Auxiliary Pneumonia Classification Algorithm Based on Pruning Compression

Figure 3

Channel pruning structure diagram in the figure; is the output of the upper convolution, and then in the convolution layer, the weight of the BN layer is used to perform a simple sorting operation from small to large, and then the pruning rate of this layer is used to BN-weight. Pruning is performed from small to large, and the remaining feature maps are used as the input of the lower convolutional layer, and the pruning operation is continued in the layer.