光谱学与光谱分析, 2019, 39 (6): 1870, 网络出版: 2019-07-10  

马铃薯冠层光谱响应特征参数优化与生长期判别

Parameter Optimization of Potato Spectral Response Characteristics and Growth Stage Identification
作者单位
1 中国农业大学现代精细农业系统集成研究教育部重点实验室, 北京 100083
2 北京工商大学, 食品安全大数据技术北京市重点实验室, 北京 100048
摘要
快速判别马铃薯作物的生长进程是指导田间关键生长期科学水肥管理的重要依据。 研究在马铃薯发棵期(M1)、 块茎形成期(M2)、 块茎膨大期(M3)和淀粉积累期(M4)四个关键生长期, 利用ASD便携式光谱仪采集80个样本区的314组作物冠层反射率数据, 并同步采集叶片测定叶绿素含量。 在光谱数据预处理后, 分析了马铃薯不同生长期的光谱反射率变化特征, 并初步选取了光谱“峰谷”响应参数, 提出了一种基于方差分析与变量减少组合的光谱参数筛选算法(variance analysis combined with variable reduction, VACVR)用于明确光谱学响应的优化指标, 采用Kennard-Stone(K-S)法划分样本集, 最终基于支持向量机(support vector machine, SVM)方法建立马铃薯关键生长期判别模型。 针对光谱数据, 首先使用变量标准化(standard normalized variable, SNV)进行光谱预处理, 在定性分析了随着生长期的推进马铃薯冠层反射特征的变化趋势的基础上, 基于作物生长期动态光谱学响应与峰谷特性选取14个参数, 包括: 8个位置参数、 2个面积参数、 4个植被指数参数。 采用K-S算法将样本按照3∶1划分为训练集(240个样本)和测试集(74个样本)。 分析马铃薯不同生长期冠层反射光谱发现, 随生长期的推进冠层光谱存在差异性: 即在400~500和740~880 nm范围内, 光谱反射率呈降低趋势; 在530~640和910~960 nm范围内, 反射率呈升高趋势; 在530~640 nm范围内, M2和M3生长期的平均光谱非常接近, M4生长期的平均光谱与其他三个生长期的差别较大。 叶绿素平均含量随生长期的进程, 从M1(28.12 mg·L-1)到M2(31.04 mg·L-1)增加, 在M2生长期达到最大值, 之后M3(22.00 mg·L-1)和M4(15.36 mg·L-1)依次减少。 光谱响应参数随着生长期的进程, 绿峰位置Lg和红谷位置Lr逐渐红移, 红边位置Lre逐渐蓝移; 蓝边面积Abe逐渐增大, 红边面积Are逐渐减小; 红边面积与蓝边面积比值依次呈现减小趋势。 根据VACVR算法筛选10个敏感光谱响应参数, 建立SVM判别模型, 训练集判别率为100%, 测试集判别率为94.59%, 该模型可在判别马铃薯的生长期的基础上为田间管理决策提供支持。
Abstract
In order to satisfy the field management requirement, the research was conducted to indicate the optimizing parameters and identify the growth stage based on the canopy spectral response of potato plants. Aiming to the four growth stages of potato, tillering stage (M1), tuber formation stage (M2), tuber expansion stage (M3) and starch accumulation stage (M4), 80 sample plots were divided in the potato field. The 314 groups data of canopy spectral reflectance were collected by ASD Handheld2 portable spectrometer. The potato leaves were collected synchronously in per sample plot to determine the chlorophyll content. After spectral pretreatment, the spectral reflectance changes of potato crop at different growth stages were analyzed. The spectral response parameters of potato growth stages were selected according to the “peak-valley” reflectance characteristics. A new algorithm was proposed to select sensitive spectral response parameters based on the variance analysis combined with variable reduction (VACVR) method. The Kennard-Stone (K-S) algorithm was used to divide the all samples into training sets and test sets. The identification model of potato growth stages was established by the support vector machine (SVM) method. For spectral reflectance, the standard normalized variable (SNV) was used for spectral pretreatment. Based on the qualitative analysis of the canopy reflection characteristics change trend as potato growth stage progress, the 14 spectral response parameters, including the 8 position parameters, the 2 area parameters and the 4 vegetation index parameters, were selected combining with spectral “peak-valley” characteristics and the dynamicspectral response of potato growth stages. The K-S algorithm was used to divide the overall sample according to 3∶1 into a training set (240 samples) and a test set (74 samples). In general, the canopy spectral reflectance varied with the growth stages progress. In the range of 400~500 and 740~880 nm, the spectral reflectance decreased. In the range of 530~640 and 910~960 nm, the spectral reflectance increased. In the range of 530~640 nm, the canopy average spectral reflectance of the M2 and M3 growth stage were very close. The canopy average spectral reflectance of the M4 growth stage was significantly different from that of the other three growth stages. The average chlorophyll content increased from M1 (28.12 mg·L-1) to M2 (31.04 mg·L-1), reaching a maximum in the M2 growth stage. And the average chlorophyll content of M3 (22.00 mg·L-1) and M4 (15.36 mg·L-1) reduced successively. With the progress of the growth stage, the green peak position and the red valley position gradually red-shifted, the red edge position gradually blue-shifted, the blue edge area gradually increased, the red edge area decreased gradually, and the ratio and normalized ratio of red edge area to blue edge decreased in turn. According to the VACVR algorithm, 10 sensitive spectral response parameters were selected to establish the SVM identification model. The identification rate of the training set was 100%, and the identification rate of the test set was 94.59% (70/74). Therefore, the model can identify the potato growth stage to support the water and fertilizer management in the potato field.

孙红, 刘宁, 邢子正, 张智勇, 李民赞, 吴静珠. 马铃薯冠层光谱响应特征参数优化与生长期判别[J]. 光谱学与光谱分析, 2019, 39(6): 1870. SUN Hong, LIU Ning, XING Zi-zheng, ZHANG Zhi-yong, LI Min-zan, WU Jing-zhu. Parameter Optimization of Potato Spectral Response Characteristics and Growth Stage Identification[J]. Spectroscopy and Spectral Analysis, 2019, 39(6): 1870.

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!