利用大量图像数据对卷积神经网络算法进行训练，通过卷积、池化、下采样以及全连接层训练后的卷积神经网络在图像识别精度越来越高。(By using a large number of image data to train the convolutional neural network algorithm, the accuracy of the image recognition is higher and higher by convolution, pooling, down sampling