Figure 4.

The network architecture that can apply the CAM method. The key layer is the global average pooling (GAP) layer, which establishes the mapping relationship between feature maps and classes. ${\mathrm{S}}_k\left(x\right)$ represents the output sequence of the last convolutional layer on the channel $\mathrm{k}$. ${\mathrm{f}}_k$ is the output of channel $\mathrm{k}$ after the GAP.${\omega }_k^c$ represents the weight of the features of each channel, $\mathrm{k}$, to different class, $\mathrm{c}$.