LIGHT-YOLOv11: An Efficient Small Object Detection Model for UAV Images

Object recognition in UAV images is very challenging due to the frequent appearance of numerous small and truncated entities. Object detection is an essential component of computer vision for accurate object localization and identification in complex and dynamic aerial scenarios. We introduce LIGHT-YOLOv11, a modified object detection model that integrates architectural enhancements to address the specific challenges of aerial imagery. To improve feature extraction and emphasize essential regions, the network’s initial block has been restructured into a Focus Block. Additionally, a Transformer block has been integrated into the backbone, replacing two blocks in the final architecture, to opti-mize spatial and contextual understanding. With these adjustments, detection accuracy and processing efficiency are remarkably balanced, especially for small and closely spaced objects. In-depth analyses of the VisDrone2019-DET dataset show that LIGHT-YOLOv11 outperforms other popular techniques, producing impressive outcomes across a range of situations.