RESUMEN
Aflatoxins in moldy peanuts are seriously toxic to humans. These kernels need to be screened in the production process. Hyperspectral imaging techniques can be used to identify moldy peanuts. However, the changes in spectral information and texture information caused by the difference in moisture content in peanuts will affect the identification accuracy. To reduce and eliminate the influence of this factor, a data augmentation method based on interpolation was proposed to improve the generalization ability and robustness of the model. Firstly, the near-infrared hyperspectral images of 5 varieties, 4 classes, and 3 moisture content gradients with 39,119 kernels were collected. Then, the data augmentation method called the difference of spectral mean (DSM) was constructed. K-nearest neighbors (KNN), support vector machines (SVM), and MobileViT-xs models were used to verify the effectiveness of the data augmentation method on data with two gradients and three gradients. The experimental results show that the data augmentation can effectively reduce the influence of the difference in moisture content on the model identification accuracy. The DSM method has the highest accuracy improvement in 5 varieties of peanut datasets. In particular, the accuracy of KNN, SVM, and MobileViT-xs using the data of two gradients was improved by 3.55%, 4.42%, and 5.9%, respectively. Furthermore, this study provides a new method for improving the identification accuracy of moldy peanuts and also provides a reference basis for the screening of related foods such as corn, orange, and mango.
RESUMEN
Peanuts with fungal contamination may contain aflatoxin, a highly carcinogenic substance. We propose the use of hyperspectral imaging to quickly and noninvasively identify fungi-contaminated peanuts. The spectral data and spatial information of hyperspectral images were exploited to improve identification accuracy. In addition, successive projection was adopted to select the bands sensitive to fungal contamination. Furthermore, the joint sparse representation based classification (JSRC), which considers neighboring pixels as belonging to the same class, was adopted, and the support vector machine (SVM) classifier was used for comparison. Experimental results show that JSRC outperforms SVM regarding robustness against random noise and considering pixels at the edge of the peanut kernel. The classification accuracy of JSRC reached 99.2% and 98.8% at pixel scale, at least 98.4% and 96.8% at kernel scale for two peanut varieties, retrieving more accurate and consistent results than SVM. Moreover, fungi-contaminated peanuts were correctly marked in both learning and test images.
RESUMEN
It's well-known fungi-contaminated peanuts contain potent carcinogen. Efficiently identifying and separating the contaminated can help prevent aflatoxin entering in food chain. In this study, shortwave infrared (SWIR) hyperspectral images for identifying the prepared contaminated kernels. Feature selection method of analysis of variance (ANOVA) and feature extraction method of nonparametric weighted feature extraction (NWFE) were used to concentrate spectral information into a subspace where contaminated and healthy peanuts can have favorable separability. Then, peanut pixels were classified using SVM. Moreover, image segmentation method of region growing was applied to segment the image as kernel-scale patches and meanwhile to number the kernels. The result shows that pixel-wise classification accuracies are 99.13% for breed A, 96.72% for B and 99.73% for C in learning images, and are 96.32%, 94.2% and 97.51% in validation images. Contaminated peanuts were correctly marked as aberrant kernels in both learning images and validation images.
