Algorithm 1 GBDT training with fusion features for broiler weight estimation

INPUT:  S = { Xmn, Ym }: Training set, Xmn is fusion features with 2073 dimension, Ym           is body weight     K: The number of regression decision trees     ρ: The learning rate

OUTPUT:  FK(x): Trained gradient boosting regression tree

1. Set squared loss function: $L\left(Y_m,f(X_{mn})\right)=\frac{1}{2}{\left(Y_m,f(X_{mn})\right)}^2$ 2. Initializ $f_0(X)=\overline{Y}$, where $\overline{Y}$ is the average weight, and m represents the number of samples. 3. for each tree k in K do: 4.  Calculate the negative gradient according to the loss function.      ${\overrightarrow{Y}}_{\mathrm{m}}=-{\left[\frac{\partial L\left(Y_m,f(X_{mn})\right)}{\partial f(X_{mn})}\right]}_{f(X)=f_{k-1}(X)}=Y_m-f_{k-1}(X_{mn})$ 5.  Establish a binary regression decision tree with J leaf region $R_{jk}$. 6.  Calculate the predicted weight $C_{jk}$ of each $R_{jk}$.       $C_{jk}=\underset{c}{{\displaystyle \arg \min }}{\displaystyle \sum _{X_{mn}\in R_{jk}}L(Y_m,f_{k-1}(X_{mn})+c)=\underset{X_{mn}\in R_{jk}}{{\displaystyle avg}}{\overrightarrow{Y}}_m}$ 7.  Obtain the prediction of the kth regression decision tree and update result.        $f_k(X)=f_{k-1}(X)+\rho {\displaystyle \sum _{j=1}^J{C}_{jk}I(X\in R_{jk})}$ 8. end for