Research On Diagnosis Of Pear Leaves Deficiency Based On Infrared Spectroscopy

Nutrient deficiencies often occur during the growth of pear trees.therefore,rapid,costeffective monitoring of the nutritional deficiency status of pear leaves is of great value for effective cultivation management.Traditional pear leaf deficiency identification methods include expert diagnosis based on appearance and physical and chemical analysis.Expert diagnosis has low recognition Accuracy and requires skilled fruit farmers.Physicochemical experiments diagnose deficiencies by detecting the chemical composition of pear leaves,which is costly and has a slow turnaround of results.Near-infrared(NIR)spectroscopy has the advantages of nondestructive,rapid,efficient,and accurate.nutrient-deficient pear leaf samples collected in national High-tech Agricultural Park,Anhui Agricultural University were analyzed with a handheld miniature near-infrared(NIR)spectrometer operating at a reflectance spectrum of 900–1700 nm.The content of this manuscripts is as follows:(1)Analyze the effect of different pre-processing methods on the Accuracy of pear leaf deficiency recognition model.In this manuscript,six single pre-treatment methods,including the first derivative transformation(FD),second derivative transformation(SD),Savitzky-Golay Smoothing(SGM),Log transformation,multiple scattering correction(MSC)and standard normal variate(SNV)methods,as well as six mixed pre-processing methods—SGM+MSC,SGM+SNV,SGM+MSC+FD,SGM+MSC+SD,SGM+SNV+FD,SGM+SNV+SD.Combined with different pre-processing methods,65 recognition models were established by support vector machine(SVM),NAS,random forest(RF),gradient boosting decision tree(GBDT)and extreme gradient boosting(XGBoost).The spectra preprocessed by SNV show good performance in different modelling methods,Modeling with SVM performances best.SNV-SVM model has good predictive performance with 93.02%Accuracy and 92.26% Macro-F1-Score on the training set and 79.73% Accuracy and 77.81%Macro-F1-Score on the test set.(2)We designed and proposed a genetic algorithm based on particle swarm optimization(PSO＿GA)for feature extraction and modeling hyperparameter optimization.Combined with GA,PSO,PSO＿GA,3 recognition models were established by SVM.It shows that,compared with GA-SVM and PSO-SVM recognition models,the performance of PSO＿GASVM model has been improved.The Macro-F1-Score and Accuracy of PSO＿GA-SVM are90.88%.and 91.03%.Therefore,the application of PSO＿GA algorithm in feature extraction and optimization modeling of hyperparameters can effectively improve the Accuracy of recognition model and the ability of NIR spectrum in diagnosing pear leaf deficiency types.(3)An diagnosis system for pear leaf deficiency was developed.The front-end development of webpage is based on bootstrap framework.The back-end development of webpage is based on Py Charm platform.My SQL is used for persistent data storage.The system integrates four functional modules of pear leaf deficiency diagnosis,user setting,data processing and authority management.Users can recognize pear leaf deficiency through the system.To sum up,in this manuscript,pear leaves were collected in The National High-tech Industrial Park of Anhui Agricultural University.We explored the method to diagnose the deficiency of pear leaf based on near-infrared spectroscopy with handheld Macro-nearinfrared spectrometer.To provide theoretical guidance for fruit growers to fertilize scientifically and promote the sustainable development of pear industry.