说明：  卷积神经网络（ＣＮＮ）是一种深度学习技术，其目标是构建一个接近人类的智能来执行任何知识的任务．使用基于交叉熵损失函数的模拟退火算法对ｄｒｏｐｏｕｔ正则化，即有些单元值被抑制的ＣＮＮ进行优化训练，而在测试阶段所使用的模型平均方法则同时考虑了保留概率和池化区域内单元值所占概率．在ＭＮＩＳＴ手写数据库和ＣＭＵ－ＰＩＥ的部分图像库的实验结果表明，在同一ＣＮＮ结构和相同迭代次数时，此方法均优于其他方法，不仅能够 得到更好的识别率而且更能防止过拟合的问题．
(Convolutional Neural Network (CNN) is a deep learning technology whose goal is to build an intelligence close to humans to perform any knowledge task. The simulated annealing algorithm based on the cross-entropy loss function is used to regularize droopout, which means that some CNNs whose cell values are suppressed are optimized for training. The model averaging method used in the testing phase takes into account both the retention probability and the cell value in the pooling area. Occupied probability.)