One of the key algorithms for target identification, recognition, and other applications is YOLOv8. In this research, we use the YOLOv8 target detection method to perform model pruning before combining the great techniques of BIFPN module and EfficientRepHead to present an improved YOLOv8 algorithm. The algorithm’s studies on the VisDrone2019 dataset yielded notable results, including a reduction of 22.6% in the number of parameters to 679,484 and a halving of the GFLOPs to 4.0, showing a considerable reduction in computational complexity. With a 25.4% reduction in size to 1.5 m, the model is now more lightweight. In terms of performance, MAP50 improves slightly to 0.231 (an increase of 0.006), while MAP50-95 improves slightly to 0.126 (an increase of 0.002), with an inference time of 0.00073 s, which is approximately 27% faster than the baseline model. These outcomes show how creative and successful our strategy is.