nuclear" it is a small matrix. The convolution kernel

performs the element of each element of the convolution core in the input data through the sliding window. Add the output results of all the product product to obtain the convolution operation. Different convolution kernels can capture different features such as edges, textures, shapes, etc. How can the B -side product manager quickly get started with the new business to enter the workflow more smoothly? View details> In the training process, the convolutional neural network will learn the optimal convolution nuclear parameters to make the network better extract the input input The local characteristics of the data. 

Because the convolution nuclear is relatively small, even if we finish the convolutiona Rich People Phone Number List l processing image, it is still large. Pond or average pooling to reduce the size of the feature map. For example, the original picture is*We use a*sampling window to perform the original picture down the original picture. Finally, the original picture can be sampled into a small picture. This operation is actually to reduce the pixels of pictures from high -definition pictures to blur pictures. This retains the main feature information and effectively filter out the interference information in the picture to reduce the amount of data transmitted to the next stage. The pooling layer can more effectively reduce the data dimension that greatly reduce parameters and calculations while avoiding the occurrence of overfitting phenomena. 

The features extracted from the convolutional layer and the pooling layer are input to the full connection layer for training and learning. Due to the great reduction of parameter quantity, the discarded interference features have greatly reduced the training cost and complexity. 4. Application scenario CNN has a wide range of applications in the field of computer vision. The following is some CNN's application scenario image classification can divide images into different categories such as identifying handwriting numbers, identification animals, and identification objects. Target detection can position and identify