Algorithm 1. Method to construct multi-constraint image sets.

Input: Training set with place labels $\left\{(x_i,y_i)\right\},i=1,2,\dots ,n$; Output: Multi-constraint image sets X; 1: Extract the feature vector for each training image; 2: for each i in {1, 2, ……, n} do 3:  Find u images from the same place with xi; 4:  Calculate the center point cp; 5:  for each j in {1, 2, ……, n} and j ≠ i do 6:    Find images from the same place & satisfy the distance relationship ${‖f(x_j^p)-c_{}^p‖}_2^2>\gamma$. Add these images into positive image set $X_{tem}^p$; 7:    Find images from different places & satisfy the distance relationship ${‖f(x_j^{})-c_{}^p‖}_2^2<{\gamma }^{\prime }$. Add these images into negative image set $X_{tem}^n$; 8:  end for 9:  if $len(X_{tem}^n)>B$ then 10:    Randomly select B images from $X_{tem}^n$; 11:  end if 12:  if $0<len(X_{tem}^n)<B$ or ($len(X_{tem}^n)=0$ and $len(X_{tem}^p)\ne 0$) then 13:    Randomly select $(B-len(X_{tem}^n))$ images from different places. 14:  end if 15:  if $len(X_{tem}^p)>A$ then 16:    Randomly select A images from $X_{tem}^p$; 17:  end if 18:  if $0<len(X_{tem}^p)<A$ or ($len(X_{tem}^p)=0$ and $len(X_{tem}^n)\ne 0$) then 19:    Randomly copy $(A-len(X_{tem}^p))$ images from the same place. 20:  end if 21:  if $X^p$ and $X^n$ exist then 22:    Add $\left(X^p,X^n\right)$ into X; 23:  end if 24: end for 25: return X