Fig. 1. The framework of aerial object recognition model

Fig. 2. Framework of domain invariant deep feature extraction module

Fig. 3. Histograms of the deep features in source and target domains，（a）the sample in the source domain，（b）the histogram of the deep features in the source domain，（c）the sample in the target domain，（d）the histogram of the deep features in the target domain

Fig. 4. Bar charts of shallow features，（a）the sample，（b）SIFT，（c）LBP，（d）Harris，（e）HOG，（f）grayscale histogram

Fig. 5. Framework of GCN-based feature fusion module

Fig. 6. Histograms of the fused features in source and target domains，（a）the sample in source domain，（b）the histogram of the fused features in source domain，（c）the sample in target domain，（d）the histogram of the fused features in target domain

Fig. 7. Samples of typical aerial object images

Fig. 8. Experimental results on the sensitivity of hgperparameter λ_wd

Fig. 9. Ressults of feature visualization for D-SLGM algorithm，（a） visualization of the deep features extracted by baseline CNN，（b）visualization of the domain invariant deep features extracted by D-SLGM，（c）visualization of the final features extracted by D-SLGM