According to the previous convolution process, we can achieve the role of feature extraction. Basically judged who is who D. C Under can do a further pooling. Database connection pooling remember it? Put a lot of database connection in a pool, pick one to use when you want use. Get so much done here convolution data, like a pond, pool for this data, we can continue to do a variety of operations, such as maximum or average pooled pooling. Pooling removed name suggests is the maximum from the maximum pond. Note maximum value, whichever is greater than the whole, but the neighborhood maximum value. The result is shown below, the maximum value of the first region (only four numbers) is 5, the maximum of the second region is 1. And so on.
It has experienced more than two steps after the convolution and pooling the results we get, really good enough. First, after convolution, that is, feature extraction, our success has been the result of 5. The higher the value, the more prominent features will be described, the determination to increase the final accuracy of the results. The second step of the pool, but also the characteristics of the part is not discarded, played to dry results, but also for our future. . . . . . . . . . . . . . . . . .
Reprinted from the original article: https://blog.csdn.net/qq_44596980/article/details/100620372