摘要
阐述卷积神经网络因其在图像识别领域的优良表现,它被广泛应用于车牌识别、自动银行支票处理等领域,但卷积神经网络的参数量大、运算量大,且随着网络层数的加深运算量成倍增加,使其运算速度较慢,影响了模型落地及应用部署。模型轻量化部署及加速研究显得尤为重要,模型压缩、算子优化、硬件加速成为发展方向。以手写体数字识别的卷积神经网络为例,探讨卷积神经的结构特点、实施加速的方向和基于FPGA的硬件加速应用。
Because of its excellent performance in the field of image recognition, convolutional neural network is widely used in license plate recognition, automatic bank check processing and other fields. However, convolutional neural network has a large amount of parameters and computation, and with the deepening of the number of network layers, the computation increases exponentially, making its operation speed slow, affecting the model landing and application deployment. Model lightweight deployment and acceleration research is particularly important. Model compression, operator optimization and hardware acceleration have become the development direction. Taking convolutional neural network for handwritten digit recognition as an example, it discusses the structural characteristics of convolutional neural network, the direction of implementation acceleration and the hardware acceleration application based on FPGA.
作者
张玉娇
ZHANG Yujiao(School of Electronic Information Engineering,Hebi Vocational and Technical College,Henan 458030,China)
出处
《集成电路应用》
2022年第6期63-65,共3页
Application of IC
基金
中国高校产学研创新基金-异构智能计算项目课题(2020HYA01001)。
关键词
卷积神经网络
手写体数字
FPGA
硬件加速
convolutional neural network
handwritten digits
FPGA
hardware acceleration