CNN手写数字识别系统的ZYNQ实现
CNN Handwritten Digit Recognition System ZYNQ Implementation

针对卷积神经网络手写数字识别系统在软件平台上运行速度慢，功耗高的问题，同时更好地满足便携性的需求，利用FPGA并行计算的优势，在ZYNQ平台的逻辑端对卷积神经网络中的卷积层和池化层进行硬件加速，使用MT9V034摄像头采集图像通过屏幕实时显示识别后的数字。与软件平台相比较，处理一帧图像的速度提高了至少178倍，综合后的总片上功耗为1.969 W，逻辑资源的使用量为9.260 K，实现了对手写数字的低功耗快速识别。
In order to solve the problems of slow running speed and high power consumption of the convolutional neural network handwritten digit recognition system on the software platform, and to better meet the needs of portability, this paper uses the advantages of FPGA parallel computing to accelerate the convolutional layer and pooling layer in the convolutional neural network on the logical end of ZYNQ platform. MT9V034 camera is used to collect images and display the recognized numbers in real time through the screen. Compared with the software platform, the speed of processing a frame image is increased by at least 178 times, the total on-chip power consumption is 1.969 W, and the usage of logical resources is 9.260 K, realizing the low power consumption and fast recognition of hand written digits.