Research On Non-destructive Detection Of Table Grape Internal Qualities Based On Hyperspectral Imaging Technology

As people pay more and more attention to the scientific diet,people also focus more on the internal quality of table grapes when eating grapes.Soluble solids content and total acidity are two important quality indicators in table grapes.They not only reflect the nutritional value of grapes,but also directly affect the taste of grapes.With the development of science and technology,how to realize the non-destructive detection of soluble solids content and total acidity in table grapes has become the focus of current research.This paper takes table grapes as the research object,and performs non-destructive detection on the internal quality of table grapes based on the hyperspectral imaging technology.The purpose aims to prove the feasibility and improve the accuracy of the quality detection of table grapes.It will provide reference and reference for the online quality detection of soluble solids content and total acidity in table grapes.This paper has mainly completed the following three parts of the work:(1)Prediction of soluble solids content in table grapes based on the hyperspectral imaging technology.First,a 30×30 pixels ROI is selected on the hyperspectral image to obtain an average spectral curve.Then,the actual value of soluble solids content in grapes is measured by physical and chemical experiments.The principal component analysis combining with Mahalanobis distance method is used to remove 13 abnormal samples,and the remaining spectral data is preprocessed to eliminate noise and interference.Different preprocessing methods are used for comparison and analysis,and the 1Der+SG preprocessing method is preferred.The UVE-SPA method is used to extract 10 feature wavelengths variables to eliminate redundant and irrelevant variables in the data.Different predictive models are established based on the feature wavelengths variables,and the PSO-LSSVM model is proposed to predict the soluble solids content of table grapes.By comparing the evaluation criteria of the models,the PSO-LSSVM model predicts the soluble solids content of table grapes well.The correlation coefficient and root mean square error of the prediction set are0.705 and 1.973,respectively.(2)Prediction of total acidity in table grapes based on the hyperspectral imaging technology.Firstly,the actual value of the total acidity is measured through physical and chemical experiments.Then the hyperspectral image is processed,and the amount of data is expanded using data augmentation.The sample set is increased by six times by rotating the hyperspectral images by 90°,180°,270°,and horizontally flipping and vertically flipping.The images are cropped,and cut into individual grape grains images with a size of 160×160 pixels,and manually labeled.Based on the structure of the AlexNet network,a modified AlexNet network is established to predict the total acidity.The modified AlexNet network is added a batch normalization layer and a global average pooling layer to avoid overfitting.The training of network also uses Adam optimizer and mean square error loss function.The modified AlexNet network has better predictive results with 94.17%accuracy rate.The modified AlexNet network is compared with the PLS and LS-SVM models,showing that the modified AlexNet network has better predictive results,and its R_p and RMSEP are 0.853 and 0.419.(3)Application design of table grape internal quality detection system.Based on the non-destructive detection of soluble solids content and total acidity,an internal quality detection system for table grapes is designed.The design of the system is divided into two parts:front-end pages and back-end modules.The front-end page is mainly the design of the detection interface and the division of functions.The back-end module encapsulates sub-modules such as preprocessing methods,feature extraction and models as the back-end data through the communication between Matlab and Java by Myeclipse.The call to the back-end database is implemented through the front-end interface.In summary,this paper study the non-destructive detection of the internal quality of table grapes.The soluble solids content and total acidity are the two most important indicators of its internal quality.The system for predicting the soluble solids and total acidity and designing the system can facilitate the online detection of its internal quality.