Acceleration and optimization of artificial intelligence CNN image recognition based on FPGA

Abstract

Before CNN, image recognition methods mainly relied on artificial design features, which can only represent the medium and low-level information in the image, and it is difficult to extract the deep-level information of the image. CNN simulates the human brain by establishing a deep neural network to analyze, learn and interpret data. It has strong expression ability and generalization ability, and can better represent the deep-seated information of images. At present, CNN has been widely used in many scenes, including image classification, speech recognition, video analysis, document analysis and so on. Most of the existing artificial intelligence image recognition systems are implemented by software and often accelerated by GPU. However, GPU has high power consumption and is not suitable for CNN reasoning, so it can not make use of the parallel computing power of CNN. To solve the above problems, this paper studies the application method of CNN image recognition acceleration and Optimization Based on FPGA, designs and optimizes the CNN forward model by using Intel FPGA board. The experimental results show that FPGA has low power consumption, the power consumption of CPU is 2.1 times, while the power consumption of GPU is 6.5 times. The average recognition time of the algorithm is 50% shorter than that of Lenet-5,AlexNet and VGG16, and the recognition time of 10000 sample pictures is 165μs. Compared with 426.6μs required by DSP chip system. Compared with the methods proposed in the literature in related fields in recent years, the proposed method has higher throughput and computational performance.