Abstract: Cervical lesion cell detection using object detection algorithms has been popular research in recent years. However, the problems of occasional clumping of cells in slides, high similarity of squamous lesion cell classes, and different distributions of cell properties pose great challenges to the cell classification task. Therefore, to solve the above problems, we propose to design an improved YOLOv8 model. The model achieved to alleviate the above problems while reducing the computation as much as possible by adding additional detection heads, SwinTransformer Block, and GC Block, respectively. The experimental results show that compared with the latest YOLOv8s model, the present model improves by 10.6 percentage points and can effectively detect squamous cervical lesion cells, while resource costed is similar.