Satellite spatial object detection is crucial for urban planning, environmental monitoring, and catastrophe management. The DETR-DC5 (Data Efficient Transformer with DC5 Backbone) is a cutting-edge object detection model. This study introduces DETR-DC5 framework-based satellite spatial object detection improvements. DETR-DC5's strengths are used to improve feature extraction and post-processing. First, a specific feature augmentation technique extracts richer and more discriminative spatial features to let the model capture subtle object attributes. Second, a semi-supervised learning training technique is used to address the lack of labeled satellite imagery data and improve the model's generalization. A new satellite imagery post-processing refining method uses non-maximum suppression methods. This refinement stage reduces false positives and improves localization, especially in densely populated metropolitan regions and complicated natural surroundings. On benchmark satellite imagery datasets, the suggested methods outperform baseline DETR-DC5 models in spatial object detection. Experimental findings indicate significant improvements in precision, recall, and F1-score, proving the upgrades work. This study uses the DETR-DC5 architecture, targeted feature augmentation, semi-supervised training, and refined post-processing algorithms to recognize spatial objects in satellite data. These techniques have the potential to dramatically improve satellite-based object detection, which has implications for urban planning, environmental monitoring, and disaster response. © 2024 Elsevier B.V., All rights reserved.