机构地区: 华南农业大学工程学院南方农业机械与装备关键技术省部共建教育部重点实验室
出 处: 《江苏大学学报(自然科学版)》 2013年第5期529-535,共7页
摘 要: 为探究柑橘树钾素质量分数无损检测的技术途径,基于反射光谱建立不同物候期柑橘叶片钾素水平预测模型.以117株园栽萝岗橙树为试验对象,采集4个物候期健康鲜叶数据,用高光谱仪测量叶片反射光谱值,用火焰光度法测定同期同批叶片钾素质量分数.对不同物候期钾素敏感特征波段和钾素质量分数建模进行试验和分析,结果表明:不同物候期钾素质量分数敏感特征波段存在漂移现象;相比多元线性回归,支持矢量回归(SVR)和偏最小二乘法(PLS)用钾素敏感特征波段建模能较好预测K素质量分数;在不同物候期特征波段漂移和模型性能差异情况下,SVR基于反射光谱建立全发育期钾素质量分数模型仍有良好的预测性能,其在测试集上的决定系数R2为0.994,均方误差为0.120,平均相对误差为1.33%. Based on reflectance spectra, the potassium (K) content prediction model was established to realize non-destructive testing of K content in citrus trees. Field experiments were conducted on 117 plan- ted Luogang citrus trees in the Crab Village, and the data was collected on fresh and healthy citrus leaves in four dominant phenological periods. The hyper-spectrometer ASD FieldSpec3 and the flame photometry were used to detect spectral reflectance data and K-contents,respectively. A series of experiments were conducted to analyze the sensitive frequency band of K-contents and the modeling regularity of prediction in different phonological periods. The results show that there is frequency drift of K-contents relevant sen- sitive band in different phenological periods. Compared with MLR, SVR and PLS, better prediction re- sults can be obtained based on K-contents relevant sensitive frequency band. The R2 of 0. 994 and themean square error of 0. 120 with mean relative error of 1.33% are obtained in SVR model on validation set, which illuminates that SVR can well predict K-contents in whole growth periods based on reflectance spectra, regardless of frequency drift and the discrepant model performance.