Algorithm 1. Approach of pruning channel and layer in YOLOv4

Input: $N$ layers and $M$ shortcut layers of YOLOv4, channel pruning rate $\alpha$ and layer pruning $t$

Output: The remaining layers after pruning Sparsity training $N$ layers and $M$ shortcut layers and get ${\gamma }_k^i$ of the $k$-$th$ channel of $i$-$th$ layer Sort ${\gamma }_k^i$ of $N$ layers and $M$ shortcut layers from small to large and then get array $W$ Threshold $t=W\left[int\left(\alpha ·len\left(W\right)\right)\right]$ for $i=1$ to $N$ do if ${\gamma }_{k=\left\{1,2,\dots \right\}}^i<t$ Remove these channels $k=\left\{1,2,\dots \right\}$ of $i$-$th$ layer end for $A~F$ is shown as Figure 3. $A^i$ is the $A$ layer of $i$-$th$ shortcut layer structure. for $i=1$ to $M$ do   if ${\gamma }_{k=\left\{1,2,\dots \right\}}^i<t$ Mark $k=\left\{1,2,\dots \right\}$ which is the index of channels of $A^i$ layer   for $j=A^i$ to [$A^i$,$C^i$,$F^i$] do   Remove $k=\left\{1,2,\dots \right\}$ channels of $j$ layer  end for end for Evalute the mean value $m_{s=\left\{1,2,\dots ,23\right\}}$ of ${\gamma }_{k=\left\{1,2,\dots \right\}}^i$ for each $M$ shortcut layers, then sort $m$ from small to large for $i=1$ to $t$ do Get the index of shortcut layer $s=m_s\left[i\right]$ Remove $C^s,D^s$ and $E^s$ layers end for