X‐ray computed tomography for quality inspection of agricultural products: A review

Abstract The quality of agricultural products relates to the internal structure, which has long been a matter of interest in agricultural scientists. However, inspection methods of the opaque nature of internal information on agricultural products are usually destructive and require sample separation or preparation. X‐ray computed tomography (X‐ray CT) technology is one of the important nondestructive testing (NDT) technologies without sample separation and preparation. In this study, X‐ray CT technology is used to obtain two‐dimensional slice images and three‐dimensional tomographic images of samples. The purpose of the review was to provide an overview of the working principle of X‐ray CT technology, image processing, and analysis. This review aims to focus on the development of the agricultural products (e.g., wheat, maize, rice, apple, beef) and its applications (e.g., internal quality evaluation, microstructure observation, mechanical property measurement, and others) using CT scanner. This paper covers the aspects regarding the advantages and disadvantages of NDT technology, especially the unique advantages and limitations of X‐ray CT technology on the quality inspection of agricultural products. Future prospects of X‐ray CT technology are also put forward to become indispensable to the quality evaluation and product development on agricultural products.

Interest in CT technology utilized in the agricultural field led to rapid interest in the visualization of inner dynamic conditions among agricultural researchers. The development of X-ray CT technology is transformed from external quality inspection to internal quality inspection. This technology is encouraged to inspect and visualize the internal information on two-dimensional (2D) structure and threedimensional (3D) structure of agricultural products (Fox & Manley, 2009;Girvin & Gupta, 2006;Guelpa, Plessis, & Manley, 2016). The pore topology, tiller numbers, internal corruption, grain hardness, moisture, temperature, volumes and densities, and mechanical damage could be achieved with CT scanner (Arendse, Fawole, Magwaza, & Opara, 2016;Chen, Xu, Yin, & Tang, 2017;Dhondt, Vanhaeren, Loo, Cnudde, & Inze, 2010;Longuetaud, Leban, Mothe, Kerrien, & Berger, 2004;Stuppy, Maisano, Colbert, Rudall, & Rowe, 2003;Yang et al., 2011;Yu & Qi, 2007). X-ray CT technology is becoming a routine tool for the nondestructive 3D characterization and the reconstruction of tomographic imaging. Takhar et al. analyzed the transport routes and transport mechanisms of moisture on maize grains using micro-CT scanner to solve the stress equation for predicting the drying parameters at the intermittent drying conditions. And the experimental average moisture content could be expected by the model with reasonable accuracy of R 2 = .88-.99 and CV = 3.5%-9.5% (Takhar, Maier, Campanella, & Chen, 2011). X-ray CT technology was utilized to observe the formation of stress cracks in maize grain that could be caused by high temperature and excess moisture (Carvalho, Aelst, Eck, & Hoekstra, 1999). The quality information of agricultural products could be inspected nondestructively by X-ray CT technology.
In this paper, we are focused on the X-ray CT technology for the quantitative and qualitative analysis of internal structure and components with cereals (wheat, rice, and maize), fruit (apple and pear), and meat (beef and lamb). This review demonstrates that the recent progress of X-ray CT as a nondestructive technology for 2D and 3D investigations related to agricultural products. The theory and systems of X-ray CT technology are briefly reviewed. The core content discusses the previous research and applications of agricultural products, unique advantages, and limitations of NDT technology on the quality inspection of agricultural products. In addition, the future trends of X-ray CT technology are discussed for the quality inspection and product development on agricultural products.

| Basic principle
The first prototype of the CT scanner was installed to examine the first patient in the Atkinson Morley Hospital. A typical CT equipment consists of three parts: scanning part, computer system, and image system. The scanning part is made up of the X-ray source, detector, and gantry. And the data, which are collected by the scanning part, are stored on the computer system. Besides, image system displays the computer-processed and reconstructed images on a screen (Neethirajan, Jayas, & Ndg, 2007;Wang & Wang, 2001).
The core principle of X-ray CT technology, which was introduced in 1917 by Johann Radon, was Radon transform. Based on the attenuation of a transmitted X-ray beam, X-ray CT technology has different mass attenuation characteristics with a consequence of the different materials comprising samples (Kays, Barton, & Windham, 2000a;Kotwaliwale, Weckler, & Brusewitz, 2006;Kotwaliwale et al., 2014;Tretiak & Metz, 1980). According to the decay characteristics of X-ray on the material, the working principle of X-ray CT technology is to obtain the radiation attenuation information when X-ray passes through the sample at different directions. The decay characteristics of different materials represent different selected X-ray energies. The X-ray energy of medical radiation is generally 140 keV, while that of industrial CT is selected as 50 keV-16 MeV. In the inspection of agricultural products, the selected energy is usually as variable as variety, and in some cases more so.
The traditional CT is an imaging procedure where an X-ray tube rotates around a sample or samples and then the attenuation is recorded on a detector. The focus of X-ray tube limits the spatial resolution, while actual resolution depends on the product size and its magnification times. Some radial projections are captured at different angles on a sample by the X-ray, usually completing a 180-degree rotation. Generally, a charge-coupled device camera is applied as an enlarged radiography. Afterward, it is reasonable to obtain 2D slice images from different angle projections and then to superimpose the sequence of 2D slice images for obtaining 3D volume (Cosmi & Bernasconi, 2013;Wang, 2004). The formation processes of CT images include image scan stage, reconstruction stage, and display stage. For CT images, voxel and pixel are the basic structural unit of an image and the basic unit of the image on a screen, respectively.
The image clarity depends on the voxel and pixel. Besides, a 3D image could be produced by the restructure of scanned numerous slice images of the sample (Mooney, Pridmore, Helliwell, & Bennett, 2012;Stock, 2008;Zhuang, 1992), as shown in Figure 1.
X-rays are both invisible light and electromagnetic waves. X-ray is produced by a speeding electron impact on a sample, where the energy of X-ray photon is equal to the total kinetic energy of the electron. X-ray photon energy E could be defined as follows: where h is Planck constant, Js; c is light velocity, m/s; and λ is X-ray wavelength.
Usually, the unit of X-ray photon energy E is eV, and 1 eV = 1.62 × 10 -34 J. h and c are approximately equal to 6.63 × 10 -34 Js and 3 × 108 m/s, respectively. Most X-rays have a wavelength ranging from 0.1 mm to 0.01 nm corresponding to energies in the range of 10 eV to 450 keV. Photons in a soft X-ray beam, when these pass through a sample, are being absorbed, transmitted, or scattered. And in general, the higher the material density and atomic number of constituents are, the greater the X-ray absorption is. As a result, the attenuation intensity is indicated by the Beer-Lambert law (Curry, Dowdey, & Murry, 1990;Wu, Yang, & Zhou, 2008;Ying & Han, 2005).
where I is the intensity of X-ray exiting through a sample in eV; I 0 is the intensity of X-ray in eV; μ n is the linear attenuation coefficient of a sample on the wavelength in eV/mm; and l is the path length through a sample in mm.
In CT system, the material difference of physical density could be visualized by the changes of image intensity. And X-ray attenuation capabilities seem to reflect the difference. Moreover, these capabilities on voxel could be represented as CT number in a 3D image. The CT number is the attenuation coefficient which is calculated using water, and its unit is Hounsfield (Hu) (Cnudde et al., 2006;Donis-Gonzalez, Guyer, Pease, & Barthel, 2014;Taina, Heck, & Elliot, 2008). The calculation formula of CT number is as follows: where μ is the linear attenuation coefficient of a sample, and μ w is the linear attenuation coefficient of water (approximately 0.195). (1) CT number = − w w × 1000, F I G U R E 1 Working principle of X-ray CT technology. Some radial projections are captured at different angles on a sample by the X-ray to obtain 2D slice images. Besides, a 3D image could be produced by the restructure of scanned numerous slice images of the sample The measurement range of CT number is from -1,000 Hu to + 4,000 Hu. Besides, CT number for air is -1,000 Hu and the number for water is 0 Hu (Ogawa, 1998).

| Image processing and analysis
The internal information of agricultural products includes the microstructural information (cell size, cell wall size, pores number, and porosity), and physical, chemical, and mechanical properties (Jha, Kingsly, & Chopra, 2006;Schoeman, Williams, Plessis, & Manley, 2016). The image processing is required to extract the suitable information from images and to obtain CT number accurately. Then, the accurate visualization and inner component identification could be carried out using analysis of CT images. By combining with the image processing technology, image analysis could provide the information of the microstructural variations on agricultural products (Leonard, Blacher, Nimmol, & Devahastin, 2008).
In order to distinguish phase materials (air and samples), the image segmentation is used to separate the voxel group of region of interests (ROIs). The image processing technology is usually performed with threshold segmentation. The textural features of CT images could be extracted using histogram groups, histogram, and shape moments to identify the ROIs on the sample (Karunakaran, Jayas, & White, 2003a, 2003b. And some small quantities of pixels could be removed by a cleaning step (Baker et al., 2012). The random noise, detector defect, and artifact defect appear in 2D slice images after image segmentation. For reducing the noise and correcting the defects, the median filter, Gaussian filter, and noise reduction technology are used (Frisullo, Laverse, Marino, & Nobile, 2009;Shahin, Tollner, & Prussia, 1999). According to the matrix arrangement, CT images are composed of different grayscale pixels from black to white. Hence, the area of high-attenuation regions (high Hu) is brighter and the area of low attenuation (negative Hu) is darker.
Because of the varying distribution in density regions of a sample, the internal structure of agricultural products could be visualized and analyzed through 3D tomographic images of volume rendering.
The image analysis is used to obtain the quantitative and qualitative information on the sample. Currently, some software, such as ImageJ, Avizo, Amira, and VGStudio Max, are usually used to analyze the quantitative, qualitative, and statistical data for sample structure (Mao, Kumi, Li, & Han, 2016). Image processing and analysis proce-

| X-R AY C T APPLI C ATI ON ON AG RICULTUR AL PRODUC TS
In recent years, X-ray CT technology has become a useful tool to assess the quality of agricultural products, enabling a better understanding of the composition, physicochemical characteristic, and internal structure on samples. More specifically, an extensive range of agricultural products such as cereal (wheat, corn, and rice), fruit (apple and pear), and meat (beef and lamb) could be analyzed by Xray CT technology with the high-resolution 2D and 3D visualization.
And the applications include internal quality evaluation, microstructure observation, mechanical property measurement, and others (Schoeman et al., 2016). The 2D structure of agricultural products could be observed with 2D slice images using CT scanner. Besides, the 3D model, which merges numerous 2D slice images, could be established (Trinh, Lowe, Campbell, Withers, & Martin, 2013). An overview of CT applications related to various agricultural products is shown in Table 1.

| Internal quality evaluation
At present, X-ray CT technology has been applied in the internal quality evaluation of agricultural products such as maturity, sugar, acidity, oil content, and tissue breakdown. The CT number of slice images is an important index to evaluate the internal quality. CT number followed a positive linear relationship with moisture con- F I G U R E 2 Image processing and analysis procedure. 2D slice images are generated to reconstruct the 3D model. Gaussian or median filter is used to reduce noise with 3D raw gray-level images. The threshold segmentation is used to segment the image based on the gray value histogram of different regions. The final step is the qualitative and quantitative analysis on CT data of ROIs TA B L E 1 Previous study on the use of CT in agricultural products. An overview of quality inspection of various agricultural products using X-ray CT is shown. The quality inspection includes internal quality evaluation, microstructure observation, mechanical property measurement, and others. The agricultural products are wheat, corn, rice, apple, pear, beef, and others  The bitter pit is an important physiological disorder with roundish brown lesions, resulting in serious economic losses for apples. To identify the internal symptoms of apples, X-ray CT technology and image processing algorithms were used by Si and Sankaran (2016) to acquire the qualitative and quantitative data from the sample with bitter pit. In the same way, Honeycrisp apples with bitter pit before and after harvesting could be identified with CT slice images. And the classification accuracy of bitter pit and healthy fruit ranged between 70% and 96% (Jarolmasjed, Espinoza, Sankaran, & Khot, 2016). Moreover, X-ray CT technology is also applied for the quantification of salt concentrations on meat (Vestergaard, Risum, & Adler-Nissen, 2004).

| Microstructure observation
The microstructure of agricultural products could determine their physical texture and sensory properties, which could be observed  Figure 4). In X-ray CT system, the X-ray tube had a spot size of 5 mm with operating voltage of 46 kV and current of 75 μA.
The high-resolution 2D microstructure and 3D structural characterization of rice grain could be provided by SEM and CT images, respectively (Zhu et al., 2012). Spatial distribution of core breakdown disorder symptoms in pears could be analyzed by a useful technology which included the magnetic resonance imaging (MRI) and X-ray CT technology (Lammertyn et al., 2003).
Due to the water demand of starch mass and the porous microstructure of rice grains, Mohoric et al. (2009) found mesostructure of grains had a positive correlation with its microstructure.
Roast has a direct impact on the microstructure of agricultural products. Schoeman et al. (2016) investigated the roasted wheat grains using CT images to find that the volume, porosity, and expansion ratio gradually increased with the decreased of relative density. Furthermore, Mebatsion et al. (2009) used synchrotron Xray CT scanner and transmission electron microscopy to establish 3D models with three important components of fruit cortex tissue, cell wall, pore network, and cells. A more study by Schulze, Peth, Hubbermann, and Schwarz (2012) found that the differences of pore size distribution, porosity, stiffness, and cell morphology in the tissue of apples had a strong influence on enrichment process. X-ray F I G U R E 4 Spatial distribution of density within kernels by X-ray CT. WTR is wild-type rice; HAR is high-amylose rice (Zhu et al., 2012). The instruments are SEM with an accelerating voltage of 30 kV and a high-resolution X-ray CT system with operating voltage of 46 kV and current of 75 μA. The amount of void was the least near the tip part of the WTR grain and varied throughout grain. Within the HAR grain, the part near the tip also had the least amount of void, but no difference was found in other parts of the grain CT technology makes an inspection on the microlevel, which is beneficial to the agricultural products because the transport properties of tissues, cellular structure, dimension, and distribution of pores could lead to the improvement of physical and sensory properties (Chevallier, Reguerre, Bail, & Valle, 2014;Vicent, Verboven, Ndoye, Alvarez, & Nicolai, 2017).

| Mechanical property measurement
The tiller number is one of the most essential agronomic traits and mechanical properties, which determines the wheat and rice yield (Li et al., 2003;Xing & Zhang, 2010). The 3D structure of tillers could be reconstructed by the algorithms of filtered back-projection and graphics processing unit with CT images (Wu, Yang, Niu, & Huang, 2017 (Arendse et al., 2016;Jiang et al., 2012). Furthermore, the nondestructive inspection of morphological characteristic parameters, such as tiller number, tiller angle, tiller stem thickness, and tiller wall thickness, could be achieved by X-ray CT technology.  . The images were acquired by X-ray CT at 17 kV potential, 65 μA current, and a resolution of 60 pixels/mm. Nonvitreous kernels have more optically dense regions than the translucent vitreous kernels F I G U R E 6 Comparison of a longitudinal digital image and CT image slice. (a) A longitudinal digital image; (b) 2D CT image slice (Guelpa et al., 2015). The voxel size, voltage, and scan time are 13.4 μm, 60 kV, and 30 min, respectively. The same maize grain is depicting the internal structure of the maize grain, that is, flour and vitreous endosperm, germ, and pedicle. In CT image, the brighter gray region represents the denser vitreous endosperm and the darker region the less dense floury endosperm. The vitreous endosperm thus appeared translucent (Figure a) due to no light being reflected In addition, the physical and mechanical properties of agricultural products also include color, size, texture, shape, density, and hardness (Witek et al., 2010). However, some features have different types of values based on the different measuring way. And the inspection on physical and mechanical properties of agricultural products becomes real challenges without a standard method (Dhondt et al., 2010). Currently, X-ray CT technology has the great potential to estimate and measure the density and porosity in comparison with traditional technologies (Kelkar, Stella, Boushey, & Okos, 2011).
Neethirajan et al. found that the wheat grains with different strains could be represented with the differences of starch granule shapes and pore shapes. The images were acquired by X-ray CT at 17 kV potential, 65 μA current, and a resolution of 60 pixels/mm. Figure 5 shows that the vitreous and nonvitreous of durum grains could be identified and classified by X-ray CT technology and transmitted light system (Neethirajan, Karunakaran, Symons, & Jayas, 2006). In order to determine the pore interconnectivity, strut size, porosity, and pore size as well as overall 3D micro-architecture, Darling and Sun reconstructed 3D models with CT-scanned serial images. The micro-CT scanner had a focal spot size of several microns with operating at 100 kV and 19.1-μm resolution (Darling & Sun, 2004). The potential function of X-ray CT technology was studied by Gustin et al. (2013) to determine the volume and density of maize grains. Figure 6 displays the internal structure of maize grains by CT scanner at voltage of 60 kV and voxel size of 13.4 μm. Guelpa, Plessis, Kidd, and Manley (2015) calculated the density, percentage porosity, and percentage cavity of maize grains to estimate the grain hardness from CT images. With the development of CT technology, the cost-effective and less time-consuming methodology was introduced to measure the entire volumes and densities of maize grains (Stuppy et al., 2003).
Additionally, the internal stress, interior heat, and mass transfer are the cause of the variations in physical and mechanical properties on agricultural products, which could be viewed on CT images.

| Others
From the previous research, X-ray CT technology is rapidly becoming a very useful technology for the measurement and analysis of internal structure on agricultural products, such as fat content, fat distribution, fat microstructure, extruded starches, lean meat content, lean weight, and meat yield (Hollo, Szucs, Tozser, Hollo, & Repa, 2007;Jay, Van, & Hopkins, 2014). To increase the meat eating quality, Lambe et al. (2017) used CT slice images of packaged lamb cuts to predict the intramuscular fat content. Based on the best CT prediction equation, the prediction level of samples was higher on average than this prediction by the taste panel. Furthermore, X-ray CT technology could be applied to visualize the ice crystal structure formation during the freezing of meat, chicken, fish, and carrot (Mousavi, Miri, Cox, & Fryer, 2007). A similar method was utilized when Haseth, Kongsro, Kohiler, Sorheim, and Egelandsdal (2008) measured the quantity of sodium chloride in ground pork and dry-cured hams with CT images at different voltages (80, 110, and 130 kV). Meanwhile, some experts focused on the soil aggregate structure of rice to increase the rice yield. Li, Zhou, Chen, Peng, and Yu (2014) found that the combined application of organic manure and chemical fertilizer could build up soil fertility to maintain the soil in good aeration for improving the yield. In practice, some CT applications have proved to predict intramuscular fat content from meat cuts of pork, beef, and lamb, and even vacuum-packaged meat (Furnols, Brun, Tous, & Gispert, 2013;Prieto et al., 2010). In addition, CT's security inspector is an important application that you can use to effectively identify organic, inorganic, and mixtures in baggage for security check in civil aviation airport, customs, and large-scale exhibition.
X-ray CT technology enables examination at the microstructural level, which is useful to the agricultural products, as the accurate calculation of the moisture content, density texture, hardness, and density could lead to improve the ability of internal quality evaluation, microstructure observation, and mechanical property measurement.
And they could be inspected with the quality feature by this technology extending to other products besides the inspection of quality feature on some agricultural products in this paper. The CT scanner is suitable to be popularized and applied in the quality inspection of products. X-ray CT technology, as a novel technology, has much literature focus on characterizing the agricultural products from scientific point of view in the laboratory and not in a commercial environment. For larger cultivated area and plant diversity of agricultural products, the portability and operability of X-ray CT equipment should be taken into account to enhance the quality inspection efficiency and accuracy. Thus, the application of this technology in agricultural production line has room for future investigation and development.

| COMPARISON OF X-R AY C T AND OTHER NDT TECHNOLOG IE S
In recent years, NDT technology is an emerging high-tech inspection technology for quality inspection of agricultural products. The properties of agricultural products, such as physical, electrical, thermal, electromagnetic, mechanical, optical, and sonic, are utilized with NDT technology to inspect the maturity degree, sugar content, sugar-acid ratio, moisture detection, fat content determination, contaminants determination, product classification, and physical inspection (Li, Rao, & Ying, 2011 (Sun et al., 2008).

NMR technology may often yield different internal information
of agricultural products compared with other NDT technologies (Shao et al., 2006;Zhao et al., 1996) (Herman, 1995). The cost of CT equipment and the corresponding system is the main question, especially for a wide variety of agricultural products with generally low prices. In addition, CT equipment only could be employed in the crosssectional scan, because of the limitations on the hardware structure (Cnudde & Boone, 2013).

| FUTURE PROS PEC TS
Based on the review of agricultural products with X-ray CT technology, the following need to develop for the future.
The scanning time of X-ray CT equipment remains a concern. A single maize kernel needed 30-min scanning time at the resolution of 13.4 μm, and while 2-3 hr were needed to obtain the resolution of 6 μm (Guelpa et al., 2015). The moisture content, mechanical damage, internal stress, and even microstructure of agricultural products are changed with scanning time, which leads to collect inaccurate data by CT equipment. Besides, the image analysis is also a time-consuming work in the use of CT technology (Schoeman et al., 2016). Development of models and algorithms further reduce the consumption time of image acquisition and analysis on the complex internal structure of agricultural products.
Different X-ray CT parameters such as the tube voltage, current, and exposure time are chosen at random, and this ultimately affects the image quality (Cnudde & Boone, 2013). In terms of quality inspection on agricultural products, the parameter setup is an important question for image quality. As low prices of agricultural products have a diverse variety of shapes, sizes, and compositions, a widely accepted parameter setup and protocol are needed to provide automatically by CT scanner.

| CON CLUS IONS
From this literature review, it is proved that X-ray CT technology is increasingly being employed as an effective tool to identify and describe the quality inspection on agricultural products. More specifically, an extensive range of agricultural products (e.g., wheat, corn, rice, apple, beef) and its applications (e.g., internal quality evaluation, microstructure observation mechanical property measurement, and others) could be investigated with a CT scanner. The popular agricultural crop on CT scan is wheat. The moisture content is the most promising quality feature that can be acquired by X-ray CT system.
The inspection of pesticide residue on agricultural products is difficult for CT scan with current development and possible will be done in.
Besides, the unique advantages of X-ray CT technology as well as its limitations, compared with other NDT technologies, are also introduced in this paper. A commercial system using portable X-ray CT with high-throughput requirements will remain a challenge to accurately inspect in the field. With the improvements in instruments and computational ability, the X-ray CT technological progress is transformed from external quality to internal quality and single-item inspection to comprehensive omnidirectional inspection. It is foreseen that the real-time and high efficient X-ray CT technology could be used to become indispensable in quality evaluation and product development of agricultural products.

ACK N OWLED G EM ENTS
The authors are grateful to the financial support by Six Talent Peaks

CO N FLI C T O F I NTE R E S T
The authors declare that they do not have any conflict of interest.

E TH I C A L A PPROVA L
This study does not involve any human or animal testing.

I N FO R M E D CO N S E NT
Written informed consent was obtained from all study participants.