Algorithm 2.

Pseudocode of our 18-way DA on w-th raw image.

Input	Input a raw preprocessed w-th training image e(w).
Step A	We attain Yr[e(w)], r = 1, …, R1. See Equation (18). Each enhanced set comprises R2 new images. See Equation (19).
Step B	An HMI is produced as $e^{\prime }(w)={\eta }_1\left[e(w)\right]$. See Equation (20).
Step C	we obtain ${\mathrm{\text{Y}}}_r\left[e^{\prime }(w)\right],r=1,\cdots ,R_1$. See Equation (21).
Step D	e(w), e′(w), Yr[e(w)], r = 1, …, R1, and ${\mathrm{\text{Y}}}_r\left[e^{{}^{\prime }}(w)\right],r=1,\cdots ,R_1$ are combined via η2. See Equation (22).
Output	A new dataset M(w) is produced based on e(w ). The image number of M(w) is R3 = 2 × R1 × R2 + 2. See Equation (23).