In this study, a 400-1000nm hyperspectral camera was applied, and FS13, a product of Hangzhou Color Spectrum Technology Co., LTD., could be used for related research. The spectral range is 400-1000nm, the wavelength resolution is better than 2.5nm, and up to 1200 spectral channels can be reached. The acquisition speed can reach 128FPS in the full spectrum, and the maximum after band selection is 3300Hz (support multi-region band selection).
China's rice production accounts for more than 30% of the world's rice production, and "Meihe rice" in Jilin Province is a geographical indication product of China's japonica rice, and its production area is located in the world's golden grain production belt (45° N latitude). In practical life, there are many kinds of Meihe rice, and chemical methods such as Kjellod nitrogen determination and spectrophotometry are usually used to determine the protein content of different varieties of rice, but these traditional chemical methods are not only destructive to the sample itself, but also complicated steps and too long detection cycle. As a rapid non-destructive detection method, infrared spectroscopy has been widely used in the detection of the main components of rice (protein ≥, fat β, starch III, water), but it can only obtain the content of the components according to the spectral information, and can not achieve a more intuitive expression, that is, the visualization of the content. Hyperspectrum is a three-dimensional cube data including image information and spectral information. The obtained hyperspectral image contains both internal information of rice (internal physical structure and chemical composition information) and external information of rice (grain type, defects, etc.), which can make up for the lack of image that NIR can not quickly identify the spatial distribution of a certain substance. In this paper, rice of 3 varieties (Daohuahua, Akita Omachi and Jijing 60) from 4 producing areas in Meihe City, Jilin Province was selected as the research object. Hyperspectral imaging technology was used to detect the collected rice and obtain the average spectrum of the region of interest of the rice. In order to reduce the signal to noise ratio of the spectrum and obtain a relatively robust model, Three kinds of prediction models of rice protein content, including partial least square regression, principal component regression and error backpropagation neural network, were established by means of convolutional smoothing, mean centralization and multiple scattering correction. SPA was used to select the characteristic wavelength, establish the characteristic wavelength model, and transform the rice hyperspectral image into the protein content distribution map to realize the visualization of the protein content of rice from different varieties.
The feasibility of visualization of protein content distribution in rice was studied by using hyperspectral imaging technology. A simplified and efficient PLSR protein content prediction model was obtained by MC spectral pretreatment method and the selection of SPA characteristic bands. Based on the quantitative model, the protein content distribution in rice of different varieties and different origin was visualized. Due to the similar shape of rice among the same varieties, it is difficult to distinguish the rice by ordinary RGB images. Imaging the protein content distribution can provide ideas for identifying the rice origin, and comparing the protein content distribution maps of rice among different varieties can provide evidence for the later breeding of rice varieties.