Object detection is a fundamental capability that enables drones to perform various tasks. However, achieving a suitable equilibrium between performance, efficiency, and lightweight design continues to be a significant challenge for current algorithms. To address this issue, we propose an enhanced small object detection transformer model called ESO-DETR. First, we present a gated single-head attention backbone block, known as the GSHA block, which enhances the extraction of local details. Besides, ESO-DETR utilizes the multiscale multihead self-attention mechanism (MMSA) to efficiently manage complex features within its backbone network. We also introduce a novel and efficient feature fusion pyramid network for enhanced small object detection, termed ESO-FPN. This network integrates large convolutional kernels with dual-domain attention mechanisms. Lastly, we introduce the EMASlideVariFocal loss (ESVF Loss), which dynamically adjusts the weights to improve the model’s focus on more challenging samples. In comparison with the baseline model, ESO-DETR demonstrates enhancements of 3.9% and 4.0% in the $mA{P}_{50}$ metric on the VisDrone and HIT-UAV datasets, respectively, while also reducing parameters by 25%. These results highlight the capability of ESO-DETR to improve detection accuracy while maintaining a lightweight and efficient structure.

Cross-view geo-localization (CVGL) aims to determine the capture location of street-view images by matching them with corresponding 2D maps, such as satellite imagery. While recent bird’s eye view (BEV)-based methods have advanced this task by addressing viewpoint and appearance differences, the existing approaches typically rely solely on either OpenStreetMap (OSM) data or satellite imagery, limiting localization robustness due to single-modality constraints. This paper presents a novel CVGL method that fuses OSM data with satellite imagery, leveraging their complementary strengths to enhance localization robustness. We integrate the semantic richness and structural information from OSM with the high-resolution visual details of satellite imagery, creating a unified 2D geospatial representation. Additionally, we employ a transformer-based BEV perception module that utilizes attention mechanisms to construct fine-grained BEV features from street-view images for matching with fused map features. Compared to state-of-the-art methods that utilize only OSM data, our approach achieves substantial improvements, with 12.05% and 12.06% recall enhancements on the KITTI benchmark for lateral and longitudinal localization within a 1-m error, respectively.