This paper proposes a mini-batch graph convolutional network (miniGCN) for hyperspectral (HS) image classification, which addresses the computational inefficiency of traditional graph convolutional networks (GCNs) in large-scale remote sensing tasks. The authors compare CNNs and GCNs in terms of their ability to extract spatial-spectral features from HS images. They introduce miniGCN, which allows for mini-batch training and can infer out-of-sample data without retraining. Additionally, they explore three fusion strategies (additive, element-wise multiplicative, and concatenation) to combine features extracted from CNNs and miniGCNs, aiming to improve classification performance. Extensive experiments on three HS datasets demonstrate that miniGCN outperforms traditional GCNs and that the fusion strategies significantly enhance classification accuracy compared to single models. The proposed method is effective in handling the challenges of HS image classification, including spectral mixing and noise, and provides a flexible and efficient approach for large-scale HS image analysis. The results show that miniGCN, combined with fusion strategies, achieves superior performance in HS image classification.This paper proposes a mini-batch graph convolutional network (miniGCN) for hyperspectral (HS) image classification, which addresses the computational inefficiency of traditional graph convolutional networks (GCNs) in large-scale remote sensing tasks. The authors compare CNNs and GCNs in terms of their ability to extract spatial-spectral features from HS images. They introduce miniGCN, which allows for mini-batch training and can infer out-of-sample data without retraining. Additionally, they explore three fusion strategies (additive, element-wise multiplicative, and concatenation) to combine features extracted from CNNs and miniGCNs, aiming to improve classification performance. Extensive experiments on three HS datasets demonstrate that miniGCN outperforms traditional GCNs and that the fusion strategies significantly enhance classification accuracy compared to single models. The proposed method is effective in handling the challenges of HS image classification, including spectral mixing and noise, and provides a flexible and efficient approach for large-scale HS image analysis. The results show that miniGCN, combined with fusion strategies, achieves superior performance in HS image classification.