Premium
Nondestructive identification of barley seeds varieties using hyperspectral data from two sides of barley seeds
Author(s) -
Sun Heng,
Zhang Liu,
Li Hao,
Rao Zhenhong,
Ji Haiyan
Publication year - 2021
Publication title -
journal of food process engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.507
H-Index - 45
eISSN - 1745-4530
pISSN - 0145-8876
DOI - 10.1111/jfpe.13769
Subject(s) - hyperspectral imaging , pattern recognition (psychology) , linear discriminant analysis , preprocessor , data pre processing , artificial intelligence , mathematics , data set , feature selection , computer science , support vector machine , multivariate statistics , set (abstract data type) , statistics , programming language
The identification of seed varieties is of significance to ensure seed purity. In this study, 1,600 barley seeds (400 per variety) were selected as the samples. By using a near‐infrared hyperspectral imaging system, the hyperspectral cube data on the reverse and ventral sides of each barley seed were obtained. Three preprocessing methods (standard normal variate, multivariate scatter correction, and Savitzky–Golay first derivative) and three discriminant models (k‐nearest neighbors, support vector machine [SVM], and random forest) were selected for preprocessing and modeling, respectively. The results indicated that the reverse spectral data preprocessed by multivariate scatter correction and the ventral spectral data preprocessed by Savitzky–Golay first derivative had the best modeling results. In order to further simplify the model, feature selection algorithm (successive projections algorithm) was used for feature wavelength selection. After multivariate data analysis, the combination of k‐nearest neighbors and successive projections algorithm based on the reverse spectral data was selected as the best discriminant model. The accuracy of the calibration set and the prediction set were 95.52% and 93.71%, and the Kappa value were 0.9424 and 0.9156, respectively. The reverse spectral data of barley seeds of the validation set were used to verify the accuracy of the discriminant model and visualize the classification results. Finally, considering the practical application, the mixed spectral data containing different ratios of the reverse spectral data was analyzed, and satisfactory results were obtained. It can be concluded from this study that hyperspectral imaging technology was a very promising method for identification of barley seed varieties. Practical applications With the circulation of the seed market, the traditional methods used for variety identification can no longer meet the needs of large‐scale, high‐quality production. As a nondestructive determination method, hyperspectral imaging technology combined machine vision and near‐infrared spectroscopy. In this study, hyperspectral imaging technology was used to identify barley seed varieties, and satisfactory results were obtained. It can provide a theoretical reference for future multispectral online detection technology.