As a distributed computing paradigm, edge computing (EC) technology has been widely used in intelligent terminal applications in recent years. In order to improve the performance of these applications, this paper deeply studies two main performance improvement methods: hardware acceleration and deep learning model lightweight. By comprehensively using the hardware acceleration technology of Graphics Processing Unit (GPU) and Field Programmable Gate Array (FPGA), we have effectively improved the speed of computing tasks, and at the same time, we have successfully reduced the complexity of the model by adopting the lightweight design based on MobileNet. The experimental results show that these two methods show excellent adaptability and flexibility in different scenarios. In terms of hardware acceleration, the integration of GPU and FPGA enables intelligent terminal devices to handle computation-intensive tasks more efficiently. The implementation of hybrid acceleration scheme effectively improves the parallelism and cooperative working ability of computing tasks and further optimizes the overall performance. At the same time, the lightweight of deep learning model improves the reasoning speed and power consumption efficiency of the model in resource-constrained environment by reducing the number of parameters and computational complexity of the model. This research provides useful reference and guidance for promoting the further development of EC technology.