作物研究2025，Vol.39Issue(4)：319-327,9.
基于Soft-NMS改进YOLOv8的田间烟株计数方法研究
Research on the Tobacco Seedling Counting Method in the Field Based on Soft-NMS Improved YOLOv8
摘要
Abstract
To address the issues of high labor costs,significant errors,and low efficiency in traditional methods of counting flue-cured tobacco transplanting plants,this study utilized field-grown tobacco plants as the subject material.By introdu-cing Soft-NMS to replace traditional NMS in YOLOv8 and incorporating data augmentation strategies,the research focused on rapid counting of small-target tobacco plants in the field.The results showed that S-YOLOv8 significantly outperformed YOLOv8,YOLOv5,and Faster R-CNN in key metrics such as precision,recall,F1-score,mean average precision(mAP),and visualization outcomes.Specifically,S-YOLOv8 achieved a precision of 98.8%,representing a 1.5 points of percentage improvement over YOLOv8.The recall of S-YOLOv8 was 90.6%,0.9 points of percentage increase over YOLOv5.The F1-score reached 93.8%,1.2 points of percentage enhancement over YOLOv8,while the mAP was 91.2%,1.7 points of percentage boost compared to YOLOv8.Additionally,S-YOLOv8 demonstrated faster processing speeds,more uniformly distributed detection boxes in visual inspections,and higher confidence levels,significantly reducing the impact of inaccurate target identification.In conclusion,the combination of Soft-NMS and data augmentation effectively improves the detection accuracy of small-target tobacco plants in the field,enabling rapid and precise determination of flue-cured toba-cco transplanting numbers.This approach provides reliable data support for formulating subsequent production plans and procurement strategies.关键词
烟株/计数/YOLOv8/Soft-NMS/目标识别Key words
tobacco seedling/counting/YOLOv8/Soft-NMS/objectrecognition分类
农业科技引用本文复制引用
于振伟,范江,汪正鑫,李海平,张毅,胡志明..基于Soft-NMS改进YOLOv8的田间烟株计数方法研究[J].作物研究,2025,39(4):319-327,9.基金项目
中国烟草总公司云南省公司科技项目(2022530000242001). （2022530000242001）
