皮带秤用神经网络基本元件的FPGA实现方法研究

Research on implementation method of basic components of neural network for belt weigher by FPGA

本文基于应用于皮带秤精度优化的神经网络算法,研究了神经网络基本元件的FPGA实现方法。制定了22位定点数的数据表示方法,并基于此方法设计实现了加法器和乘法器等计算元件;提出了一种神经网络激活函数(sigmoid函数)的硬件实现方法,分别将matlab仿真、硬件实现的仿真输出与被逼近函数进行了比较,验证了神经网络基本元件的FPGA实现方法的效果。

Based on the neural network algorithm which is applied to the optimization of belt weigher accuracy, this paper studies the FPGA implementation method of the basic components of neural network. A data representation method for 22 bit fixed-point numbers is developed, and computational components such as adders and multipliers are designed and implemented based on this method. A hardware implementation method of neural network activation function(sigmoid function) is proposed.The matlab simulation output and the hardware-implemented simulation output are compared with the approximated function, and the effect of the FPGA implementation method of the basic components of the neural network is verified.