Identification of crack features in fresh jujube using Vis/NIR hyperspectral imaging combined with image processing

Abstract

Detection of crack defect in fresh jujube is a critical process to guarantee jujube quality and meet processing demands of fresh jujube fruit. This study presented a novel method for identification of fresh jujube crack feature using hyperspectral imaging in visible and near infrared (Vis/NIR) region (380–1030nm) combined with image processing. Hyperspectral image data of samples were used to extract the characteristic wavebands by chemometrics, which integrated the method of partial least squares regression (PLSR), principal component analysis (PCA) of spatial hyperspectral image (SPCA) and independent component analysis (ICA) of spatial hyperspectral image (SICA). On the basis of the selected wavebands, least-squares support vector machine (LS-SVM) discrimination models were established to correctly distinguish between cracked and sound fresh jujube. The performance of discriminating model was evaluated using receiver operating characteristics (ROC) curve analysis. The results demonstrated that PLSR–LS-SVM discrimination model with a high accuracy of 100% had the optimal performance of “area”=1 and “std”=0. For acquiring rich crack feature information, SPCA was also carried on images at the five characteristic wavebands (467, 544, 639, 673 and 682nm) selected by PLSR. Finally, the SPC-4 image was explored to identify the location and area of crack feature through a developed image processing algorithm. The results revealed that hyperspectral imaging combined with image processing technique could achieve the rapid identification of crack features in fresh jujube.