Prediction Of Grafting Healing In Melon Based On Hyperspectral Data

The traditional judgment of the healing state of grafted seedlings is mainly based on the experienced agricultural personnel to judge the seedlings 10 days after grafting.In large-scale production,if the healing state of grafted seedlings can be predicted in advance,it can not only effectively improve the utilization rate of grafting healing devices,but also bring huge economic effects to melon planting.In this paper,we tried a more efficient method to predict the healing of melon grafted seedlings.By obtaining spectral data of grafted seedlings,ENVI software was used to extract Region of Interest(ROI)and calculate the average spectral value.The First derivative(FD),Second derivative(SD),Standard Normal transform(SNV),Detrend,Smooth 21,Multiplicative Scatter Correction(MSC)and more than two combined pretreatment methods,Three classification models including Support Vector Machine(SVM),Decision Tree and XGBoost were established.Principal Component Analysis(PCA),Competitive Adaptive Reweighting Sampling(CARS),Genetic Algorithms(GA)and Successive Projections Algorithm(SPA)were used to select characteristic variables,and then the characteristic variables were used to establish the classification and discrimination model.It provides theoretical basis for the prediction of grafting healing of grafted seedlings.The results of the classification discriminant model established by feature variables show that,in the classification results of SVM model,two different data pretreatment methods combined with three feature variable selection algorithms,the F1＿score values of model calibration set and prediction set established by the selected feature variables are58%-87% and 53%-80%,respectively.The original spectral data were selected by CARS feature,and the F1＿score value predicted by the model was the highest,reaching 80%.In the classification results of decision tree model,the F1＿score value of correction set and prediction set is 58% ～ 89% and 53% ～ 85% respectively.After FD data preprocessing and dimensionality reduction by GA and PCA algorithms,the F1＿score obtained reaches85%.In the classification results of XGBoost model,the model calibration set and prediction set F1＿score established by combining the feature variables selected by the three algorithms with the two pretreatment methods are 83%-95% and 80%-93%,respectively.After combining FD data preprocessing with GA algorithm,the model F1＿score reached 93%.Fd-ga-xgboost model works best among all the combinations of data preprocessing methods and classification models.The prediction method of grafting healing of melon seedlings based on hyperspectral technology can accurately judge grafting healing of melon seedlings at 6 days after grafting by using spectral data of grafting seedlings at 4,5 and 6 days after grafting.