Analysis And Application Of Near Infrared Spectroscopy Data Of The Navel Orange Infected With Huanglongbing

The citrus industry is facing the most dangerous challenge from citrus Huanglongbing?HLB?,Prevention mainly depends on the time When we discover the infected trees.The sooner we find it,the sooner we can remove it.The laboratory diagnosis method is not practical because of the long time and high cost.The main method to discover the infected trees in the field is to observe the symptoms,however results can be unreliable due to personal experience and subjectivity,especially in the absence of specific symptoms,which result in farmer's questioning and obstruct.In this paper,a rapid near-infrared screening technique for HLB In the field was developed.On the basis of predecessors,the following research was carried out:1.Compare and choose the optimized preprocessing method for leaf spectral data.We collected four kinds of navel orange leaves which were confirmed by classical laboratory methods:uninfected leaves with mineral element deficiency?uninfected leaves without mineral element deficiency?infected and symptomatic leaves?infected but not symptomatic leaves.Leaf spectral datas with portable near infrared spectrometer were collected,wihch were processed with different data preprocessing methods to improve the robustness and performance of the model.Select the best data preprocessing method by comparing the corresponding model parameters.The results show that:Compared with normalization,standardization,standard normal distribution?SNV?,first derivative and second derivative,multivariate scattering correction?MSC?method has the best data preprocessing ability and the total recognition rate reaches 95.0%.2.The study and application of leaf-based near infrared rapid detection model for HLB.After determining the optimal data preprocessing method,different modeling methods are used.The optimal modeling method is determined by comparing model parameters.The results show that:Compared with K proximity algorithm,random forest method,naive bayes and integrated learning method,partial least squares linear discriminant analysis is the best modeling method,and the total recognition rate reaches 99.47%.The model was built into a portable near infrared rapid detector and used to detect 720 leaf samples in the field.The results were compared with PCR.The results show that:the predictive recognition rate of the model is 97%and the false positive rate is less than 1%.3.A new method for standardized peeling of bark phloem.Use a ring barker to peel the trunk lengthwise,the results show that:The bark width was 0.5cm,the average length was 11.34cm,and the average thickness was 0.15cm.4.Compare and choose the optimized preprocessing method for bark phloem spectral data.The bark of uninfected and infected fruit trees was collected,and spectral data of bark phloem were collected by near infrared spectrometer,then different data preprocessing methods are used to establish the model.The results show that:Compared with standard normal distribution method?SNV?,multivariate scattering correction method?MSC?,First derivative and Second derivative,normalization has the best data preprocessing ability,and the root mean square error of the corresponding model is 1.8890×10^-5.5.The effects of different sampling sites on the model were compared.We gather the leaves and bark of the same trees at the same time,and use the same method to preprocessing and modeling.The results show that:The performance of the model is different with monitoring positions.The root-mean-square error?RMSEP?of the models were all in the order of 10^-5,and the RMSEP of leaves(RMSEPL,1.6909×10^-5)<the RMSEP of bark(RMSEPB,1.8890×10^-5)<the RMSEP of composite(RMSEPC,2.5676×10^-5).The determination coefficient?r²?of prediction set was all above 0.9,and the r² of leaves?rL²,0.9396?<the r² of bark?rB²,0.9415?<the r² of composite?rC²,0.9603?.It shows that the models established by the three sampling schemes have good accuracy and predictive ability,at the same time,the precision of the leaf sampling scheme is the highest,but the predictive ability is the worst.The comprehensive sampling scheme has the best predictive ability,but the model precision is the lowest.The model precision(RMSEPB,1.8890×10^-5)and the predictive ability?rB²,0.9415?of the bark sampling scheme could be maintained at a good level.