technology in edge computing, proposes lightweight methods such as pruning, quantization, and knowledge distillation, and verifies the performance improvement of pruning technology on different edge devices through experiments. Experimental results show that lightweight technology can improve the inference speed on edge devices, reduce the computing time by more than 40%, and reduce power consumption by about 30%. On devices such as Jetson Nano and Raspberry Pi 4B, the lightweight model reduces power consumption by about 35% and increases in inference speed by about 50% while ensuring more than 95% accuracy. This study provides an effective solution for AI applications in edge computing and demonstrates the potential for widespread application of lightweight technology in future smart devices.