南京林业大学学报(自然科学版) ›› 2020, Vol. 44 ›› Issue (2): 150-158.doi: 10.3969/j.issn.1000-2006.201903017

• 研究论文 • 上一篇    下一篇

基于Logistic回归模型的红松立木腐朽分级预测

郝泉龄1(), 徐国祺1,*(), 王立海2, 时小龙1, 许明贤1, 纪文文1, 张广晖1   

  1. 1.东北林业大学工程技术学院,黑龙江 哈尔滨 150040
    2.东北林业大学,森林持续经营与环境微生物工程黑龙江省重点实验室,黑龙江 哈尔滨 150040
  • 收稿日期:2019-03-06 修回日期:2019-05-14 出版日期:2020-03-30 发布日期:2020-04-01
  • 通讯作者: 徐国祺
  • 基金资助:
    国家自然科学基金项目(31570547);国家自然科学基金项目(31500470)

Logistic regression-based prediction of wood decay in standing Korean pine (Pinus koreiensis)

HAO Quanling1(), XU Guoqi1,*(), WANG Lihai2, SHI Xiaolong1, XU Mingxian1, JI Wenwen1, ZHANG Guanghui1   

  1. 1. College of Engineering and Technology, Northeast Forestry University, Harbin 150040, China
    2. Key Laboratory of Forest Sustainable Management and Environmental Microorganism Engineering of Heilongjiang Province, Northeast Forestry University, Harbin 150040, China
  • Received:2019-03-06 Revised:2019-05-14 Online:2020-03-30 Published:2020-04-01
  • Contact: XU Guoqi

摘要:

【目的】红松活立木在生长过程中极易受到各种生物的侵袭,其中木腐菌造成的腐朽是最为广泛和严重的危害之一,活立木腐朽将会造成木材资源的损失,降低木材的力学性能。为了监测红松活立木的健康状况,减少因腐朽引起的相关损失,须定期检测红松活立木腐朽状况,并采用腐朽分级的方法进行评估。【方法】使用阻抗仪和电阻断层成像技术(ERT)两种无损检测手段对黑龙江省五营国家森林公园内的红松活立木进行检测,并在检测截面使用生长锥钻取木芯。将野外测量数据导入电脑,获取立木的阻力曲线和电阻断层图像,分析处理微针阻力的下降幅度,得到样木的阻力损失;根据立木心材与边材腐朽在ERT图像上呈现的不同特征,首次采用一种基于OpenCV的图像处理技术识别腐朽样木电阻断层图像缺陷部分,实现了心材和边材缺陷面积的定量表征;采用烘干法测定质量损失率,根据每株样木的质量损失率将所有样木分为5个腐朽等级。将多分类有序因变量的Logistic回归分析引入到立木腐朽程度分级问题中,以阻力损失、心材缺陷面积比和边材缺陷面积比为自变量,腐朽等级为因变量,建立腐朽分级预测模型。检验模型的有效性、拟合优度及预测能力。【结果】Logistic回归模型性能良好,五营国家森林公园30株红松样木的综合预测准确率为86.67%,各等级预测准确率分布不均,高等级腐朽的预测准确率相对较高。【结论】运用提取心材与边材腐朽的方法处理ERT图像,有效地提高了利用ERT图像识别腐朽的准确率。建立Logistic回归模型可以较准确地预测出立木腐朽的等级,较之烘干法测定腐朽程度,其腐朽评估效率有了明显提高,且对立木损伤更小。

关键词: 红松, 腐朽程度, 无损检测, ERT图像处理, 多分类有序Logistic回归, 概率预测模型

Abstract:

【Objective】The Korean pine (Pinus koraiensis) is an ancient and valuable tree species with high research value. Living Korean pine trees are prone to attack by various organisms during their growth; the decay caused by wood rot fungi is one of the most extensive and serious growth hazards. The decay of living wood results in the loss of wood resources and reduces the mechanical properties of the wood. To monitor the health status of Korean pine trees and reduce the loss caused by rotting, the decay status of Korean pine was detected periodically and evaluated by the method of decay grading. 【Method】Thirty standing Korean pine trees in Wuying National Forest Park of Heilongjiang Province were tested using a resistograph and electrical resistance tomography (ERT). The cores drilling were made at the same section to obtain the resistograph readings. The field measurement data were imported into a computer and the resistance curve and resistance tomography images were obtained. After analyzing and processing the drop range of the microneedle resistance, the resistance loss of the sample wood was obtained. According to the different characteristics presented by the decayed heartwood and sapwood in the ERT images, an image processing technique based on OpenCV was used to identify the defected part of the resistance tomographic image of the decayed sample wood; further, quantitative characterization of the defected area of the heartwood and sapwood was performed. The defect area ratios of the heartwood and sapwood were calculated separately. The mass loss rate was determined using the drying method. According to the mass loss rate of each sample wood, all samples were divided into five decadent grades. Multi-category ordered-dependent-variable Logistic regression was introduced into the decay grade classification of the standing trees. The prediction model of decay classification was established by considering resistance loss, area ratio of heartwood defect, and area ratio of sapwood defect as independent variables and the decay grade as the dependent variable. The validity, goodness of fit, and prediction ability of the model were tested.【Result】The performance of the Logistic regression model was good. The comprehensive prediction accuracy of the thirty Korean pine trees was 86.67%, and the accuracy rate of each grade prediction was uneven. The prediction accuracy of high-grade decay was relatively high. 【Conclusion】The Logistic regression model could accurately predict the rank of decay in standing trees. Compared with the drying method, the evaluation efficiency of Logistic regression model was higher and the damage to the standing trees was less.

Key words: Korean pine, decay grade, nondestructive testing, ERT image identify, multi-category ordered-dependent-variable Logistic regression, probabilistic prediction model

中图分类号: