Algorithm 1: Structure-Consistent CFAR (SC-CFAR)

Input: RD Map $X\in R^{M\times N}$; background window $B$ size $N_w$; local             neighborhood $\mathsf{\Omega }$ (e.g., 5 × 5); intensity tolerance $\delta$; base false             alarm rate $P_{fa}$; parameters $\alpha >0,\gamma \in [0,1].$ Output: Binary detection map $D\in {\{0,1\}}^{M\times N}$ (1 = target, 0 = background).

1:Initialize $D\leftarrow 0_{M\times N};$  2:for each pixel $(x,y)$ in $X$ (excluding margin required by $B$ and $\mathsf{\Omega }$) do  3:$B_{xy}\leftarrow$ set of background cells for sliding window around $(x,y)$;  4: ${\mu }_n\leftarrow {\displaystyle \frac{1}{N_w}}\sum _{(i,j)\in B_{xy}}X(i,j);$  5: ${\sigma }_n\leftarrow \sqrt{{\displaystyle \frac{1}{N_w}}\sum _{(i,j)\in B_{xy}}{(X(i,j)-{\mu }_n)}^2};$  6:${\mathsf{\Omega }}_{xy}\leftarrow$ set of coordinates in local neighborhood centered at $(x,y)$  7: $c\leftarrow 0;$  8:       for each $(i,j)\in {\mathsf{\Omega }}_{xy}$ do   9:             if $|X(i,j)-X(x,y)|<\delta$then   10: $c\leftarrow c+1$  11:          end  12:   end  13: $C(x,y)\leftarrow c/\left|{\mathsf{\Omega }}_{xy}\right|;$  14: $T(x,y)\leftarrow {\mu }_n+\alpha \cdot {\sigma }_n\cdot \sqrt{{\displaystyle \frac{\ln (1/P_{fa})}{N_w}}}\cdot (1-\gamma \cdot C(x,y));$  15:   if $X(x,y)>T(x,y)$then   16: $D(\mathrm{x},\mathrm{y})\leftarrow 1;$  17:   else  18: $D(\mathrm{x},\mathrm{y})\leftarrow 0$  19:   end  20:return $D$;