摘要
从网络结构、图像卷积、池化运算、激活函数等方面对深度学习与卷积神经网络进行了介绍,并基于卷积神经网络与TensorFlow软件对手写数字识别进行了研究,给出了训练流程,并对试验结果进行了分析。应用所研究的方法,对手写数字的识别准确率可以达到98%,效果良好。
The deep learning and CNN were introduced from the aspects of network structure, image convolution, pooling operation and activation function. Handwritten digit recognition was studied based on CNN and TensorFlow software. The training process was given and the experimental results were analyzed. Application of the technique under study, the accuracy of handwritten digits recognition can reach up to 98~ with sound effect.
出处
《上海电气技术》
2018年第1期31-34,61,共5页
Journal of Shanghai Electric Technology
