Context-aware multi-label classification: Synergistic effects of masked attention and graph neural networks

We propose the MLC-GCN for remote sensing image classification. This model utilizes attention mechanisms to enhance the classification of multi-label remote sensing images by effectively capturing complex spatial and contextual features. The network integrates several key modules: feature extraction using ResNet-50, context processing, category feature extraction and fusion, label semantic mining, and a dual graph network. Experiments on three datasets — AID, UCM, and MLRSNet — show that MLC-GCN outperforms existing models. On the AID dataset, it achieves a 5.42% improvement in mAP and a 1.43% increase in accuracy. On the UCM dataset, MLC-GCN improves mAP by 0.62% and Precision by 6.45%. On the MLRSNet dataset, the network ranks in the top three for several performance metrics, including mAP, Micro-Precision, and Weight-Precision. The model’s ability to process complex remote sensing images is validated through extensive experiments and ablation studies, highlighting its robustness and efficiency.