光谱学与光谱分析, 2019, 39 (3): 865, 网络出版: 2019-03-19  

松材线虫危害下马尾松光谱特征与估测模型研究

Spectral Characteristics and Evaluation Model of Pinus Massoniana Suffering from Bursaphelenchus Xylophilus Disease
作者单位
1 长江师范学院计算机工程学院, 重庆 408100
2 长江师范学院三峡库区环境监测与灾害防治工程研究中心, 重庆 408100
3 长江师范学院武陵山区特色资源开发与利用研究中心, 重庆 408100
摘要
松材线虫病又叫松树枯萎病, 是由于松材线虫寄生在松树上引起的毁灭性死亡病害, 其发病速度快、 传播迅速、 防治难度大。 如何识别松材线虫害并对其程度进行估测, 对我国森林资源及生态环境保护具有重要意义。 研究表明, 马尾松叶绿素、 水含量会随着虫害程度的加深逐渐减少, 不同虫害程度的马尾松光谱反射率呈现较大差异, 因此光谱分析技术在虫害程度估测方面具有独特优势。 针对不同虫害程度的马尾松样本, 研究了其光谱特征参数的变化规律, 以实测光谱特征参数为自变量, 样本虫害程度量化值为因变量, 利用线性回归方程构建了虫害程度估测模型。 该研究在光谱特征指标选择和估测模型方法上作了有价值的探索, 对评估松材线虫病害有一定的指导意义, 可为相关研究及当地精准农业提供科学支持和应用参考。 首先针对不同虫害程度的马尾松样本, 研究其在绿光、 红光及近红外波段内的光谱反射率变化规律, 构建指示样本虫害程度的六个光谱特征参数: 绿峰反射率(RGP)、 绿峰位置(GPP)、 红谷反射率(FRB)、 红谷位置(RBP)、 红边斜率(RES)、 红边位置(REP), 分析光谱特征参数与虫害程度的相关性。 然后构建虫害程度估测模型, 其步骤可描述为: (1)计算健康、 轻度、 中度、 重度四种不同虫害程度下的样本光谱特征参数RGP, FRB和RES; (2)量化健康、 轻度、 中度、 重度四种样本虫害程度值; (3)以实测光谱特征参数为自变量, 样本虫害程度量化值为因变量, 利用线性回归方程构建虫害程度估测模型。 实验中选取重庆市涪陵区永胜林场、 冒合寨工区的马尾松林为研究对象, 随机选取健康、 染病、 完全枯死的马尾松植株进行监测。 数据采集过程中使用ASD野外光谱分析仪FieldSepc4, 采集波段范围为从可见光400 nm到近红外波段1 100 nm处, 分辨率为1 nm。 共采集了70条马尾松植株的有效光谱数据, 根据不同虫害程度, 将其划分为健康、 轻度、 中度、 重度和枯死五种类型, 并利用Matlab软件进行处理分析, 得到其光谱反射率曲线。 选择涵盖绿光区(510~580 nm)、 红光区(620~680 nm)和近红外区(680~780 nm)三个波段, 计算各个波段的光谱特征参数, 构建虫害程度估测模型。 实验结果表明: (1)针对枯死样本, 其“绿峰”和“红谷”特征消失, 红边陡峭上升趋势被拉平。 其他几种类型样本光谱特征参数RGP, FRB和RES与虫害程度呈负相关, 虫害程度越深, 其光谱特征参数值越小, 即健康(RGP)>轻度(RGP)>中度(RGP)>重度(RGP), 健康(FRB)>轻度(FRB)>中度(FRB)>重度(FRB), 健康(RES)>轻度(RES)>中度(RES)>重度(RES)。 (2)随着虫害程度加深, 光谱特征参数GPP向长波方向移动, 即存在“红移”现象, 而光谱特征参数RBP和REP向短波方向移动, 即存在“蓝移”现象。 (3)与一元线性估测模型相比, 二元线性估测模型具有较大的相关系数R2, 较小的估计误差E以及残差。 实验中对两棵马尾松样本虫害程度进行估测, 二元线性估测模型的结果为PD=2.990 7和PD=3.679, 与实际情况相符。 在后续研究中将对1 100~2 500 nm波段特征进行相关性分析。
Abstract
Bursaphelenches xylophilus disease, also known as pine wilt disease, is a fatal one caused by the parasitism of pine wilt nematode in pine trees. It’s difficult to prevent and control the disease because of rapid infection and spread. The recognition and estimation of the disease play a significant role in the protection of forest resources and ecological environment in China. Studies have shown that the chlorophyll and water content in Pinus Massoniana will reduce gradually when the pest degree deepens and the spectral reflectance of Pinus Massoniana in different pest degree appears to be greatly different. Therefore, the spectral analysis technique has unique advantages in pest degree estimation. In this paper, the variation regularities of the spectral characteristic parameters were studied for the samples with different pest degrees. Then the measured spectral characteristic parameters were taken as independent variables and the quantization of samples’ disease degree as dependent variables to construct an estimation model for the pest degree with the help of linear regression equation. Valuable efforts made on the spectral characteristic selection and the evaluation model could provide significant guidance for the estimation of Bursaphelenchus xylophilus disease, as well as providing scientific support and application reference for related research and local precision agriculture.Firstly, the variation of the spectral reflectance in the green, red and near infrared bands was studied; six spectral characteristic parameters indicating the degree of pest damage were conducted, including the peak reflectance and their corresponding wavelengths (positions) of the green, and red bands, as well as the slope and position of the red edge; the correlation between spectral characteristic parameter and pest degree was analyzed. Next, the liner models for estimating the pest degree of Pinus Massoniana samples were constructed. The steps consisted of (1) calculating the reflectance of spectral characteristic parameters such as green peak (RGP), reflection of red band (FRB) and red edge slope (RES) for samples in healthy, mild, moderate and severe pest degree; (2) quantizing the pest degree of these samples; (3) taking the measured spectral characteristic parameters as the independent variables and the quantitative value of the pest degree as the dependent variables, and constructing the pest degree estimation models with the linear regression equation. In the experiments, the Pinus Massoniana samples from Yongsheng Forest Farm and the area of Maohe Zhai in Fuling District of Chongqing were investigated and Pinus Massoniana trees belonging to healthy, infected and completely dead categories were tested and monitored separately and randomly. An ASD field spectrometer, FieldSepc4 with a range of 350 to 1 100 nm and a resolution of 1nm, was used to collect spectral data for Pinus Massoniana samples. 70 records of effective spectral data for Pinus Massoniana trees collected were divided into five levels, i.e. healthy, infected mildly, moderately, severely as well as dead according to the different pest levels. Spectral data was then processed by Matlab software to generate the spectral reflectance curves. The spectral characteristic parameters with wavelength covering the green region (510~580 nm), the red region (620~680 nm) and the near infrared region (680~780 nm) were calculated and the estimation models for pest degree were constructed. The results demonstrated that: (1) the spectral characteristics for dead samples such as green peak and red band disappear, at the same time, the steep uptrend of the red edge is leveled. For the remaining kinds of samples, the spectral parameters RGP, FRB and RES are negatively correlated with the pest degree. The deeper the pest degree is, the smaller the parameter is, that is Health (RGP)>Mild (RGP)>Moderate (RGP)>Severe (RGP), Health (FRB)>Mild (FRB)>Moderate (FRB)>Severe (FRB) and Health (RES)>Mild (RES)>Moderate (RES)>Severe (RES); (2) with the deepening of pest degrees, GPP moves towards the longer wave direction called “red shift” in the green peak position while RBP and REP move towards the shorter wave direction called “blue shift” in the red valley position as well as the red edge position; (3) compared with univariate linear estimation models, the bivariate models generate higher correlation coefficients, but smaller estimation error and residual. In the experiment, the two Pinus Massoniana trees were estimated. The results for the bivariate linear estimation models were PD=2.990 7 and PD=3.679 and corresponded with the actual observations. In our following research, the correlation analysis on the spectral characteristics will be extended to the 1 100~2 500 nm bands.

张素兰, 覃菊, 唐晓东, 王宇杰, 黄金龙, 宋清亮, 闵佳园. 松材线虫危害下马尾松光谱特征与估测模型研究[J]. 光谱学与光谱分析, 2019, 39(3): 865. ZHANG Su-lan, QIN Ju, TANG Xiao-dong, WANG Yu-jie, HUANG Jin-long, SONG Qing-liang, MIN Jia-yuan. Spectral Characteristics and Evaluation Model of Pinus Massoniana Suffering from Bursaphelenchus Xylophilus Disease[J]. Spectroscopy and Spectral Analysis, 2019, 39(3): 865.

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!