Algorithm 2: Lightweight Multi-Scale Feature Fusion (LMSFF) Module

Input:  $\mathrm{Feature} \mathrm{map} \mathrm{tensor} F\in R^{C\times H\times W}$

Output: $\mathrm{Multi}-\mathrm{scale} \mathrm{fused} \mathrm{feature} \mathrm{map} F_{fused}\in R^{C^{’}\times H\times W}$

Step 1: Multi-Scale Convolutional Feature Extraction:

$F_1=DSCon{v}_{3\times 3,r=2}\left(F\right)$ $F_2=DSCon{v}_{5\times 5,r=1}\left(F\right)$ $F_3=DSCon{v}_{5\times 5,r=2}\left(F\right)$ $F_4=DSCon{v}_{7\times 7,r=3}\left(F\right)$

Step 2: BN + ReLU:

$F_i\leftarrow \mathrm{Re}LU(BN(F_i)), For i\in \left\{1,2,3,4\right\}$

Step 3: Feature Fusion:

$F_{concat}\leftarrow Concat(F_1,F_2,F_3,F_4)$