基于FPGA与卷积神经网络的零件识别系统  

Part Recognition System Based on FPGA and ConvolutionalNeural Network

作　　者：赵攀峰 王一群 ZHAO Pan-feng;WANG Yi-qun(Renmin University of China,Beijing 100872,China;Beijing University of Information Technology,Beijing 100192,China)

机构地区：[1]中国人民大学,北京100872 [2]北京信息科技大学,北京100192

出　　处：《仪表技术与传感器》2023年第4期72-76,82,共6页Instrument Technique and Sensor

基　　金：科技创新2030-“新一代人工智能”重大项目(2021ZD0113603)。

摘　　要：针对工业领域零件生产线分拣系统存在识别速度慢、准确率较低的问题,设计了基于FPGA与卷积神经网络的零件识别系统。采用FPGA作为硬件平台,采集零件图像,对图像进行预处理,读取Hu不变矩作为零件形状特征,将该特征作为卷积神经网络的输入,实现不同零件类型的自动识别,完成了系统硬件和软件的设计。实验表明系统检测的平均准确率为98.24%,速度为875 ms/次,对光照和姿态有较强的鲁棒性,在生产线零件分拣系统应用中有一定的推广价值。Aiming at the problems of low recognition speed and low accuracy in the sorting system of the part production line in the industrial field,a part recognition system based on FPGA and convolutional neural network was designed.FPGA was used as the hardware platform to collect part images,preprocess the images,read Hu moment invariants as part shape features,and this feature was used as the input of convolutional neural network to realize automatic recognition of different part types.The hardware and software design of the system is completed.The experiment shows that the average accuracy of the system is 98.24%,the speed is 875 ms/time,and the system has strong robustness to light and attitude.It has certain application value to the parts sorting system of the production line.

关 键 词：FPGA 卷积神经网络 零件识别 HU不变矩 

分 类 号：TN919[电子电信—通信与信息系统]