Research Of Discrimination Maturity Of Apricots Based On Hyperspectral Imaging Technique

Apricot fruit is favored by people for its abundant nutritional value and high medicinal value.However, during the process of practical production, by reason that its picking period coincides with the high temperature season with a short mature period and that the apricot fruit is the typical climacteric fruit featured by putrescibility, deterioration and difficult storage, the development of the apricot industry is restricted to a great extent. Therefore, the study on how to determine the appropriate maturity levels of fresh apricot optimal for harvest and rapid classification of fresh apricot according to different industrial requirements in the post-harvest has significant meaning.The’shajin’ red apricot of taigu was selected as the research object.The maturity of apricots is classified and distinguished in terms of two aspects by using hyperspectral imaging technology, namely spectrum and image information, which provides data support and theoretical basis for the rapid and lossless on-line inspection of the apricots’ maturity.Conclusions of the research are as follow:(1)Analyse and compare the effect of different proceeding methods to the partial least-square(PLS) models.The spectral information (400～1000nm,900～1700nm) was preprocessed by using 9 different pretreatment methods, extracted by ENVI software from the hyperspectral image data of fresh apricot,and the PLS discriminant models was established.It proved that the best spectral pretreatment method was the standard normal variate(SNV) in Vis-NIR wave range,while the best spectral pretreatment method was the Multiple Scattering Correction(MSC) in Vis-NIR wave range.(2)The maturity of fresh apricots was classified and distinguished based on spectral information in Vis-NIR wave range. Spectral information preprocessed by SNV was used; 9 principal components and 9 specific wavelengths were extracted by using principal component analysis(PCA) and regression coefficient(RC). The PLS、PCR、LS-SVM and ELM discrimination models had been built respectively on the basis of the data of full spectrum, principal component and specific wavelength, and its model parameter and predictive effect had been compared.The result showed that LS-SVM model based on full spectrum or principal component worked best,the identifying rate both was 96.67%.(3)The maturity of fresh apricots was classified and distinguished based on spectral information in NIR wave range.Spectral information preprocessed by MSC was used.12 principal components and 6 specific wavelengths were extracted by using PCA and RC. The PL、PCR、LS-SVM and ELM discrimination models had been built respectively on the basis of the data of full spectrum, principal component and specific wavelength, and its model parameter and predictive effect had been compared.The result showed that the LS-SVM model based on full spectrum was best,the identifying rate was 94.17%.(4)The maturity of fresh apricots was classified and distinguished based on image information in Vis-NIR wave range. The PLS discriminant model of apricot maturity was established based on the color characteristic parameter extracted via the original hyperspectral image or extracted via the image through contraste enhancement respectively, and its model parameter and predictive effect had been compared. In general consideration, the PLS distinguishing model had the best effect when built from the characteristic value of colors extracted from unprocessed images, with a distinguishing rate of 88.75%.(5) Discriminating analysis on the maturity of fresh apricot was carried out based on spectral and image information of Vis-NIR wave range. The discriminant model of apricot maturity was respective established based on principal components and specific wavelengths respectively, together with the characteristic colors of the image. Then its model parameter and predictive effect had been compared. The result showed that LS-SVM model had the best effect based on principal components and the color characteristic parameter extracted from the hyperspectral image, with a distinguishing rate of 94.17%.