Small targets from UAV perspective provide limited feature information and are susceptible to external conditions such as weather, lighting, and background noise. Thus, small object detection from the UAV perspective remains a formidable challenge. In this research, a novel object detection method named Flight-YOLO, improved from YOLOv8, is proposed to address challenges related to object detection in UAV images. Firstly, to enhance the ability to extract features across different scales while maintaining computational efficiency, a novel multi-scale feature extraction (MFE) module, which is designed using a newly developed attention mechanism and partial convolution, and LSKA is integrated into the SPFF. Secondly, a triple feature fusion (TFF) module is developed to concatenate feature maps from three distinct layers, significantly increasing the feature fusion capacity. Thirdly, the initial head is replaced with DyHeadv3, which employs DCNv3, further improving the capacity to identify tiny targets. The experimental results demonstrate Flight-YOLO greatly enhances the performance metrics P, R and \(mAP\) by 9.8%, 15.2%, and 16.9%, respectively, compared to YOLOv8s. The proposed method performs superiorly to other baseline methods, making it highly suitable for detection tasks, especially from the UAV perspective.