木材科学与技术2025，Vol.39Issue(3)：7-18,12.DOI:10.12326/j.2096-9694.2025008
基于改进YOLOv8的树木年轮实例分割及信息检测方法
Study on Tree Ring Instance Segmentation and Information Detection Based on Improved YOLOv8
摘要
Abstract
In climatology,ecology,and archaeology,tree rings hold irreplaceable value for studying environmental change and historical succession.To improve the efficiency and reduce human interference in traditional tree-ring detection,this study proposes DCW-YOLOv8,an instance segmentation model enhanced with dilation-wise residual(DWR)attention module,lightweight CARAFE upsampling,and dynamic Wise-IoU loss.A detection method using model-generated masks is designed to quantify ring counts and widths.The ablation and comparative experiments show DCW-YOLOv8 achieves superior mask mAP(86.4%for mAP@0.50,53.6%for mAP@0.50∶0.95)versus state-of-the-art models.Attention visualization confirms stronger focus on tree-ring features.Detection results include 86.2%count-accuracy,70%width measurements within±0.5 mm error,and a mean width error of 0.295 mm.This model provides a novel automated approach for tree-ring analysis.关键词
树木年轮/深度学习/实例分割/年轮宽度/年轮数量Key words
tree ring/deep learning/instance segmentation/ring width/ring number分类
农业科技引用本文复制引用
王品博,王雨晨,王双永,周海宾..基于改进YOLOv8的树木年轮实例分割及信息检测方法[J].木材科学与技术,2025,39(3):7-18,12.基金项目
"十四五"国家重点研发计划课题"应县木塔本体信息采集、挖掘与数据模型构建"(2023YFF0906301). （2023YFF0906301）
