Small object detection remains a challenging task due to limited pixel resolution, complex backgrounds, and high sensitivity to bounding box variations in aerial images. Although Detection Transformer (DETR)-based methods have made progress, they still face significant limitations in small object detection, primarily due to their reliance on global features, which fail to capture fine-grained details and are sensitive to background noise and bounding box variations. This study proposes Progressive Fusion (PF)-DETR, a model specifically designed to refine small object detection through progressive feature fusion techniques. Central to our approach is the Cross-Scale Feature Fusion with S2 (S2-CCFF) module, which integrates multi-level features with an S2 layer to preserve small object details. Coupled with SPace-to-Depth convolution (SPDConv) downsampling, this module reduces computational cost while maintaining critical information. Additionally, Cross Stage Partial Omni-Kernel Fusion (CSPOK-Fusion) Module achieves progressive fusion by gradually integrating multi-scale features from local, large, and global branches through successive convolutional layers, effectively refining the feature representation at each stage, mitigating background interference and occlusion effects to enhance cross-scale spatial representation. We further introduce a Parallelized Patch-Aware (PPA) attention module in the Backbone network to prioritize small object features, significantly addressing information loss. Finally, Normalized Wasserstein Distance (NWD) loss function is incorporated to heighten robustness against minor localization errors by aligning bounding box positioning and shape, thus boosting detection accuracy. Experimental results on the VisDrone and NWPU VHR-10 datasets revealed that PF-DETR surpasses existing state-of-the-art methods, establishing its effectiveness and adaptability in complex aerial detection tasks.
Liu et al. (Mon,) studied this question.