Development Of Multi-quality Testing Device For Korla Pear

Korla pear is very popular among consumers because of its taste and juiciness,rich nutrition and other advantages.With the improvement of people’s living standard,consumers pay more and more attention to the quality of balsam pear,not only focus on the colour,fruit shape,fruit weight and other related sensory quality,more attention to soluble solids,hardness and other edible quality.At present,the testing of balsam pear related quality parameters are mostly carried out according to the fruit and vegetable quality testing standards,which is time-consuming and laborious,and also causes damage to the balsam pear,which is suitable for random testing,and is not conducive to the quality testing of balsam pear in large quantities.In this paper,a multi-quality inspection device for balsam pears was developed using visible/near infrared spectroscopy,machine vision technology,feature variable screening algorithms and quantitative prediction algorithms combined with an embedded system.The research content and results are as follows:(1)Construction of a multi-point detection device for simultaneous acquisition of images and spectra.The device consists of a spectral image acquisition unit,a control and processing unit,a light source unit,a display unit,a power supply unit and a multi-point detection structure.(2)Korla pear multi-quality inspection software was developed.The inspection software has a user identification function,spectrometer and industrial camera identification and parameter setting function,spectral and image pre-processing function,quality parameter prediction function,real-time display module and communication module.The software was developed in Python under Linux.(3)The prediction model of sensory quality and edible quality of Korla pear was studied.The sensory quality(color,fruit weight)and edible quality(hardness,soluble solids content(SSC))of Korla fragrant pear were selected as the quality indexes of Korla fragrant pear.The spectral data were preprocessed by algorithms such as multivariate scattering correction and standard normal variable transformation.The characteristic wavelengths were screened by continuous projection algorithm.The quality prediction model based on spectrum,image data and their fusion was established by partial least squares regression algorithm.The pear was intercepted by a rectangular frame(800 pixel × 900 pixel).The Otsu method was used to calculate the image threshold and binary processing.The morphology,edge detection operator and filling operation were used to extract the defect contour and obtain the defect area.The quality prediction model is established by four features of long axis,short axis,perimeter and area,and the image texture features are extracted by gray level co-occurrence matrix.A quality prediction model is established by combining image texture features with spectral data corresponding to spectral characteristic wavelengths.Under the feature fusion,the correlation coefficients between the calibration set and the prediction set of the SSC prediction model are 0.933 and 0.931,the root mean square errors are 0.432 % and 0.436 %,the correlation coefficients between the calibration set and the prediction set of the hardness prediction model are 0.941 and 0.926,the root mean square errors are 0.418 kgf / cm2 and 0.478 kgf / cm2,the correlation coefficients between the calibration set and the prediction set of the color a * are 0.946 and 0.935,and the root mean square errors are 0.406 and 0.426.The best model is substituted into the prototype.In the prototype test,the prototype can complete the acquisition of quality information,and the results are highly consistent with the manual detection.