Table 1 Notation description.

D

Collection of multi-source datasets

V

Number of information sources

\({L}^{v}\)

Labeled dataset of the v-th information source

\({U}^{v}\)

Unlabeled dataset of the v-th information source

\({x}_{i}^{v}\)

The i-th sample of the v-th information source

\({y}_{i}\)

Label of the i-th sample in each source

k

Total number of labeled samples in each source

m

Total number of unlabeled samples in each source

C

Total number of categories in the dataset

\(C\left( I\right)\)

Curvelet pooling layer

T

Confidence threshold for pseudo-labeling

\(\lambda\)

Momentum decay coefficient of EMA

\(\mu\)

The ratio of unlabeled to labeled data batch sizes

\({L}_{s}\)

Supervised learning loss

\({L}_{c}\)

Multi-source feature interaction constraint loss

\({L}_{u}\)

Unsupervised learning loss

\({L}_{f}\)

Self-adaptive class fairness regularization loss

\({{w}}_{c}\)

Weight of multi-source feature interaction constraint loss

\({{w}}_{u}\)

Weight of unsupervised learning loss

\({{w}}_{f}\)

Weight of self-adaptive class fairness regularization loss