CH-YOLO-Lite: a lightweight object detection model with context-aware progressive aggregation and hierarchical feature aggregation for aerial imagery

Object detection in aerial images poses significant challenges due to the high proportion of small objects, drastic scale variations, and resource constraints. We propose a lightweight object detection model, CH-YOLO-Lite, based on context-aware progressive aggregation and hierarchical feature augmentation. First, we designed a context-aware progressive aggregator (CPA) to integrate multi-scale contextual information through a progressive fusion strategy, enhancing the model’s perception of its environment and improving small object detection in complex scenes. Second, we designed a hierarchical feature aggregation (HFA) module and an improved adaptive spatial feature fusion (iASFF) method to construct the feature augmentation neck, which significantly enhances the model’s multi-scale detection capability in scenarios dominated by small objects. The HFA introduces more low-level detail features into the feature pyramid through cross-level branches, strengthening the model’s feature representation for small objects. Meanwhile, iASFF adaptively learns the spatial fusion weights of feature maps at different scales, effectively solving the issue of scale inconsistency during feature fusion, further improving detection accuracy. Finally, to further simplify the model, channel pruning is used, which significantly reduces the parameters of the model while maintaining the detection performance. Experiments on the test set of the VisDrone dataset show that CH-YOLO-Lite significantly improves mAP0.5 by 7.2% and reduces parameters by 67% compared with the baseline model YOLOv7-Tiny. The mAP0.5 of CH-YOLO-Lite is also improved by 4.9% on NWPU VHR-10 datasets, which validates the generalization of our proposed algorithm.