Quantitative research of convolutional neural network and FPGA deployment

Convolutional neural networks have become indispensable in the field of artificial intelligence. It is very important to implement the convolutional neural networks in FPGAs especially for some specific applications. However, the characteristics of convolutional neural networks such as large model and multiple parameters pose great challenges to the deployment of that in FPGAs. Typically, the convolutional neural network model need to be quantized and compressed while deploying it in FPGAs. In this paper, we trained the weight binary network and implemented it in the FPGA platform. The HDL descriptions are mainly converted from C++ high-level language with XILINX tool. Experimental results show that the implemented scheme has low power consumption, high precision and performance.