Design of Fast Image Recognition Accelerator Based on Convolutional Neural Network

Due to the limited resources of the Internet of things terminal, the speed of image recognition is difficult to meet the application requirements. A design method of a fast image recognition accelerator based on a convolutional neural network (CNN) is proposed. A pipeline processing scheme combining software and hardware is designed. The operation strategy of the parallel image block, parallel input channel, and parallel output channel is adopted. Based on this strategy, a model of terminal resources and recognition time is established. By solving the model, the optimal number of image partition blocks and convolution parallel parameters are obtained. The experimental results show that the computational performance of the proposed accelerator is improved from 8.86 GOPs to 12.26 GOPs, which effectively improves the speed of image recognition.