基于改进YOLOv5的枸杞虫害检测

杜丁健,高遵海,陈倬

Wolfberry pest detection based on improved YOLOv5

Dingjian DU,Zunhai GAO,Zhuo CHEN

表 2 模型改进前后的检测性能对比

Tab.2 Comparison of detection performance before and after model improvement

模型 AP/% P/% R/% mAP50/% M/MB 尺蠖 大青叶蝉 负泥虫 蚜虫 毛田甲 YOLOv5m 99.0 99.4 87.6 78.6 99.4 95.8 88.8 92.8 42.2 YOLOv5-P 99.4 98.4 84.8 79.2 99.3 94.9 90.2 92.2 59.09 YOLOv5-E 99.1 98.1 84.3 72.7 98.4 94.6 86.8 90.5 56.2 YOLOv5-N 99.5 99.2 87.1 84.1 98.9 94.7 92.0 93.7 65.8 YOLOv5-NC（加权融合） 99.5 99.5 85.2 85.9 99.2 96.4 92.4 93.9 66.8 YOLOv5-NC（自适应融合） 99.4 99.5 87.3 87.1 99.3 97.0 92.1 94.5 66.8 YOLOv5-NC（级联融合） 99.4 99.5 85.4 87.2 98.2 96.1 91.3 94.0 66.9 NCF-YOLO 99.4 99.5 87.3 88.0 99.3 97.0 92.1 94.7 57.4