Cal-Yolov8: an Object Detection Algorithm for Multi-Scale Uav Aerial Images in Complex Scenes

With the rapid development of UAV technology and computer vision, UAV-based remote sensing shows broad prospects in smart cities, agricultural monitoring, and emergency rescue. However, UAV images are typically characterized by complex backgrounds, large-scale variations, and densely distributed small objects, posing significant challenges to object detection. To address these challenges, we propose Cal-YOLOv8, an improved network based on YOLOv8. First, we embed a dynamic convolution module (DynamicConv) in the backbone to enhance feature extraction in complex scenes. Second, we introduce a Spatial-Channel Synergetic Attention (SCSA) mechanism to optimize multi-scale feature representation. Finally, we design a dedicated small-object detection layer to improve the detection of tiny targets. Extensive experiments on the VisDrone2019 dataset demonstrate that our method improves mAP@ 0.5 by 4.2 % with minimal parameter overhead. Notably, accuracy improves by 8.7 % and 8.2 % for pedestrians and crowd gatherings, respectively, significantly improving performance in complex scenes.