To address challenges in drone aerial imagery such as complex backgrounds, occlusions, dense small objects, and indistinct features, a small object detection algorithm based on an improved YOLOv8n, named CSP-YOLO (cross stage partial-you only look once), was proposed. First, the space-to-depth convolution (SPD-Conv) module was designed to reduce the loss of feature information for small objects and enhance feature robustness. Second, the C-SimAM (convolution simple attention module) attention mechanism, based on residual techniques, was proposed to improve the accuracy of low-resolution feature recognition in complex backgrounds. To enhance the detection capability for dense small objects, the neck structure was redesigned with the addition of a tiny prediction head, which focused on capturing fine-grained information from shallow feature maps to optimize the ability to focus on dense small objects. Finally, to tackle the imbalance between positive and negative samples, the wise intersection over union (WIoU) loss function was introduced to accelerate model convergence and improve the precision of bounding box regression. Experimental results show that the CSP-YOLO algorithm performs excellently on the VisDrone2019 dataset, achieving an mAP0.5 of 0.405, which represents a relative performance improvement of approximately 23.1% compared to YOLOv8n. Compared to mainstream object detection methods and algorithms in the YOLO series, CSP-YOLO not only demonstrates outstanding performance and stability but also achieves a balance between detection accuracy and model size. This makes it suitable for drone-perspective object detection tasks and proves its significant practical value and broad application prospects.