This paper addresses the challenge of unsupervised domain adaptive (UDA) image segmentation in high spatial resolution remote sensing images (RSIs), where the same category from different domains can appear significantly different, leading to inconsistent distributions and limiting UDA accuracy. To tackle this issue, the authors propose a novel Deep Covariance Alignment (DCA) model. The DCA model explicitly aligns category features to learn shared domain-invariant discriminative feature representations, enhancing the model's generalization ability. Specifically, the model employs a Category Feature Pooling (CFP) module to extract category features by combining coarse outputs and deep features. A Covariance Regularization (CR) technique is introduced to enforce intra-category features to be closer and inter-category features to be further apart. This CR method aims to regularize the correlation between different dimensions of features, making the model more robust to divergent category features with imbalanced and inconsistent distributions. The DCA model is trained using a stagewise procedure to alleviate error accumulation. Experiments on the LoveDA dataset demonstrate the superiority of the proposed DCA method over other state-of-the-art UDA segmentation methods, showing significant improvements in mIoU for both Rural-to-Urban and Urban-to-Rural scenarios.This paper addresses the challenge of unsupervised domain adaptive (UDA) image segmentation in high spatial resolution remote sensing images (RSIs), where the same category from different domains can appear significantly different, leading to inconsistent distributions and limiting UDA accuracy. To tackle this issue, the authors propose a novel Deep Covariance Alignment (DCA) model. The DCA model explicitly aligns category features to learn shared domain-invariant discriminative feature representations, enhancing the model's generalization ability. Specifically, the model employs a Category Feature Pooling (CFP) module to extract category features by combining coarse outputs and deep features. A Covariance Regularization (CR) technique is introduced to enforce intra-category features to be closer and inter-category features to be further apart. This CR method aims to regularize the correlation between different dimensions of features, making the model more robust to divergent category features with imbalanced and inconsistent distributions. The DCA model is trained using a stagewise procedure to alleviate error accumulation. Experiments on the LoveDA dataset demonstrate the superiority of the proposed DCA method over other state-of-the-art UDA segmentation methods, showing significant improvements in mIoU for both Rural-to-Urban and Urban-to-Rural scenarios.