The green matrix is the l-1 neuron,

The green matrix is the l-1 neuron, the red is the l-layer, the yellow is the l-layer convolution nucleus, and the convolution nucleus of the l-1 layer is omitted from the figure. The elements of each matrix can be seen as a neuron, so, like the BP network, a local gradient can be defined for each neuron, and in order to see clearly, the elements of the matrix can be replaced by the transfer of values in the neural network as the transfer of derivatives.