改进YOLOv11n的菌落图像检测算法OA

In view of the miscounting and undercounting caused by small object sizes,high density,adhesion,clumping,and overlapping in bacterial colony image detection,this paper proposes an improved algorithm named YOLOv11n-DPD based on the YOLOv11n model,aiming to enhance detection accuracy for small object colonies.Firstly,the C3k2 module in the Backbone is replaced with the C2f_Dense module to strengthen feature reuse and model representation capabilities for small object colonies.Secondly,a P2 small object detection layer is added to improve feature integration across hierarchical levels and multi-scale feature fusion,thereby boosting the extraction of small object colony features.Finally,a dynamic detection head is adopted to replace the original decoupled detection head,which effectively integrates multi-scale information and further enhance detection accuracy for small object colonies.Experiments on the Roboflow colony dataset show that,in comparison with YOLOv11n,YOLOv11n-DPD exhibits significant improvements in all relevant metrics;its precision,recall rate,mAP@0.5,and mAP@0.5:0.95 are increased by approximately 12.5%,15.5%,14.2%,and 6.8%,respectively.The YOLOv11n-DPD exhibits strong practical applicability for bacterial colony image detection.