In this study, a 400-1000nm hyperspectral camera can be used, and FS13, a product of Hangzhou CHNSpec Technology Co., Ltd, can be used for related research. The spectral range is 400-1000nm, and the wavelength resolution is better than 2.5nm, up to 1200
Two spectral channels. The acquisition speed can reach 128FPS in the full spectrum, and the maximum after band selection is 3300Hz (support multi-region band selection).
Sorghum is one of the important food crops in China, because of its rich nutrients in the grain in the wine industry has a "good wine can not be separated from red grain" incisive judgment, the annual demand of up to 20 million t. At present, the main varieties of wine sorghum are Luzhou Red, Qinghuyang, Runuo No. 7 and other glutinous sorghum with high starch content. Because there are many kinds of sorghum and different producing areas, the contents of starch, protein, fat and tannin in the grain are very different, which leads to great differences in flavor, style, quality and yield of liquor. It can be seen that the accurate and efficient identification of sorghum varieties before the batch storage of sorghum raw materials has a very important guiding significance for the production of high-quality liquor, which can control the production process such as the time of bubbling grain, water consumption and steaming grain during the brewing process. The traditional identification methods mainly include manual empirical identification and biological sampling detection. The former is subject to subjective influence, low efficiency, and difficult to form a unified standard, while the latter is cumbersome, time-consuming and laborious. Both of them cannot meet the needs of modern liquor enterprises for identifying sorghum, so it is urgent to find a fast, accurate and simple sorghum variety classification and detection method. The objective of this study is to classify 11 sorghum varieties by combining spectral information and image information, and identify different sorghum varieties by optimizing hyperspectral technology and machine learning methods through comparison and external verification, so as to improve their accuracy and efficiency in application.
The original spectral curves of 550 samples of 11 categories of sorghum and the spectral curves after MSC pretreatment are shown in Figure 1. Each color represents a different category.
In this paper, the identification of 11 varieties of sorghum was studied based on the combination of hyperspectral spectrum and image information. The hyperspectral images of sorghum were collected, 48 feature wavelengths were selected from the spectra after MSC pre-processing by SPA algorithm, and then the texture features of the images were extracted. SVM, PLS-DA and ELM classification models were established based on the texture features, full spectrum, feature spectrum and their combined image information, respectively. Finally, the data not involved in modeling was used for external verification. The results show that the SVM classification model based on the combination of feature spectrum and texture features has the best effect. The correct recognition rate of the test set is 95.3%, and the accuracy of the verification set is 91.8%. The combination of visible spectrum and image can effectively realize the rapid recognition of wine sorghum and improve the recognition accuracy of the model. This provides a theoretical basis for the detection of different brewing raw materials and the realization of brewing automation.