Study On The Mechanical Damage And External Quality Nondestructive Testing Method Of Cuiguan Pear Based On Hyperspectral Imaging Technology

China is a big country of pear production,and Cuiguan pear market has a bright future.In order to guarantee the quality of pears,it is necessary to quickly and effectively sort out the mechanical damage and fruit defects of Cuiguan pear and realize the online detection of Cuiguan pear's quality.This paper mainly studies the non-destructive testing methods for mechanical damage and external defects of Cuiguan pear based on hyperspectral imaging technology.The mechanical crush injury on the surface of Cuiguan pear was used as the research object.The hyperspectral images of Cuiguan Pear were collected from intact samples and samples which after mechanical damage of 1-7 days,11 days and 14 days.The experiment will extract the spectral information of the surface damage area,and the spectral range is500 nm-900 nm.It adopts a centralized method to spectrally preprocess the collected spectrum.The feature wavelengths are extracted by competitive adaptive reweighted algorithm and continuous projection algorithm,respectively.Using linear discriminant analysis and partial least squares method to establish a full-spectrum and characteristic wavelength detection model which identify Cuiguan pear bruising,respectively.Through the analysis of test results,the accuracy of the recognition between intact and mechanical crush injury samples which based on the full-spectrum detection model and the characteristic wavelength detection model is above 90%,and the optimal value is 97.78%.The experiment focused on the various defect features(color difference,scab,brown rot and decay,rotten,umbilical rot)on the surface of Cuiguan pear.The hyperspectral images of Cuiguan pear were collected from intact samples and samples which the surface has color aberration,crusting,brown rot and decay,umbilical rot.After the masking process,the images at 550 nm,649 nm,and 859 nm are selected to generate false color composite images.Fruit stalks and intact areas are defined as positive.Color aberration,crusting,brown rot and decay and umbilical rot are defined as negative.The test accuracy rate is 94.28% when using convolutional neural network to model the data.The research results show that hyperspectral imaging technology can be applied to the detection of mechanical damage and appearance defects in Cuiguan pear.It provides a theoretical basis for Cuiguan pear quality hyperspectral online detection system.