Non-destructive Testing And Variety Identification Of Volatile Oil Of Zanthoxylum Bungeanum Based On Hyperspectral Technology

Zanthoxylum bungeanum has good edible and medicinal values,and the volatile oil content is an important manifestation of these two values.This article uses hyperspectral technology to realize non-destructive detection and visualization technology of volatile oil content of Zanthoxylum bungeanum,as well as variety identification,and get the following conclusion:（1）In order to carry out rapid,non-destructive,and low-cost volatile oil detection of Zanthoxylum bungeanum,the study took Zanthoxylum bungeanum in Hanyuan County as the test object.Spectral data of Zanthoxylum bungeanum samples in the wavelength range of 400～1000nm were collected,and standard normal variable transformation Spectral data was preprocessed,feature variables were extracted using the iterative retained information variable method,and an extreme learning machine regression model is established.The result of the model is²is 0.8522,RMSEC is 0.3475,²is 0.8365 and The RMSEP is0.5737.In order to further improved the prediction performance of the model,the fruit fly optimization algorithm was used to adaptively optimize the input weights of the extreme learning machine.Finally,the optimized model（IRIV-FOA-ELM）has a coefficient of determination²is 0.8792,RMSEC is 0.3323,²is 0.8659,and RMSEP is 0.3621.（2）The regression model was established for the average spectral data of the Zanthoxylum bungeanum region of interest without any preprocessing,and the band correlation method was determined as the band selection method.Four different models were trained,namely Light GBM,Bagging,Random forest regression,and Xgboost algorithm,Input the pixel point spectrum data of the intercepted hyperspectral image into the four trained models to obtain the predicted value result.The pseudo-color map technology was used to restore the Zanthoxylum bungeanum image,and the MATLAB software was used to generate the visualization image,and the prediction value interval division of the pixel was calculated.The results show that the pseudo-color image obtained by the random forest regression algorithm was more suitable for the realization of visualization technology.（3）The three kinds of prickly ash from different origins were classified and identified.The experiment took Sichuan Hanyuan Zanthoxylum bungeanum,Sichuan Maoxian Zanthoxylum bungeanum,and Shaanxi Dahongpao Zanthoxylum bungeanum as the experimental objects.The spectrum data and texture information of Zanthoxylum bungeanum were extracted,and the characteristic bands and texture features were extracted respectively.The support vector machine classification was established and compared and combined with the two features.Classifier and extreme learning machine.The results show that the accuracy of the test set of support vector machine classifier established by using the feature band and the texture feature of the gray-level co-occurrence matrix reached95.2%,achieving the classification effect.