The Bruising Grade Discrimination Of Lingwu Jujube Based On Hyperspectral Imaging Technology

Lingwu jujube,also known as "Ma Ya jujube",is an excellent fruit variety in Ningxia.It has a long breeding history,excellent varieties,fresh taste,rich juice,crisp pulp and high nutritional value.Lingwu jujube has been cultivated for more than 1300 years.Since the Tang Dynasty,Lingwu jujube has been one of the royal tributes,known as "treasures in fruit".Lingwu jujube is one of the most important economic crops in Ningxia,and its market share is increasing year by year.During the production and transportation of Lingwu jujube from field picking to grading processing,packaging and transportation,there will be loss processes such as extrusion and collision,resulting in internal bruises,which will greatly shorten the shelf life and cause huge losses.Therefore,A convenient,practical and low-cost method that can quickly identify bruised Lingwu jujube is of great value.This paper takes Ningxia Lingwu jujube as the research object,uses hyperspectral imaging technology to discriminate the bruising grade of Lingwu jujube,and provides a theoretical basis for the online quality identification of Lingwu jujube,so as to achieve efficient guidance of Lingwu jujube harvesting,post-scientific processing,transportation and storage.The main findings are as follows:(1)The bruising grade discrimination of Lingwu jujube based on Visble/near-infrared(VIS/NIR)hyperspectral imaging technology.The reflection spectrum,absorption spectrum and Kubelka-Munk spectrum of bruises Lingwu jujube were obtained by VIS/NIR hyperspectral imaging technology,and applied to non-destructive discrimination of bruise grades of bruises.Spectra were preprocessed and characteristic wavelengths were selected by Competitive adaptive reweighted sampling(CARS)and Interval variable iterativespace shrinkage approach(iVISSA).Partial least squares-discriminant analysis(PLS-DA)and Support vector machine(SVM)were used to construct the classification discriminant model.By comparison,the results show that the A-Raw-iVISSA-PLS-DA model has the smallest cross-validation error,while the number of feature variables is the smallest,and the accuracy of the calibration set and prediction set are 89%and 100%,respectively.Demonstrated the feasibility of judging the bruising grade of Lingwu jujube based on absorption spectrum.(2)Discrimination of bruise grades in Lingwu jujube based on Near-infrared(NIR)atlas fusionThe NIR hyperspectral imaging system was used to collect near-infrared spectral images of intact Lingwu jujube and five grades of bruised Lingwu jujube(3,6,9,12,and 15 times of bruising impact),and extract features based on spectral features and textural features,respectively.The spectral features were extracted using 1 1 spectral preprocessing methods and 5 feature wavelength extraction methods;Textural features using the Gray-level Co-occurrence Matrix(GLCM)algorithm to extract 8 textural features parameters from the hyperspectral principal component images of different bruise grades of Lingwu jujube;then,the textural and spectral combination features are optimized.Finally,a discriminant model is established based on a single spectral feature,a single textural feature,and feature crosses.Among all the results,the feature crosses model performed the best,with 94%accuracy on the calibration set and 98%accuracy on the prediction set.(3)The bruising grade discrimination of Lingwu jujube based on Convolutional Neural Networks(CNN)In this paper,a series of optimizations have been made on the VGGNet network used for bruising grade discrimination of Lingwu jujube and the dataset of hyperspectral images of Lingwu jujube in bruised state,including reducing the network convolution layer,reducing a fully connected layer and Introduce RMSprop optimization algorithm and random pooling.At the same time,Retinex color enhancement,rotation and mirror flip were done for the dataset.The improved network structure has higher accuracy than traditional VGGNet in identifying bruised Lingwu jujube,and the network convergence speed is greatly accelerated.The test results show that the improved convolutional neural network fully learns the bruise characteristics of Lingwu jujube,and the accuracy rate reaches 97.9%,which has a high bruise grade recognition rate.