基于不同物候期苹果树叶片光谱特征预测果实糖度

Forecasting Apple Sugar Content Based on Leaf Characteristic Spectra in Different Phenological Phases

苹果糖度是评价苹果品质的重要指标。通过分析苹果开花期、新梢生长期、萌芽开花坐果期、新梢旺长期、花芽分化期和落叶期等6个重要的生理物候期果树叶片的光谱特征,并与最终采集到的对应位置的果实糖度信息进行二维相关运算,获得了果树叶片光谱信息中反映果实糖度的敏感谱段,构建了糖度敏感光谱,通过引入计算获得的不同物候期糖度贡献权重,最终构建了带权糖度敏感光谱,并基于该光谱进行了糖度预测。通过对不同物候期的二维相关分析获得了果实糖度敏感谱段（530~570 nm和700~720 nm）,经过主成分分析分别获得了不同生理物候期的果叶光谱主成分,利用不同物候期的主成分进行了果实糖度回归分析,量化了某单一时期果树生长对果实糖度的贡献比例,并获得了光合作用强度变化等重要信息。利用各物候期的糖度贡献权重对原始果叶糖度敏感光谱进行变换,最终获得带权果叶糖度敏感光谱,并基于该光谱进行了果实糖度预测。分别建立了基于主成分分析的多元线性回归模型以及基于参数优化的支持向量机回归预测模型,结果显示,利用参数优化的支持向量机回归模型获得了较高的糖度预测精度。其测定系数Rc2为0.9202,RMSEC=0.3892 Brix,预测Rv2达到0.9443,均方根误差是0.5246 Brix。利用不同物候期果叶光谱预测苹果糖度的研究结果进一步揭示了果实糖度的积累过程。

Apple sugar content is one of the important indicators in evaluating fruit quality. The apple leaf spectra characteristics were detected separately in six important phenological periods including flowering stage, shoot-growing stage, fruit setting stage, branch shooting stage, bud differentiation stage, and defoliation stage. 2-D correlation operation between the leaf spectra and apple sugar contents was done to explore the sensitive spectral bands reflecting sugar content. After calculating, sugar contribution value of different phenological phases was obtained, which was used to construct the weighted sensitive spectra to forecast fruit sugar content. Firstly, the visible and near infrared spectral reflectance of apple leaf samples from different phenological phases were measured using the spectrophotometric method. And the sugar content of the fruit sample growing near each leaf samples was collected and measured using laboratory method. By introducing 2-D correlation analysis technology, the fruit sugar content sensitive spectra （530-570 nm and 700-720 nm） were achieved. Then the principal component analysis was conducted among each sensitive spectrum and the principal components were obtained in different phenological phases. The principal components were used to perform sugar content regression analysis, which quantified the contribution proportion to fruit sugar accumulation in each certain phase. And the other important information such as intensity change of photosynthesis in different physiological phases was obtained as well. By using of the sugar contribution weight in each different phenological phase, the original sensitive spectra were transformed and the weighted leaf characteristic spectra were achieved. Based on the characteristic spectra, two models were established, multiple linear regression model based on the principal component analysis and the model based on SVM （ Support Vector Machine） with parametric optimal solution. The SVM regression model showed good accuracy. The calibration Rc2 reached 0.9202 with the RMSEC （ Root Mean Square Error of Calibration） of 0. 3892 Brix, and the validation Rv2 reached to 0. 9443 with the RMSEP （ Root Mean Square Error of Prediction） of 0. 5246 Brix. This research indicated that it was possible to forecast apple sugar content by using apple leaf spectra information in different phenologieal phases and meanwhile the fruit sugar accumulation process was revealed.