| In recent years,the use of real-time,high-efficiency hyperspectral technology to determine the chlorophyll content has gradually replaced the traditional method of chlorophyll determination.However,the traditional hyperspectral technology can not form the information of the map and can’t realize the display,diagnosis and decision of crop growth information.The rise of near-field imaging spectroscopy technology has better integrated high-spectrum and high-spatial-resolution data to form a unified map,providing visual information for crop condition monitoring.In addition,the existing research on the prediction of fruit chlorophyll content using near-field imaging spectroscopy mostly focuses on a single period of fruit tree growth,and there are few studies on different phenological phases of fruit trees.Therefore,the use of near-field imaging spectroscopy to predict the chlorophyll content of apples during different phonological phases is of great significance for the visual monitoring and information management of apple tree growth.The study used the Qixia Apple Park in Yantai as the research area and the Fuji apple tree leaf in the fruiting period as the research object.The study was conducted in autumn shoots stop growing stage(September)of 2016;new shoots flourishing stage(May),spring shoots stop growing stage(June),autumn shoots stop growing stage(September)of 2017.SOC710 VP was used to measure the spectral data.The content of chlorophyll in apple leaves was determined by spectrophotometer.The data were sequentially subjected to reflectance extraction,pretreatment,screening of characteristic wavelengths,construction of characteristic parameters,and the optimal characteristic parameters were selected.Chlorophyll content of different species during the whole phonation period and Chla+Chlb content during different phenological phases were predicted.And through the accuracy test,compared the modeling and forecasting results,and optimized the best estimation model.The main results are as follows:(1)Determined the spectral response of chlorophyll content imagingThrough different characteristic wavelength screening methods,it was determined that the response wavelength of chlorophyll in the imaging spectrum fluctuates near the blue valley(480 nm),green peak(550 nm),red valley(680 nm)and red edge(760 nm).With the change of phenological period,that is,from the new long-term to the long-term spring shoots,and then to the long-term shoots of autumn,the spectral reflectance of apple leaves is decreasing,and the change of Chla+Chlb content is corresponding to that from the new shoots.Long-term long-term spring to stop long-term,and then to the long-term fall shoots,Chla + Chlb content increased.(2)Build and filter the optimal feature parametersIn the prediction of different types of chlorophyll content,The imaging spectral characteristic parameters constructed were optimized by the correlation between Chla,Chlb,Chla+Chlb content and near-field imaging spectroscopy data,and Chla,Chlb,Chla+Chlb content prediction and test accuracy.The results show that the prediction and verification effects of different types of chlorophyll content models are based on r(TCARI)optimization: Chla predictive model validation set has a coefficient of determination R2 of 0.8260,relative error RE of 0.1175,and root mean square error RMSE of 0.2526;Chlb prediction model validation set validation coefficient R2 is 0.7650,relative error RE is 0.1008,and root mean square error RMSE is 0.1666;The Chla+Chlb prediction model validation set determines that coefficient R2 reaches 0.8266,relative error RE is 0.0750,and root mean squared error RMSE is 0.1097.(3)Chla + Chlb is identified as the best predictive speciesIn the comparative analysis of Chla,Chlb and Chla+Chlb content prediction models,it was found that using the imaging spectrum data of apple leaves in the whole phonological period had the best prediction effect on Chla+Chlb content.Take the best prediction ccuracy model r(TCARI)as an example: The validation set of the Chla+Chlb model has a 36.1% reduction in the relative error RE,and a 56.6% reduction in the root mean square error RMSE;The set decision coefficient R2 is 8.1% higher than the Chlb model,and the relative error RE is reduced by 25.6%,and the relative error RMSE is reduced by 34.2%..(4)Determined the optimal phenological phase of Chla+Chlb content predictionIn the prediction of the Chla+Chlb content of apple leaves in different phenological phases using near-field imaging spectral data,the stepwise regression model has better prediction performance than the support vector machine model.The autumn shoots stop growing stage is optimal.The coefficient of determination R2 of the SVM model validation set increased by 12.05% and 22.57% over new shoots flourishing stage and spring shoots stop growing stage,respectively,and the RE decreased by 68.24% and 47.34% over new shoots flourishing stage and spring shoots stop growing stage,RMSE decreased by 56.70% and 47.59%,respectively,over the new shoots flourishing stage and spring shoots stop growing stage;The stepwise regression model correction set determines the coefficient R2 to be 0.9223,the RE to 0.0141,and the RMSE to 0.1148,the calibration set R2 was increased by 0.15% and 15.16% respectively compared with the new shoots flourishing stage and spring shoots stop growing stage,and the RE reduced by 69.78% and 74.46%,respectively,and the RMSE decreased by 28.38% and 48.10%,respectively;Validation set determination coefficient R2 is 0.8994,RE is 0.0270,and RMSE is 0.1263,the verification set determination coefficient R2 increased by 2.81% and 15.22% respectively ove the new shoots flourishing stage and spring shoots stop growing stage,RE were 60.93% and 54.85% longer than those of the new shoots flourishing stage and spring shoots stop growing stage,respectively,the RMSE decreased by 42.06% and 45.25%,respectively.In summary,the use of near-field imaging spectral data to predict the chlorophyll content of apple leaves in different phenological phases is feasible,providing more intuitive technical methods and theoretical support for the visual monitoring of apple tree growth. |
