题 目： 基于 DL4J 库 CNN 手写识别 原理研究与实现 摘 要卷积神经网络作为一种深度单方向传播的可用一个有向无环图表示的网络，其在计算机对图像的处理中应用效果非常好。LeNet-5 模型是一个专门为手写数字识别而设计的最经典的卷积神经网络模型。借助 MNIST 字符库数据集，优化卷积层样本的训练模式，即将每批固定输入样本和固定迭代次数的原始训练模式优化为每批不同输入样本和不同迭代次数的混合训练模式。优化后的训练方法可以提高识别速度，减少预处理的工作量。实验结果表明，优化后的混合样本输入法在保证样本训练时间相等的前提下，可以获得较高的识别率。关键词：手写数字识别；MNIST 字符集；卷积神经网络；Lenet-5 模型ABSTRACTAs a kind of deep unidirectional propagation network, convolutional neural network can be represented by a directed acyclic graph, which has been well applied in image processing by computer. Lenet-5 model is the most classical convolutional neural network model specially designed for handwritten digit recognition. With the help of MNIST character library data set, the training mode of convolution layer samples is optimized, that is, the original training mode of each batch of fixed input samples and fixed iterations is optimized to the mixed training mode of each batch of different input samples and different iterations. The optimized training method can improve the recognition speed and reduce the workload of preprocessing. The experimental results show that the optimized mixed sample input method can achieve high recognition rate on the premise of ensuring the same training time of samples.Keywords:Handwritten digit recognition; MNIST character set; convolutional neural network; lenet-5 model目 录第一章 引言.............................................................................................................................51.1 选题依据和研究意义...................................................................
