Damage Detection for Conveyor Belt Surface Based on Conditional Cycle Generative Adversarial Network

. 2022 May 3;22(9):3485. doi: 10.3390/s22093485

Algorithm 1. MCC-CycleGAN training process. The pseudocode of the proposed network training process.
1:	Input: $hyperparameters (batch size k$ $, epochs e$ $, times of generator training n, learning rate r$ ), location of dataset A and B
2:	$Establish and initialize models : G_{A}$ $, G_{B}$ $, D_{A}$ $, D_{B}$ $and C$ , setup optimizer: Adam,
2:	$Load training dataset A$ $and B$ $, samples a \in A and samples b \in B$
3:	For $e p o c h = 1$ to $e$ do
4:	For $t = 1$ to $n$ do
5:	Train $G_{A} and$ $G_{B} :$ $freeze parameters of D_{A} and$ $D_{B}$ , generate fake images
	$i m g_g_{A} = G_{A} (b)$ $and i m g_g_{B} = G_{B} (a)$ $, compute l o s s_{G A}$ $, l o s s_{G B}$ based
	$on Equation (2), and l o s s_{c y c l e}$ based on Equation (3), then update parameters of
	$model G_{A} and$ $G_{B}$
6:	end for
7:	Train $D_{A} and$ $D_{B} :$ $freeze parameters of G_{A} and$ $G_{B}$ , generate fake images
	$i m g_g_{A} = G_{A} (b)$ $and i m g_g_{B} = G_{B} (a)$ $, compute l o s s_{G A}$ $, l o s s_{G B}$ based on
	$Equation (2), then update parameters of model D_{A} and$ $D_{B}$
8:	Train $C :$ $compute the input fusion image i m g_{m i x}$ based on Equation (8), feed
	$i m g_{m i x}$ $into C$ $, compute l o s s_{c r i t i c}$ based on Equation (4), then update parameters
	$of model C$
9:	end for