Purity analysis of multi-grain rice seeds with non-destructive visible and near-infrared spectroscopy

Highlights

• Purity analysis of multi-grain rice seeds.

• Rapid and non-destructive detection technique.

• Visible and near-infrared spectroscopy.

• Equidistant wavelength combination screening.

• Wavelength step-by-step phase-out.

Abstract

Seed purity is a crucial indicator of seed breeding, production and circulation. The traditional purity analysis methods are based on the authenticity identification of single-seed, which are complex, time consuming and lowly efficient. Using visible and near-infrared (Vis-NIR) spectroscopy, a novel purity analysis method of multi-grain rice seeds was developed. The samples were mixed samples that the target variety (Y Liangyou 900) was contaminated by the other four varieties. The standard normal variate method was used for spectral pretreatment. The equidistant combination-partial least squares (EC-PLS) was adopted for large-range wavelength screening. The wavelength step-by-step phase-out PLS (WSP-PLS) was further used to eliminate interference wavelengths and improve predicted effect. In vis-short NIR region, the selected WSP-PLS model included 19 non-equidistant wavelengths. The root mean square errors and correlation coefficients for prediction in modeling (RMSEP_M, R_P,M) reached 0.115, 0.920, respectively. In validation, the root mean square errors and correlation coefficients for prediction (RMSEP_V, R_P,V) were 0.152, 0.845, respectively. In the long NIR region, the selected model included 24 non-equidistant wavelengths, the RMSEP_M, R_P,M were 0.103, 0.930 and the RMSEP_V, R_P,V were 0.129, 0.894, respectively. Results showed that the predicted and actual purity values had high correlation, and the wavelength model complexity was low. The proposed Vis-NIR detection method has the feasibility to analyse the purity of multi-grain rice seeds, which is a rapid, non-destructive and promising analytical technique. As a simple and efficient wavelength selection method, WSP-PLS is expected to be used for more analysis objects.

Introduction

Rice is one of the three major cereal crops worldwide. The mixing of different seeds during rice planting, whatever naturally or artificially caused, seriously affects the yield and quality of rice planting. The Chinese government has formulated relevant national standards for regulation (GB/T 3543.5-1995, 1995). Thus, seed purity analyses are highly necessary.

The traditional analysis methods of rice seed purity are actually based on the genuineness identification of single-seed, which have very low efficiency and poor representation. And the percentage of the sample number of the tested seed variety to the total number of samples is then calculated, that is, the purity is obtained. Traditional methods for rice-seed authenticity include artificial methods, such as field-planting (Chen et al., 2011a), morphological (Hussain et al., 2010), and seedling methods (Sasaki, 2004). All of these methods require field planting, i.e., a long period of time and significant lags. Another type of method is molecular instrumental analysis, including high-performance liquid chromatography (The ministry of agriculture of the People's Republic of China, 1985a, The ministry of agriculture of the People's Republic of China, 1985b, The ministry of agriculture of the People's Republic of China, 2008), protein thin-layer electrophoresis technology (Zhao et al., 2005), DNA molecular marker identification technology, and others (Kretzschmar et al., 2018, Liu et al., 2008). These methods are performed to determine for proteins, fats, moisture, amino acids of single seed. For different types of seeds, serious overlaps exist in the range of the above components. Thus even need to determine the structure and composition of seeds DNA. The measurements require experimental equipment that is complicated, expensive, and highly specialized. Indeed, the abovementioned methods are inconvenient for large-scale agricultural-breeding applications. Thus, a simple, rapid, and effective method of rice seed purity analysis is significant to develop.

Near-infrared (NIR) spectrum primarily reflects vibration absorption with overtones and combination frequencies for hydrogen-containing functional groups, such as CH, OH and NH. NIR absorption strength is weak, so most sample types can be measured directly without preprocessing. This technique has the obvious advantages of being rapid, real time, and nondestructive and has been successfully applied to soil (Chen et al., 2011b, Cozzolino and Moron, 2006, Pan et al., 2014, Rossel et al., 2006), agricultural products and food (Chen et al., 2006, Guo et al., 2014, Liu et al., 2013, Lyu et al., 2016, Wang et al., 2015, Wang et al., 2016), environment (Pan et al., 2012, Sousa et al., 2007), biomedicine (Du et al., 2004, Han et al., 2015, Jiang et al., 2002, Pan et al., 2013, Xie et al., 2010, Yao et al., 2016) and other fields.

The main components of rice seeds (e.g., moisture, starch, fat, protein, and various amino acids) contain a large amount of hydrogen-containing groups (X–H) with NIR absorption. NIR spectroscopy is useful for the quantitative analysis of the main components of rice seeds, such as starch, fat, and protein (Peng et al., 2017, Wang et al., 2006, Zhu et al., 2015). Due to the different rice seed varieties, the compositions and contents of starch, fat and protein are diverse, thereby producing different NIR absorption. Thus, at the molecular level, it is expected to use the characteristics of NIR spectroscopy to discriminate and analyse different rice seed varieties (Wang et al., 2015). Further, by optimizing the wavelength models, the characteristic wavelengths are extracted, which may be used to predict the purity of the target seed in multi-grain way. However, as far as we know, attempts and researches in this area have not been carried out yet.

Rice seeds are a complex system with multiple components. Thus, the NIR diffuse reflectance spectra of multi-grain seeds obtained by direct measurement are used for purity analysis, and the methodological difficulties are necessary to overcome. Standard normal variate (SNV) is a common spectral preprocessing method. It associates spectral changes with the component concentrations, increasing the difference between spectra, thereby improving the robustness and prediction ability of the models (Barnes et al., 1989, Dhanoa et al., 1994). In a previous study (Han et al., 2008), the SNV method was successfully applied to the vigor test of oat seeds through NIR diffuse-reflectance spectra. In the present work, the commonly used SNV method was also adopted for the spectral pretreatment of rice seeds.

Further, the optimization of the wavelength model is necessary to extracted the spectral information related to seed purity. Based on the PLS regression, the moving-window PLS (MW-PLS) can achieve the modeling of all wavebands by two cyclic parameters (the initial wavelength and number of wavelengths), and has successfully adopted in many applications (Chen et al., 2011b, Du et al., 2004, Han et al., 2008, Jiang et al., 2002, Pan et al., 2012, Yao et al., 2016).

The recently proposed equidistant combination-PLS (EC-PLS) method can achieve the modelling of all combination of equidistant wavelengths by three cyclic parameters (the initial wavelength, number of wavelengths and number of wavelength gaps) (Chen et al., 2018, Han et al., 2015, Han et al., 2018, Pan et al., 2014, Yao et al., 2016, Yao et al., 2017). Moreover, it covers MW-PLS in terms of algorithm. In this study, the EC-PLS method was used to optimize the wavelength model for NIR analysis of purity.

Considering that the wavelength selection of equidistant combination is difficult to avoid the interference wavelengths, the wavelength step-by-step phase-out combined with PLS regression (WSP-PLS) was adopted to further eliminate interference wavelengths in the EC-PLS models.

WSP-PLS method can be used to correct any one wavelength model (continuous, discrete) with n wavelengths. It eliminated the wavelengths as following steps. First, the wavelengths backward elimination: each time eliminated the wavelength, whose removing resulted in the lowest prediction error, until only one wavelength remained. Then, the optimal model was selected from the above-mentioned process of wavelengths elimination by step-by-step phase-out mode.

In the present study, the EC-PLS and WSP-PLS methods were applied for the wavelength selection in the Vis-NIR purity analysis of multi-grain rice seed. Experiments confirmed the feasibility of Vis-NIR spectroscopy for the purity analysis of multi-grain rice seeds.

The purity analysis based on the detection of multi-grain seeds was performed using the following experimental design. A type of excellent hybrid rice variety was used as the target sample for purity analysis, and four other rice varieties were used as interference samples. The interference samples were independently mixed into the target sample in various proportions. The Vis-NIR spectra of the mixed multi-grain seed samples were directly measured for purity analysis.