Introduction

Mango (Mangifera indica) is a tropical fruit with a worldwide production of 54.8 million tons in 2020 [1]. Mango is a climacteric fruit that is normally harvested at the mature, unripe stage and further ripened for a few days before consumption. Mango is highly perishable and the eating qualities of mango during the ripening period change very quickly. Therefore, monitoring the internal quality development is important [2]. In general, determinations of internal quality of fruits are performed using various destructive testing methods which inevitably cause fruits loss during assessment and therefore complete assessment inspection cannot be conducted. Recently, several non-destructive testing methods have been receiving much attention for detecting the internal quality attributes of mango including image analysis [3], acoustic method [4], and near-infrared spectroscopy (NIRS) and hyperspectral imaging (HSI).

NIRS and HSI are rapid and non-destructive techniques for determining the internal quality attributes of fruits based on the absorption of an electromagnetic wave in the wavelength region of 700–2500 nm [5, 6]. Determination of internal quality attributes of different cultivars of mangoes during ripening using NIRS and HSI has been reported [2]. Handheld instruments with a short NIR wavelength (700–1100 nm) were frequently studied because of their low cost and convenience for in field applications [7, 8]. For example, Rungpichayapichet, Mahayothee, Nagle, Khuwijitjaru and Müller [8] developed NIRS models (700–1100 nm) to determine total soluble solids (TSS), titratable acidity (TA), firmness, and the ripening index using data collected from three years for ‘Nam Dok Mai’ mango. However, long-wavelength NIRS, which is mostly available in benchtop NIR spectrometers usually provided better prediction from the absorption peaks of various chemical components such as organic acids and sugars [9]. On the other hand, studies on HSI analysis of mango during ripening are quite limited. Most HSI studies focused on in field monitoring, especially for dry matter content [10, 11]. Rungpichayapichet et al. [12] mapped the changes in firmness, TSS, and TA during the ripening of mangoes using an HSI instrument in the wavelength range of 400–1000 nm.

A comparison of NIRS and HSI on the prediction of internal quality attributes of fruit, however, has been rarely investigated. Therefore, in the present study, NIRS and HSI were employed to determine the quality attributes of ‘Nam Dok Mai’ mango, which is a substantial high-price export variety of fresh mango from Thailand. The NIRS instrument was a benchtop type that measured the full wavelength (800–2500 nm), while the HSI measured visible and short NIR wavelength (450–998 nm).

Materials and methods

Mango samples

A total of 188 mature mango fruits (Mangifera indica cv. Nam Dokmai, subcv. Si Thong) with all-yellow peel harvested from four growing areas (Lopburi, Udonthani, Phitsanulok, and Sakeaw provinces, Thailand) were used in this study. Fruits were harvested during the season, from February to May with the fruit mass ranging 250–350 g and a specific gravity greater than 1.00. Samples were washed with water and then soaked in 200 ppm peroxyacetic acid solution with fruit to solution ratio at 1:2 (w/v) to reduce the surface microorganisms. After that, the fruits were air-dried and kept in a basket to ripen under ambient conditions (29.7 ±  0.7 °C and 66.3 ±  4.4% relative humidity) for 8 days, with the arrival day set as the first day of ripening. On each ripening day, six fruits were randomly selected for NIRS and HSI measurements as well as the reference analyses.

Non-destructive measurement

NIR spectral measurement

A Fourier transform near-infrared (FT-NIR) spectrometer (MPA, Bruker Optik, Leipzig, Germany) was used to acquire fruit spectra in reflectance mode in the wavelength range of 800–2500 nm by averaging 32 scans at 16 cm−1 resolution. Spectra were collected at 25 °C from three different fruit sections assigned as shoulder, cheek, and tip as shown in Fig. 1. Average spectra from each position and average spectra from all sections were used for calibration model development. Spectral reference was automatically collected before each measurement.

Fig. 1
figure 1

Reference analyses a, NIRS b, and HSI c measurement positions on ‘Nam Dok Mai’ mango fruit surface

Full size image

Hyperspectral image acquisition

An HSI instrument and image acquisition in this study was set up and operated as described in our previous study [12]. Spectral data in the wavelength region between 450–998 nm were extracted from hyperspectral image files using Hyper See software (BurgerMetrics SIA, Riga, Latvia). From a total of 910 × 900 pixels image, nine region of interest (ROI) areas of 5 × 5 pixels were randomly marked in each fruit section (Fig. 1). Therefore, a reflectance spectrum for each fruit section was averaged from 225 pixels and was used for calibration model development.

Chemometrics and calibration procedure

Chemometrics and calibration model development using spectral data from both devices were performed using The Unscrambler software (version 9.7, Camo, Oslo, Norway). Spectral outliers were identified using principal component analysis (PCA) subject to Hotelling’s T-squared distribution [8]. Five different spectral pretreatment techniques namely Savitzky–Golay smoothing (SG), Savitzky–Golay first derivative (D1), Savitzky–Golay second derivative (D2) using 17 points with second-order polynomial regression, standard normal variate (SNV) transformation, and multiplicative scatter correction (MSC) were performed to improve the predictability during the development of the calibration models.

Calibration models were built using partial least square (PLS) regression. Samples were divided into calibration and validation (test) sets in the ratio 2:1 (Table 1). The physicochemical values were sorted and the samples were selected by specific intervals ‘Every 2 samples counting from 3’ using the Unscrambler program and the two highest and two lowest values were assigned to the calibration set. The performance of prediction models was evaluated using coefficient of determination (\({R}_{C}^{2}\)) and standard error of calibration (SEC) and validation was evaluated based on the coefficient of prediction (\({R}_{V}^{2}\)), standard error of prediction (SEP), bias, and the ratio of the standard deviation of the test set to SEP (RPD).

Table 1 Descriptive statistics of ‘Nam Dok Mai’ mango fruit quality attributes used for calibration and validation sets in PLS modeling
Full size table

Multiple linear regression (MLR) models were developed using specific wavelengths. Candidate variables were chosen from wavelengths which showed the relatively high absolute values in PLS regression coefficient plots and loading weight plot of the best fitting PLS model. Multifactorial analysis (backward elimination stepwise method) was used for selecting significant wavelengths. MLR calibration models were constructed using the same calibration and test set samples as PLS models.

Reference analyses

Firmness

Firmness was determined on the unpeeled fruit sample only at the center area of cheek position using a texture analyzer (TA-XT2i, Stable Micro Systems, Surrey, UK) because of the difficulty of measuring other positions with the instrument. A puncture test was performed using a 2 mm diameter cylindrical probe at a speed of 0.5 mm s−1 and a compression depth of 10 mm. The maximum force (N) was recorded and represented the firmness of the fruit [13].

Chemical properties

Fruits were peeled and cut into two halves. Each half was then divided into three sections (Fig. 1). Flesh from each section was cut into small pieces and homogenized (Ultra-Turrax T25 basic, IKA, Staufen, Germany). pH value was determined using a pH meter (PHM210, Radiometer Analytical SAS, Villeurbanne, France) by placing the electrode probe into the puree. Extracted juice obtained from squeezing 5 g of puree through filter cloth was used to measure TSS by a digital refractometer (PAL-1, Atago, Tokyo, Japan) and the value was expressed as °Brix. TA was evaluated by alkaline titration with 0.1 N NaOH to a pH of 8.1 using a 100 mL dilution prepared from 5 g of puree and reported as mass percentage (%) of citric acid [14]. All measurements were done in duplicate. In addition, ripening index (RPI) was calculated from firmness (F), TSS, and TA using Eq. (1) according to Mahayothee et al. [15] to represent the ripeness of the sample.

$$\text{RPI = }ln\text{(}\frac{100\cdot \text{F}\cdot \text{TA}}{\text{TSS}}\text{)}$$
(1)

β-Carotene content

β-Carotene was extracted from flesh via a procedure adapted from Pott et al. [16]. Approximately 5 g of homogenized flesh from each mango section was mixed with 40 mL of methanol (RCI Labscan, Bangkok, Thailand). The mixture was filtered through a sintered glass funnel. The residue was washed with methanol until colorless. The yellow-clear solution was liquid-liquid extracted with 20 mL of a mixture of acetone (RCI Labscan) and hexane (RCI Labscan) (1:1 v/v) in an amber glass separatory funnel. After separation, the hexane layer (upper phase) was firstly kept in a dark environment and the methanol layer (lower phase) was re-extracted with a mixture of acetone and hexane until colorless. Then the pooled hexane layer was washed with water to remove acetone and dried by adding 2 g of anhydrous sodium sulfate (CARLO ERBA Reagents, Milano, Italy) and 2 g of butylhydroxytoluene (BHT) (ACROS organics, Geel, Belgium) was added as an antioxidant. Hexane was evaporated using a vacuum evaporator (RV10, IKA). The residue was dissolved in isopropanol (RCI Labscan) to a volume of 10 mL and used for HPLC analysis.

The extract (20 µL) was used for quantitative analysis using an HPLC system (Shimadzu, Kyoto, Japan) as described by Rungpichayapichet et al. [17]. The separation was carried out on a reverse phase C-18 column (Inertsil ODS-3, 4.6 × 150 mm, GL Sciences, Kyoto, Japan). A mixture of methanol and acetonitrile (RCI Labscan) (7:3 v/v) was used as a mobile phase at a flow rate of 1 mL min−1. β-Carotene was detected at 450 nm using a diode array detector (SPD-M20A, Shimadzu) and a certified β-carotene (purity ≥ 95%, Sigma-Aldrich, St. Louis, MO, USA) was used to prepare a calibration curve. The β-carotene content was calculated and was expressed in mg 100 g−1 (dry basis). The measurement was repeated in duplicate.

Statistical analyses

Reference data were analyzed by analysis of variance (ANOVA) with a randomized complete block design using fruits as blocks and Duncan’s multiple range test was applied to determine significant differences between quality attributes for each fruit section with a significance level of 0.05 (SPSS version 17, Chicago, IL, USA). In addition, principal component analysis (PCA) was performed to visually examine the influence of fruit position on chemical properties i.e., TSS, TA, pH, and β-carotene content, using The Unscrambler software. The difference between actual and predicted values obtained from prediction models was analyzed by paired t-tests.

Results and discussion

Changes in internal quality attributes during ripening of mango

‘Nam Dok Mai’ mangoes used in this study were mature, unripe fruits sorted by their specific gravity higher than 1.00 [18]. The mangoes were ripened naturally at ambient temperature (29.7 ± 0.7  ̊C) without the addition of a ripening accelerator. Figure 2 shows the changes in important internal quality attributes, which are general ripeness indicators including TSS, TA, pH, and firmness. In addition, β-carotene content which is a major carotenoid found in mango flesh, was also determined. β-Carotene is a pro-vitamin A, which is present abundantly in mangoes. A typical ripening behavior i.e., decrease in firmness and TA, and an increase in TSS, pH, and β-carotene content was observed. Changes in the fruit quality attributes particularly occurred during the first 5 days of ripening and thereafter remained unchanged or only slightly changed to the end of the ripening period. Significant decreases (more than 60%) were found in firmness and TA during the initial ripening period, while the highest TSS value was found on the fifth day of ripening. As reported in many studies, declining firmness relates to a process of change in the cell wall structure and composition by the action of enzymes e.g., polygalacturonase, pectin methylesterase, and cellulase during ripening, while a decrease in TA is due to the degradation of acids in the respiration process. In addition, hydrolysis of starch into sugar during ripening results in a rise in TSS [19, 20]. β-Carotene content continuously increased along the ripening period, sharply increased on the third day, and reached the maximum on the last day of ripening. A similar pattern was reported in previous studies [19, 21, 22]. Compared to other cultivars, Nam Dokmai subcv. Si Thong contains higher β-carotene content (0.6–8.5 mg 100 g−1) than Haden and Ataulfo (0.4–2.8 mg 100 g−1) [2], and Tainong (1.9–3.3 mg 100 g−1) [20].

Figure 3 shows PCA scores plots of the first two principal components (PC1 and PC2) using chemical data (TA, TSS, and β-carotene content) from each mango section on days 2, 4, 6, and 8 of the ripening. Clear boundaries were  not observed for data from each fruit section which indicated proximity of chemical compositions between fruit sections. However, in the last phase of ripening (days 6 and 8), the shoulder position was slightly distinct  from other parts whereas cheek and tip positions were clustered together. This result was confirmed by ANOVA analysis (data not shown) that fruits at the shoulder position contained lower concentrations of β-carotene compared to cheek and tip positions on the last four days of ripening. In addition, the highest TA was found in the tip position but was not significantly different from the other parts. In a previous study [22], however, significantly higher TA in tip than cheek and shoulder were observed. The results were in contrast to Nordey et al. [23] who reported that for mango cv. Cogshall, the tip position had significantly less TA and higher TSS than cheek and shoulder positions. However, the authors also found a non-significant difference of ethylene, a ripening related hormone, between fruit positions [23]. Therefore, non-significant variation in these fruit quality attributes in this study may be explained by this reason.

Fig. 2
figure 2

Internal quality attributes of ‘Nam Dok Mai’ mango fruit during ripening, including a firmness, b ripening index (RPI), c total soluble solids (TSS), d titratable acidity (TA), e pH, and f β-carotene content. Error bars indicate standard deviation. A similar letter above the bars indicates no significant difference (p > 0.05, Duncan’s test) between ripening days

Full size image
Fig. 3
figure 3

PCA score plots of ‘Nam Dok Mai’ mango on different ripening days using chemical properties including TA, TSS, and β-carotene content from shoulder (blue dot), cheek (red dot) and tip (green dot) positions. Ellipsis indicates the 95% confidence intervals (Color figure online)

Full size image

Prediction models for physicochemical properties of mango

Postharvest changes in mango simultaneously occur in both external peel and internal pulp. These changes are regulated by ethylene content which generally accumulates in pulp closed to seed. For example, during ripening, increasing of intense yellow color can be observed in peel and pulp due to the carotenoid synthesis and accumulation. Degradation of protopectin to pectin results in fruit pulp softening and peel hardness as well as respiratory water loss. Therefore, internal changes in pulp often reflect changes in external peel. Postharvest changes are related to the alteration of chemical compositions which mostly consist of C, H, O in their structures, therefore these will affect to the absorption energy and spectroscopic signal. Average original spectra of mango from each day of the ripening measured by NIRS (800–2500 nm) and HSI (450–998 nm) are illustrated together in Fig. 4a. In the visible wavelength region (450–700 nm), a strong absorption was found in the blue-green light area (450–500 nm), which is associated with the presence of carotenoid compounds [24, 25]. The absence of absorption peak around 670–695 nm, which is related to the chlorophylls, might be attributed to the characteristic all-yellow peel of this mango cultivar [26]. Broad band absorption peaks at 900–1050, 1100–1300, 1400–1600, and 1900–2000 nm were observed in the NIR region. These absorption bands are associated with the absorptions of C–H and O–H bonds in water and organic matters such as sugar, acids, and β-carotene [25, 27, 28]. As can be seen on the spectra, there were baseline shifts and overlapping peaks which might have affected the prediction performance, therefore the spectra were adjusted prior to calibration model development using several spectral pretreatment techniques.

Both original and pretreated spectra obtained from NIRS and HSI were used to develop calibration models for firmness, TSS, TA, pH, RPI, and β-carotene content. Several spectral pretreatment techniques including Savitzky-Golay smoothing (second polynomial order, 17 point), Savitzky-Golay first derivative (second polynomial order, 17 point), Savitzky-Golay second derivative (second polynomial order, 17 point), standard normal variate (SNV), and multiplicative scatter correction (MSC) were applied to adjust the spectral resolution (Fig. 4b–f).

Fig. 4
figure 4

Average spectra of ‘Nam Dok Mai’ mango during ripening a original NIRS (solid line) and HSI (dash line) spectra, b SG-NIRS spectra, c D1-NIRS spectra, d D2-NIRS spectra, e MSC-NIRS spectra, and f SNV-NIRS spectra

Full size image

NIRS prediction models

The best PLS regression results for the prediction of mango quality in each fruit section using NIR spectra are shown in Table 2. The best prediction ability for fruit quality attributes was obtained from the model developed from average fruit spectra, which were pretreated with D1, MSC and SNV. MSC and SNV techniques are known as the methods used to remove the effects of scattering and reduce baseline shift which can improve the spectral resolution [24, 28, 29]. However, TA was most accurately predicted by using the model developed from D2 pretreatment. The maximum RPD was found for RPI model at 3.16 with SEP of 0.54. However, less prediction ability was observed on the TSS model with \({R}_{C}^{2}\), SEP and RPD ranging between 0.78 and 0.93, 1.22–1.69 ˚Brix and 1.83–2.64, respectively.

Table 2 NIRS-PLS regression results for ‘Nam Dok Mai’ mango fruit quality attributes in different sections
Full size table

Regarding the fruit position, the calibration model developed from the tip position spectra showed a lower prediction error for all quality attributes. Based on the \({R}_{V}^{2}\) and RPD values, the prediction model for pH shows the best predicting performance (\({R}_{V}^{2}\)= 0.90, RPD = 3.11) followed by those for TA (\({R}_{V}^{2}\)= 0.86, RPD = 2.70), and TSS (\({R}_{V}^{2}\)= 0.83, RPD = 2.46). In this study, RPD values obtained from NIRS prediction models were in a range of 1.83–3.16. Munawar, Zulfahrizal, Meilina and Pawelzik [28] suggested that for quantitative prediction. Models with an RPD value between 2 and 2.5 are acceptable, while an RPD value between 2.5 and 3 or above indicates a good and excellent prediction accuracy. Therefore, the prediction models for all mango quality attributes developed from NIRS average spectra in this study are promising. Similar result was reported by Xu, Xu, Xie and Ying [30] that prediction models of soluble solids content (SSC) developed from the average spectra obtained from 7 different positions of apple fruits provided the accurate prediction performance.

HSI prediction models

Table 3 shows the best PLS regression results for the prediction of mango quality in each fruit section using HSI spectra. Similar to the NIRS regression results, calibration models developed from average spectra showed the best prediction performance in all fruit attributes except for pH, which the model developed from cheek spectra provided the best prediction ability. Overall, slightly lower accuracy of HSI prediction models as compared to NIRS prediction models, which might be because only absorbance data in visible and short-wavelength regions were measured. The best-fitting regression for firmness was achieved from SG pretreated spectra with \({R}_{C}^{2}\), SEP, and RPD of 0.81, 2.62 N, and 2.75, respectively. Screening quantitative prediction performance (RPD between 1.5 and 1.9) was observed for the TSS calibration model. The results agreed with many researchers that HSI has lower prediction performance for TSS than other attributes, e.g. firmness [12, 31, 32]. This might be because HSI relies on the scattering effect and the visible range wavelength included in HSI measurement does not enhance the predicting ability for TSS [31]. A relatively good prediction of β-carotene content was obtained from SNV pretreated spectra with \({R}_{C}^{2}\), SEP, and RPD of 0.84, 1.13 mg 100 g−1, and 2.80, respectively. The better prediction accuracy on β-carotene content using HSI spectra compared to NIRS might be attributed to high absorption in the visible wavelength region around 450–500 nm.

Table 3 HSI-PLS regression results for ‘Nam Dok Mai’ mango fruit quality attributes in different sections
Full size table

Scatter plots of the best calibration results for predicting mango fruit attributes using NIRS and HSI techniques are illustrated in Figs. 5 and 6. Based on the paired t-test methods, for all calibration models, no significant differences (p > 0.05) between actual and predicted quality attributes were found from both techniques (data not shown).

Fig. 5
figure 5

Prediction results for a firmness, b total soluble solids (TSS), c titratable acidity (TA) of ‘Nam Dok Mai’ mango using NIRS calibration models (green color) and HSI calibration models (blue color) (open circle calibration set, × validation set) (Color figure online)

Full size image
Fig. 6
figure 6

Prediction results for a pH, b β-carotene content, and c RPI of ‘Nam Dok Mai’ mango using NIRS calibration models (green color) and HSI calibration models (blue color) (open circle calibration set, × validation set) (Color figure online)

Full size image

MLR calibration models

To identify the relevant wavelengths that are important to estimate each mango quality parameter, loading weight value and PLS regression coefficient were computed and analyzed. Figure 7 shows an example of NIRS-PLS regression coefficient and loading weight plot, which represent relevant variables for TSS prediction. Significantly contributed variables were mainly located in the long-wavelength region (1100–2500 nm). A set of wavelengths representing the first overtone of O–H absorptions and C–H combinations (1350–1450 nm), the first overtone of C–H bond (1600–1800 nm) and the region between 2100 and 2400 nm that corresponded to water and carbohydrate absorption was observed as relevant wavelengths for TSS prediction. For firmness prediction models, the second overtone of the C=O bond (1800–1900 nm) and combinations of C–H + C–H absorption (2200–2400 nm) were found as relevant variables. The results were in agreement with Buyukcan and Kavdir [33] that the effective wavelength ranges for SSC of apricot were between 1333 and 1836 nm and between 2173 and 2300 nm while the effective wavelength ranges for firmness were between 800 and 1836 nm and between 2173 and 2355 nm. Similar to firmness and TSS models, a set of wavelengths around 1700 and 2200–2400 nm was observed in the β-carotene content prediction model.

Fig. 7
figure 7

Regression coefficient plot and loading weight plot for TSS prediction developed from average ‘Nam Dok Mai’ mango fruit spectra measured by NIRS

Full size image

From the regression coefficient plot and loading weight plot of the best fitting PLS prediction models, the relevant wavelengths were chosen as candidate wavelengths and evaluated for significant impact using the backward elimination stepwise method and the number of significant wavelengths used for NIRS-MLR model construction were defined as describe in our previous study [17]. Table 4 shows prediction performance of NIRS-MLR prediction models developed by significant wavelengths. Lower predictive accuracy was observed on MLR models compared to PLS model due to a smaller number of variables used for model construction. The minimum number of wavelengths was found for firmness model at three wavelengths, 2310, 2344, and 2400 nm, whereas pH (1692, 1719, 1730, and 2396 nm) and β-carotene content prediction model (1721, 1735, 2212, and 2314 nm) required four significant wavelengths. Five significant variables (1350, 1392, 1878, 1931, and 1997 nm) were used for RPI prediction model while greater number of variables were chosen for TSS (7 wavelengths including 1402, 1418, 1668, 1687, 1705, 1733, and 2031 nm) and TA prediction model (1145, 1161, 1673, 1688, 1708, 1735, and 1752 nm).

Table 4 NIRS-MLR prediction results for ‘Nam Dok Mai’ mango fruit quality attributes
Full size table

PLS regression coefficient plots of TA, pH, and β-carotene content prediction developed from average HSI spectra are illustrated in Fig. 8. Candidate variables were selected and evaluated by the backward elimination stepwise method. Prediction results of HSI-MLR models developed using significant wavelengths are listed in Table 5. For HSI-MLR prediction models, the important wavelengths for TA determination were 514, 550, 646, 710, 886, and 890 nm. This set of wavelengths represented the fourth overtone (710 nm) and the third overtone (890 nm) of C-H bonds and carotenoids absorption. For pH prediction, four wavelengths (482, 514, 550, and 802 nm) were included in the MLR prediction model. These results agreed with the finding of other studies, which showed that the spectral regions at visible wavelengths are capable to predict TA and pH [34, 35]. However, the ability of visible wavelength to determine pH might be from the relation between the changes of pigments and internal fruit quality attributes during fruit ripening [36].

Table 5 HSI-MLR prediction results for ‘Nam Dok Mai’ mango fruit quality attributes
Full size table

For β-carotene content prediction, important wavelengths were 466, 482, 518, 546, 578, 678, 922, and 954 nm. As reported by Davey et al. [37], a wavelength region between 400 and 600 nm is strongly absorbed by carotenoids, especially at 520 and 540 nm. Therefore, a greater number of wavelengths around 450–600 nm were found as the important wavelengths that enhance the predicting ability of the β-carotene content prediction model. Four significant wavelengths, 514, 554, 674, and 914 nm, were responsible for firmness prediction while five variables were required for RPI (514, 524, 678, 858, and 990 nm) and TSS prediction models (506, 538, 671, 762, and 898 nm).

Fig. 8
figure 8

Regression coefficient plots for a TA b pH, and c β-carotene content prediction developed from average ‘Nam Dok Mai’ mango fruit spectra measured by HSI

Full size image

Conclusion

This study was performed to explore the potential of NIRS and HSI as non-destructive techniques for internal quality attributes assessment of mangoes. Fruit position showed only a slight influence on the prediction performance for both devices, and using the average fruit spectra gave the best prediction performance. The PLS models developed from NIRS spectra provided an acceptable and good prediction ability for firmness (RPD = 2.57), TSS (RPD = 1.83–2.64), TA (RPD = 2.12–3.07), pH (RPD = 2.55–3.11), RPI (RPD = 3.16), and β-carotene content (RPD = 1.93–2.65). Slightly lower prediction accuracy was found for the models obtained from HSI spectra with the RPD ranging between 1.47 and 3.58. Although the reduction of variable numbers in MLR prediction models resulted in slightly lower prediction ability of the models from both devices, acceptable prediction performances for general screening purposes were obtained. The overall results obtained in this study showed the feasibility of NIRS and HSI for quality control in mango sorting. However, this study was performed using mango from one cultivar therefore further study is still required involving analysis of more samples from different harvest years or other cultivars.