基于机载高光谱成像的柑橘产量预测模型研究

Development of Citrus Yield Prediction Model Based on Airborne Hyperspectral Imaging

果树的隔年结果现象严重影响果园的果实产量和经济效益。选择受隔年结果现象影响较为严重的柑橘作为研究对象,运用机载高光谱成像仪在较早生长季节(2003年4、5、6月)获取柑橘果树的高光谱图像,利用偏最小二乘回归(PLS)确定基于高光谱图像数据的模型预测变量,建立柑橘产量的多元线性回归(MLR)和人工神经网络(ANN)预测模型。研究结果表明,利用5月份获得的高光谱图像建立的模型具有最优的产量预测效果,而且PLS-MLR模型比PLS-ANN模型具有更好的稳定性和一致性。该研究结果为今后研制和开发基于高光谱成像技术的柑橘产量预测方法提供了重要的理论和技术基础。

The phenomenon of alternate bearing of fruits seriously affects the fruit yields as well as the economic benefits of orchards. The present study investigated the possibility of airborne hyperspectral images to predict the fruit yield of individual citrus trees. The hyperspectral data were first extracted from the images and the predictors were determined using partial least-squares regression （PLS）. The optimal number of PLS factors were identified, and they were used as inputs of citrus yield prediction models developed by means of multiple linear regression （MLR） and artificial neural network （ANN） modelling techniques. The results showed that the models based on the hyperspectral images obtained in May achieved the best prediction, and the PLS-MLR model has a better stability and consistency than the PLS-ANN model. These results proviode an important theoretical and technical foundation for the future research and development of hyperspectral imaging-based citrus production techniques.