基于LeNet神经网络的手写数字识别技术研究

Research on Scripted Number Recognition Technology Based on LeNet Neural Network

以机器学习的神经网络算法进行数字的识别,研究卷积神经网络,分析卷积神经网络输入层、卷积层、激励层、池化层和全连接层的原理和作用,对现有卷积神经网络LeNet-5模型进行简要分析。对图像数字的大小格式进行灰度处理,合理设计LeNet-5层级结构,进行摄像头图片的特征提取,并对参数进行设置,对向前向后传播方式进行深入了解,并对梯度下降方式进行选取以及分类层的设计;利用OpenMV摄像头模块,基于Phython语言进行了程序设计,对数字识别的试验结果进行分析比对,评估本系统的识别准确度和辨识度等性能指标。

Based on the neural network algorithm of machine learning,this paper analyzed the digital identification,convolution neural network including the principle and function of convolution neural network input layer,convolution layer,excitation layer,pool layer and all connection layer,and briefly analyzed the existing convolution neural network LeNet-5 model,properly handled gray level of size and format of image numbers,properly designed LeNet-5 structure,did picture feature extraction of cameras,set parameters,and look into propagation mode of backwards and forwards,selected gradient descent way and design the category layer.Through OpenMV camera module,and based on Phython,we did programming,comparatively analyzed the result of digital identification,evaluated performance index like recognition accuracy,inimitable symbol,and so on.