Enhancing object detection in remote sensing imagery through hybrid feature consistency alignment

Robust object detection is crucial for analyzing remote sensing images in Earth observation applications. However, existing methods face significant challenges due to variability in aerial imaging, such as atmospheric interference, varying observation angles, and different altitudes. These factors result in inconsistent feature representations, limiting the effectiveness of optimization strategies that primarily rely on model-level modifications. To address these issues, we propose the Hybrid Feature Consistency Alignment (HFCA) approach, which aims to enhance object detection in RSIs by combining both data- and model-level feature refinement strategies. The HFCA framework consists of three core components: (1) a Balanced Augmentation algorithm that preserves optical feature diversity while minimizing noise, (2) a Morphological Augmentation algorithm that enhances object diversity with maintained shape integrity, and (3) a Zigzag Feature Fusion network that mitigates feature inconsistencies across receptive fields in different model layers. These components are seamlessly integrated into HFCA-Net, a modified Faster R-CNN architecture. We evaluate HFCA-Net using two backbone variants—HFCA-Net-B1 and HFCA-Net-B4—against 36 state-of-the-art models from 2023 to 2026. Experimental results on four benchmark datasets show that HFCA-Net significantly improves detection accuracy, inference speed, and model efficiency by reducing the number of parameters compared to the baseline model. As a plug-and-play technique, HFCA can be seamlessly integrated into existing detection frameworks without extensive structural modifications, emphasizing its practical potential for widespread deployment in remote sensing applications.