Nondestructive determination of potato noodle hardness using hyperspectral imaging technology*

Zhishang Ren, Guangyao Zhang, Juan Du,Xiang Yin, Chengqian Jin, 2, Chengye Ma

(1. School of Agricultural Engineeringand Food Science, Shandong University of Technology,Zibo, 255000, China; 2. Nanjing Institute of Agricultural Mechanization, Ministry of Agriculture and Rural Affairs, Nanjing, 210014, China)

Abstract: The texture is an important index that indicates the noodle quality. In this research, the relationship between hardness and texture with respect to potato noodles was demonstrated by using 108 noodle samples with different amounts of potatoes. Noodle images would be captured by hyperspectral imaging technology and textures would be analyzed during that process. A calibration and cross-validation model were established based on the least squares regression method. 72 noodles as training samples were used to establish the model and the other 36 samples were used to verify the calibration model. The coefficient of determination and values of the calibration model and cross-validation model were 0.877 and 0.842 while the root mean square error of calibration (RMSEC) and root mean square error of cross-validation (RMSECV) values were 11.405 and 13.166. The coefficient of determination of the prediction set was 0.988 and the root mean square error of prediction (RMSEP) was 3.585. Results showed that the proposed determination method of potato noodle hardness was of high robustness and stability and could be used as a non-destructive detection technology for detecting the noodle hardness.

Keywords: nondestructive determination; hyperspectral imaging; potato noodle hardness; partial least squares regression

0 Introduction

Nowadays, noodles have become one of the two main wheat products (bread and noodles) in the world. The production of noodles has been industrialized in China because it is at the forefront of Chinese fast food consumption. Therefore, the quality control of noodles is very important[1]. Noodles are one of the flour products with a large output value, accounting for about 9% of China’s flour consumption[2]. At present, many researchers mixed vegetables, fruits, and flour to make noodles to increase its nutritional content. Potatoes are native to the Andes in Peru and Bolivia in South America, and they are abundant in yield and are currently the fourth major food crop in the world after wheat, rice, and corn[3]. Potato flour is the main processed potato product[4]. It is rich in dietary fiber, protein, vitamin C, mineral potassium, etc., and has high development and utilization value due to its rich nutrients[5]. Today, many researchers have done researches on potato noodle making methods and quality inspection. Chemical methods were used to detect the composition of noodles. But chemical methods are slow, contaminative and destructive to the environment. The detection methods for noodle quality are mainly texture analyzer detection and sensory evaluation. However, the detection of texture analyzer is time-consuming, the samples was destroyed, and sensory evaluation is subjective, which can be affected by many factors.

Hyperspectral imaging (HSI) is an emerging innovative technology that integrates traditional spectral and digital imaging technologies into one system, making it possible to simultaneously provide the spectral and spatial information of an object[6]. In recent years, the application of hyperspectral imaging technology in food quality and safety evaluation has been greatly developed[7]. There has been many studies on the detection of chemical components of fruits, vegetables, meats, and grains, but there are barely any studies on the texture characteristics of foods, and no research has been reported on the texture of potato noodles. Hardness is related to the ability of starch and protein to adhere to each other[8]. In this study, the hardness of potato noodles was used as an index to investigate the possibility of applying hyperspectral imaging technology to the detection of noodle texture.

1 Materials and Methods

1.1 Preparation of potato noodles

In order to establish a method for rapid determination of noodle hardness, 108 noodle samples would be prepared. Noodles vary in hardness depending on the potato content. In this study, potato pulp and wheat flour were used as raw materials to prepare noodle samples. The potato (Atlantic potato) were purchased from a local grocery store in Zibo, China. Wheat flour was obtained from Chunhao Flour Co., Ltd. Potato noodles was prepared with the following formula: 100 g of wheat folour, 0～55 g of raw potato pulp, added water to make all noodle dough have 32% moisture. Mixed the ingredients together to form a dough and press it to make a dough sheet then cut it into noodles.

1.2 Measuring the hardness of potato noodles

A texture analyzer was used to measure the hardness of the noodles. The potato noodles were put on the stage, selected the cutting mode, the speed before testing was 2.0 mm/s, the speed was 0.17 mm/s, and the speed was 2.0 mm/s, triggering force 5 g, cutting degree 90%.

According to the texture analyzer, the hardness of 108 noodles ranges from 111.786～220.45 g.

1.3 Hyperspectral imaging system

Black box (IRCP0076-2COMB Isuzu Optical Corp., Taiwan, China) includes visible and infrared switching mechanism, base plate, support frame, transmission light source, diffuse light source mountain and hood cover; infrared light source (IRCP0078-1COMB, Isuzu Optical Corp., Taiwan, China), double-branched halogen light source (1 000～2 500 nm), 1 000～2 500 nm hyperspectral imager (N25E-SWIR, Specim, Finland), wavelength range: 1 000～2 500 nm; a displacement platform (IRCP0076-1COMB, Isuzu Optical Corp., Taiwan, China); spectral resolution (30 μm slit): 12 nautical miles; lens (LES30, Specim, Finland), focal length: 30.7 mm, C-Mount, 900～2 500 nm), acquisition and analysis software (IRCP0072-1COMB, Isuzu Optical, Taiwan, China), spectral imaging system software, image processor (Dell, USA): 4G memory and 1t hard disk.

1.4 Image acquisition, correction and extraction

The study required the researchers to put the potato noodles in a petri dish and then place the noodles on the moving platform. Under the camera and under the light source of the halogen lamp, spectral images could be obtained by setting the moving speed and exposure time to control the platform movement using the software. In order to improve the accuracy of the hyperspectral data and reduce the effect of changes in light intensity on the hyperspectral data, the original collected hyperspectral image was corrected to obtain the spectral relative reflection information. The calculation formula is as follows[9]

(1)

Where:RTis the spectral reflectance;Irawis the original hyperspectral image;Idarkis the blackboard calibration image;Iwhiteis the whiteboard calibration image.

Each acquired image is a three-dimensional (3-D) data cube, which contains two-dimensional images arranged in sequence at different wavelengths[10]. A hyperspectral image can be viewed as the spectrumI(l) of each individual pixel (x,y), or as an imageI(x,y) of each individual wavelengthl[11]. Using the spectrum analysis software of Isuzu Optical Instruments to select the scanned image of each sample, the spectral data of absorption intensity and wavelength can be obtained automatically. The text file of the spectrum was converted with origin mapping software to obtain a curve image with the wavelength on the horizontal axis and Absorption intensity on the vertical axis.

2 Chemometric analysis

Many spectral-based technologies such as partial least squares (PLS), principal component analysis (PCA), neural network (ANN), and spectral angle mapper, can be used to process massive hyperspectral data sets[12]. In this study, the partial least squares regression (PLSR) method was used to predict the hardness of potato noodles using near-infrared spectral information extracted from hyperspectral images of samples. PLSR is a classic nonlinear supervised regression method, which has been commonly used to establish empirical correction models and prediction models in hyperspectral experiments[13]. A total of 108 samples were used in this study, 72 samples were used to correct the model, and the other 36 samples were used to verify the model. This process was performed in Unscrambler 10. 4.

3 Model validation and evaluation

Model validation is an important step in the multivariate data analysis. This experiment uses full-wavelength data for modeling analysis. The leave-one-out method is being used to select the optimal number of principal components of the PLSR model. The correction set model determines the coefficientR2Cand the correction set mean square. The root error RMSEC, the validation set model determination coefficientR2cv, and the root mean square error RMSECV of the validation set are used as evaluation indicators. Then, the purpose of the prediction capability of the correction model is to establish a PLSR prediction set. It’s being done by comparing the actual measured value with the predicted value. In general, a good model should have highR2C,R2cv, andR2P, as well as low RMSEC, RMSECV, and RMSEP, and the slight differences between them[14-15].

4 Results and discussion

4.1 Spectral characteristics

Figure 1 presents the spectral reflectance curve of potato noodles showing the average reflection spectra extracted from different noodles.

Fig. 1 Average spectral data of the noodles hardness in 1 000～2 300 nm

The wavelength range of 1 000～2 300 nm has been selected as the research focus. The spectral wavelength is 1 000～2 300 nm, and the reflectivity is 0.2～1.0. Absorption peaks around 1 200 and 1 940 nm may be associated with the O-H bond, and absorption peaks around 1 450 nm may be associated with the first overtone of O-H[16-17]. The O-H bond may be related to the water-locking ability of proteins and starch. Hardness were likely related to protein, starch, and color differences[18]. In the wavelength range of 1 000～2 300 nm, the spectral reflectance of all samples showed a similar trend, but there were some differences between the spectra of various samples. The kind of noodles’ ingredients resulted in the change of spectral. The unevenness of the sample surface structure and the uneven scattering of the sample surface also affect the spectral reflectance[19]. When the wavelength is more than 2 100 nm, the curve shows an irregular trend, which is caused by excessive energy and reduced noise reduction during the scanning of the band[20].

Tab. 1 Calibration and prediction results of noodles hardness by using full spectral range

4.2 Full spectrum model correction

(a) calibration model scatter plot

(b) cross-validation model scatter

(c) predict model scatter plot

The study plots the experimental measurements and the predictions from their calibration and prediction sets, as shown in Figure 2. The prediction model hadR2of 0.988 and RMSEP of 3.585. The results show that the method has better prediction accuracy for noodle hardness. In general, whenR2is greater than 0.80, the prediction accuracy of the regression model is better[21-25]. In addition, when the established PLSR model is accurate, the results show that this method has strong prediction ability and strong stability, and it can be used as a non-destructive testing technology for detecting the hardness of noodles.

5 Conclusions

Hyperspectral imaging technology is feasible in terms of predicting the hardness of noodles. The full-wavelength correction model and cross-validation model have a highR2, and the prediction model has a very highR2, indicating that the model has a better prediction ability. Hyperspectral imaging technology is a fast and non-destructive detection method, which has great application potential in detection of the noodle quality.