Convolutional neural network and its feature graph visualization

Reference link: https://www.jianshu.com/p/362b637e2242

Reference link: https://blog.csdn.net/dcrmg/article/details/81255498/

Reference link: https://zhuanlan.zhihu.com/p/59917842

Reference link: https://www.jb51.net/article/171016.htm

1. Convolution and feature map

A simple deep network model, mainly:

input -> conv -> feature_map -> maxpooling -> flatten -> fully connected -> output

CNN: Convolutional layer, pooling layer, connection layer

Convolution layer

       The convolution layer consists of a set of convolution units (also called "convolution kernels"). These convolution units can be understood as filters, and each filter extracts a specific feature. The input image is 32 * 32 * 3, 3 is its depth (that is, R, G, B three-color channel), the convolution layer is a 5 * 5 * 3 filter, note here: the depth of the filter must be the same as the input image The same depth. A 28 * 28 * 1 feature map can be obtained by convolving a filter with the input image. If two filters are used, two feature maps are obtained.

For example, input, filter, outpu