Abstract:In recent years, thanks to the enhancement of computing power of computers and the vast amount of data generated by the internet, Deep Learning(DL) technology has achieved rapid development. Among them, the most notable Convolutional Neural Networks(CNN) have successfully been commercialized in fields such as image recognition,object detection, and natural language processing. However, as the network layers become deeper, the demand for computing power and memory has risen sharply. How to accelerate convolutional neural networks and deploy them on hardware accelerators has gradually become a hot topic in academic research. Starting from the advantages of developing neural networks with Field-Programmable Gate Arrays(FPGA), various development methods of FPGA are introduced, various optimization strategies for deploying and accelerating convolutional neural networks are discussed in detail, and the performance of FPGA convolutional neural network accelerators using different optimization strategies is presented. Finally, the future development direction of FPGA convolutional neural network accelerators is expected.