Abstract: To address problems such as low accuracy, a high missed detection rate, and model redundancy in UAV aerial image detection, an improved multi-scale feature collaborative MSF-YOLO detection algorithm based on YOLOv8 was proposed in this study. First, the RepViTBlock and EMA attention mechanisms were used to construct RBE-C2f to replace the C2f module in the backbone network and improve feature extraction capability. Second, the neck network was reconstructed using the TFE, SSFF, and small-target detection layers to achieve multi-feature interactive fusion. Finally, a shared convolution detection head was designed in the detection head to make different detection heads share features, reduce redundant features, and improve the efficiency of the model detection. Compared with the original model in the VisDrone2019 data set, the accuracy rate, recall rate, mAP₅₀, and mAP_50-95 of the improved model increased by 3.1%, 4.0%, 4.6%, and 2.8%, respectively; the number of parameters decreased by 42%, and the model size decreased by 36% to only 3.8 MB. Based on an experimental comparison in a complex environment, the improved model exhibited a better detection effect than the basic YOLOv8 algorithm.