Premium
Detection of viability of soybean seed based on fluorescence hyperspectra and CARS‐SVM‐AdaBoost model
Author(s) -
Li Yating,
Sun Jun,
Wu Xiaohong,
Chen Quansheng,
Lu Bing,
Dai Chunxia
Publication year - 2019
Publication title -
journal of food processing and preservation
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.511
H-Index - 48
eISSN - 1745-4549
pISSN - 0145-8892
DOI - 10.1111/jfpp.14238
Subject(s) - hyperspectral imaging , adaboost , support vector machine , artificial intelligence , computer science , boosting (machine learning) , pattern recognition (psychology) , machine learning
In this study, the feasibility of the fluorescence hyperspectral imaging (FHSI) technology to detect the viability of soybean seeds was investigated. Viable and nonviable seed samples were obtained by artificial aging method. Hyperspectral images of samples were collected by the FHSI device and then the spectral data were collected. Characteristic wavelengths were respectively selected by three variable selection methods, eliminating a large number of redundant information irrelevant to the viability of soybean seeds. Support vector machine (SVM) models based on the full spectra and the optimal spectral data were developed to identify the viability of soybean seeds. To further improve the accuracy of the model, the adaptive boosting (AdaBoost) algorithm was used. The results showed that the accuracy of the calibration and validation sets in the CARS‐SVM‐AdaBoost model (22 characteristic wavelengths) reached 100%, indicating that the combination of FHSI technology and the optimization model can greatly improve the recognition accuracy. Practical applications A rapid and accurate nondestructive identification method of viability of soybean seeds can contribute to the construction of the online seed viability detection system. FHSI technology has the advantages of high sensitivity and comprehensive analysis of sample information. Combined with the optimization model proposed in this paper, the recognition accuracy can be greatly improved. It can be applied to the online seed viability detection by seed companies, seed quality inspection departments, and soybean breeding units.