【作者】 丁娇； 丛爽； 林文瑞； 张飞； 王大欣；

【Author】 Ding Jiao;Cong Shuang;Lin Wenrui;Zhang Fei;Wang Daxin;Department of Automation,University of Science and Technology of China;

【机构】 中国科学技术大学自动化系；

【摘要】 LeNet网络是最早成功应用于图像识别的卷积神经网络。本文在MATLAB环境下,利用深度卷积神经网络工具箱MatConvNet,设计出一个标准的LeNet网络,并进行训练和测试,实现0-9十个手写数字的识别。实验结果表明,只训练1次的分类器也可达到90%以上的识别准确率,生成的20个不同训练次数的分类器的性能随着分类器的训练次数的增加是有一个逐步提高的正相关关系。更多还原

【Abstract】 LeNet network was the first convolutional neural network successfully applied to image recognition.A standard LeNet network is designed by using the deep convolutional neural network toolbox MatConvNet in the MATLAB environment in this paper.The training and testing of the LeNet network are implemented to realize the recognition of 0-9 ten handwritten digits.The experimental results show that a classifier trained only once can achieve a recognition accuracy of more than 90%,and the performance of the 20 generated classifiers with different times trains have a gradually increasing positive correlation as the number of classifier training increases.更多还原