【摘要】 高分辨率空天遥感影像的智能解译是空天信息处理领域的重要研究方向。复杂地表环境，如建筑与拆除（Construction and demolition, C&D）废弃物填埋场等，对遥感影像分割模型的鲁棒性提出了较高要求。传统卷积神经网络（Convolutional neural networks,CNNs）受限于局部感受野，难以捕获全局依赖关系；而基于Transformer的模型虽具备长距离建模能力，却容易忽略细粒度空间结构，导致异质遥感目标分割精度不足。为此，本文提出一种全局与局部协同网络（Global-local collaborative network, GLC-Net），面向空天遥感影像的智能分割任务。该模型融合了高效Transformer模块以建模全局依赖关系，并引入局部增强模块用于细节结构优化。此外，设计了多尺度空间聚合与增强模块（Multi-scale spatial aggregation and enhancement, MSAE）以强化上下文特征表征并抑制背景干扰，同时通过深层监督机制提升多层次语义学习能力。基于两组高分辨率遥感数据集（昌平与大兴）的实验结果表明，GLC-Net在平均交并比（mean intersection over union,mIoU）指标上较现有先进方法提升1.5%～3.2%，并在边界刻画与语义一致性方面表现更优。结果验证了全局-局部协同建模在空天遥感影像智能分割与环境监测中的有效性与潜力。更多还原

【Abstract】 Intelligent interpretation of high-resolution remote sensing imagery is a fundamental challenge in aerospace information processing. Complex ground environments such as construction and demolition(C&D) waste landfills exemplify the need for robust segmentation models that can handle diverse spatial and spectral patterns. Conventional convolutional neural networks(CNNs) are limited by their local receptive fields, whereas Transformer-based architectures often lose fine spatial detail, resulting in incomplete delineation of heterogeneous remote sensing targets. To address these issues, we propose a global-local collaborative network(GLC-Net), which is designed for intelligent remote sensing image segmentation. The model integrates an efficient Transformer block to capture global dependencies and a local enhancement block to refine structural details. Furthermore, a multi-scale spatial aggregation and enhancement(MSAE) module is introduced to strengthen contextual representation and suppress background noise. Deep supervision facilitates hierarchical feature learning. Experiments on two high-resolution remote sensing datasets(Changping and Daxing) demonstrate that GLC-Net surpasses state-of-the-art baselines by 1.5%—3.2% in mean intersection over union(mIoU), while achieving superior boundary precision and semantic consistency. These results confirm that global-local collaborative modeling provides an effective pathway for intelligent remote sensing image segmentation in aerospace environmental monitoring.更多还原