基于高光谱成像技术的结球甘蓝叶片水分监测  

Detection for moisture of cabbage leaves based on hyperspectral image method

作　　者：宋忠厅 王帅 李增 李建设[1] 王继涛 曹云娥[1] SONG Zhong-ting;WANG Shuai;LI Zeng;LI Jian-she;WANG Ji-tao;CAO Yun-e(College of Wine and Horticulture,Ningxia University,Yinchuan 750021,China;Horticultural Technology Extension Station,Ningxia Hui Autonomous Region,Yinchuan 750011,China)

机构地区：[1]宁夏大学葡萄酒与园艺学院,银川750021 [2]宁夏回族自治区园艺技术推广站,银川750011

出　　处：《西南农业学报》2024年第5期1131-1140,共10页Southwest China Journal of Agricultural Sciences

基　　金：国家重点研发计划项目(2021YFD1600300);六盘山区冷凉蔬菜产业关键技术集成研究与应用示范(2021YFD1600302)。

摘　　要：【目的】实现对结球甘蓝叶片含水量的无损检测,为结球甘蓝田间水分管理提供参考依据。【方法】在结球甘蓝的整个生长季设置23.33 L/株(CK)、11.33 L/株(W1)、15.33 L/株(W2)、19.33 L/株(W3)、27.33 L/株(W4)、28.33 L/株(W5)共6个灌水量处理,收集结球甘蓝莲座期和结球期各99个结球甘蓝叶片样本光谱数据,采用卷积平滑法(Savitzky-Golay)、移动平均法(Moving average)、归一化法(Normalization)和多元散射矫正(Multiple scattering correction)进行预处理,连续投影算法(Successive project-ion algorithm)进行特征提取,多元线性回归模型(Multiple linear regression)进行建模。【结果】在波段400~1000 nm,不同灌水量的结球甘蓝叶片含水量模型的预测精度差异明显。经过预处理的原始光谱建立的预测模型精度均有一定程度提高,其中Savitzky-Golay预处理效果最好。特征提取方法提高了光谱数据质量和准确性;在结球甘蓝莲座期和结球期优选出连续投影算法进行特征波段提取最佳,分别筛选出5和6个特征波长,2个时期筛选出的特征波长基本都在一定的水分敏感波段范围。在结球甘蓝莲座期和结球期,通过多元线性回归建模方法建立的结球甘蓝叶片含水量预测模型都取得了比主成分回归和偏最小二乘回归更高的预测精度;多元线性回归模型的校正集相关系数和均方根误差在莲座期为0.8927和0.8757,在结球期为0.9167和0.9014。【结论】高光谱技术可成功监测结球甘蓝叶片在不同生育期的含水量,为农田水分管理提供依据,为精准灌溉提供技术支撑。[Objective]The study aimed to realize the noninvasive detection of water content of cabbage leaves and provide reference for field water management of cabbage.[Method]In the study,a total of 6 treatments including 23.33 L/plant(CK),11.33 L/plant(W1),15.33 L/plant(W2),19.33 L/plant(W3),27.33 L/plant(W4)and 28.33 L/plant(W5)were set up.The spectral data of 99 cabbage leaf samples at rosette stage and heading stage were respectively collected.Savitzky-Golay,moving average,normalization and multiple scattering correction were used for sample pretreatment,successive projection algorithm were used for feature extraction,and multiple linear regression was used for modeling.[Result] In the range of 400-1000 nanometers,the prediction accuracy of water content model of cabbage leaves with different amounts of rrigation had significant dfference.The accuracy of the prediction models established by the original spectra after pretreatment had a certain degree of improvement,among which Savitzky-Golay had the best pretreatment fect.The feature extraction method improved the quality and accuracy of spectral date.In rosette stage and heading stage,the continuous projection algorithm was selected to extract the feature bands as the best,5 and 6 characteristic wavelengths were selected respectively,and the characteristic wavelengths of the two periods were basically in a certain water sensitive band range.In rosette stage and heading stage,the prediction model of water content of cabbage leaves established by multiple linear regression had higher prediction accuracy than principal component regression and partial least squares regression.The correlation coefficient of calibration set and root-mean-square error of multiple liner regression model were O.8927 and 0.8757 in rosette stage,and 0.9167 and 0.9014 in heading stage.[Conclusion]Hyperspectral technology can successfully monitor the water content of cabbage leaves in different growth stages and provide reference for farmland water management and technical support for precision

关 键 词：高光谱 结球甘蓝 叶片含水量 建模分析 精准灌溉 

分 类 号：S127[农业科学—农业基础科学]