Figure 3.

The overall 3D convolution operation. The input data dimension is $W\times H\times B\times C_1$, where $\mathrm{B}$ is the band number and $C_1$ is the channel number; the 3D convolution kernel size is $\mathrm{k}\times \mathrm{k}\times \mathrm{k}$ and the last $\mathrm{k}$ denotes the coverage of convolution kernel over spectral dimension in each convolution operation; if padding is not used and the stride is 1, then the output data dimension of single kernel is $\left(W-k+1\right)\left(H-k+1\right)\left(B-k+1\right)$ and the final output data generated by $\mathrm{p}$ kernels is a 4D tensor.