The Estimation of Chemical Properties of Pepper Treated with Natural Fertilizers Based on Image Texture Parameters

The cultivar and fertilization can affect the physicochemical properties of pepper fruit. This study aimed at estimating the content of α-carotene, β-carotene, total carotenoids, and the total sugars of unfertilized pepper and samples treated with natural fertilizers based on texture parameters determined using image analysis. Pearson’s correlation coefficients, scatter plots, regression equations, and coefficients of determination were determined. For red pepper Sprinter F1, the correlation coefficient (R) reached 0.9999 for a texture from color channel B and −0.9999 for a texture from channel Y for the content of α-carotene, −0.9998 (channel a) for β-carotene, 0.9999 (channel a) and −0.9999 (channel L) for total carotenoids, as well as 0.9998 (channel R) and −0.9998 (channel a) for total sugars. The image textures of yellow pepper Devito F1 were correlated with the content of total carotenoids and total sugars with the correlation coefficient reaching −0.9993 (channel b) and 0.9999 (channel Y), respectively. The coefficient of determination (R2) of up to 0.9999 for α-carotene content and the texture from color channel Y for pepper Sprinter F1 and 0.9998 for total sugars and the texture from color channel Y for pepper Devito F1 were found. Furthermore, very high coefficients of correlation and determination, as well as successful regression equations regardless of the cultivar were determined.


Introduction
Nowadays, the consequences associated with conventional agricultural production are observed, such as the disruption of natural diversity, soil and water pollution by fertilizers and plant protection products, reduced biomass production, and an increase in diseases. Therefore, there is an increasing trend in agricultural production based on the agroecological concept. With the ever-increasing nutritional awareness of society, the group of consumers looking for organically produced food is growing, and thus the demand for products with high health-promoting value and free of pesticides is increasing. The use of chemical plant protection products is an easy way of combating organisms that are harmful to plants, but it carries a risk to the health of people, animals, and the environment. More and more modern preparations of natural origin for use in plant cultivation are appearing on the market, which have a beneficial effect on plant-life processes and, at the same time, can influence the composition and content of bioactive and nutritional compounds and protect plants from diseases and pests [1,2].
Sweet pepper (Capsicum annuum L.), otherwise known as the annual pepper plant, is classified as a plant of the Solanaceae family. About 39 pepper species have been identified, including five domesticated species, C. annuum, C. baccatum, C. chinense, C. frutescens, and C. pubescens [3]. It is a vegetable with low energy value and low index and glycemic load. It is one of the vegetables characterized by high health-promoting compounds, carotenoids (especially beta-carotene, α-carotene, and lutein); phenolic compounds; and vitamins, such can be analyzed using specialized software packages, for example, MaZda, which enables quantitative analysis, including texture parameters computation, as well as texture selection and extraction, data classification and visualization, and image segmentation tools. MaZda provides options for the transformation of color images to grayscale and conversions to individual color channels (components) [31][32][33]. Several channels are most commonly used. There are color channels R (red), G (green), and B (blue) from the RGB color space; L (lightness component from black to white), a (red for positive and green for negative values), and b (yellow for positive and blue for negative values) from the Lab color space; and X (component with color information), Y (lightness), and Z (component with color information) from the XYZ color space [29,34].
The objective of this study was to estimate the chemical properties of peppers belonging to two cultivars treated with natural fertilizers in a fast and inexpensive manner based on image texture parameters. A great novelty of the present study is related to the estimation of α-carotene, β-carotene, total carotenoids, and total sugars of untreated red pepper Sprinter F 1 and yellow pepper Devito F 1 and samples subjected to different organic fertilization, such as NaturalCrop ® SL, Bio-algeen S90, and nettle fertilizer. The innovative regression equations including features selected from a set of 1629 image textures from nine color channels were determined. The developed procedure could enable the selection of fruit with the desired properties based on the image features without the need to perform more labor-intensive and expensive measurements.

Experimental Design
The experimental material consisted of pepper growing in greenhouses at the Laboratory of Cultivation of Vegetable and Edible Mushrooms of the National Institute of Horticultural Research in Skierniewice in Poland. Peppers belonged to two cultivars of Sprinter F 1 characterized by red fruit and Devito F 1 with yellow fruit. The experiment was carried out between March and October 2021. During the experiment, daily climatological data were collected inside the greenhouse, using a thermo-hygrometer. Organic pepper seeds were used and potted seedlings were planted in the soil in an ecological greenhouse under the principles of organic farming. The experiment was set up in the soil with a pH of 6.0. Before planting, the soil was fertilized with organic plant compost at a dose of 300 kg per 100 m 2 , obtaining the soil fertility at the level (mg dm −3 ) of N-230, P-310, K-400, and Mg-350. The seedlings were planted on April 20, in a row-row system, at a spacing of 35 cm × 60 cm (3.6 plants per 1 m 2 ). Plants were grown in a four-shoot system. Pepper fertigation was carried out with the use of a drip system and fertilizers approved for organic cultivation depending on the current humidity and nutritional requirements of the pepper in particular growth stages [35]. The fruit was harvested at the stage of colorful ripeness.

Image Acquisition and Processing
The image acquisition was performed using an Epson Perfection V600 flatbed scanner. The scanner was placed in a black box to obtain a black background of images. The images of pepper slices (cross-sections) with a thickness of 10 mm obtained using a slicing machine were considered. In the case of each cultivar and treatment, images of 50 slices were acquired, as follows: − Fifty slices of a control sample of Sprinter F 1 pepper, − Fifty slices of Sprinter F 1 pepper treated with NaturalCrop ® SL, − Fifty slices of Sprinter F 1 pepper treated with Bio-algeen S90, − Fifty slices of Sprinter F 1 pepper treated with nettle fertilizer, − Fifty slices of a control sample of Devito F 1 pepper, − Fifty slices of Devito F 1 pepper treated with NaturalCrop ® SL, − Fifty slices of Devito F 1 pepper treated with Bio-algeen S90, − Fifty slices of Devito F 1 pepper treated with nettle fertilizer.
The acquired images were saved in a TIFF format at 800 dpi resolution. Before processing in the Mazda software (Łódź University of Technology, Institute of Electronics, Łódź, Poland) [31][32][33], the pepper slice images were converted to a BMP format. Then, texture parameter extraction was carried out for each of the color channels L, a, b, R, G, B, X, Y, and Z. The sample images are presented in Figure 1. The image segmentation into pepper slices and the background was performed manually based on the pixel brightness intensity. Each pepper slice (the area of flesh without skin) was considered as a region of interest (ROI). For each ROI, in total 1629 image textures, including 181 textures for each color channel were computed.

Chemical Properties
The samples for chemical analysis were prepared from 10 pepper fruits from each repetition. Pepper fruits were cut into quarters, frozen in liquid nitrogen, crushed in a Blixer homogenizer and stored at −20 °C.

Chemical Properties
The samples for chemical analysis were prepared from 10 pepper fruits from each repetition. Pepper fruits were cut into quarters, frozen in liquid nitrogen, crushed in a Blixer homogenizer and stored at −20 • C.

HPLC Analysis of Sugars
An HPLC analysis of total sugars (sucrose, glucose, and fructose) was determined by high-performance liquid chromatography (Agilent 1200 HPLC system, equipped with a differential refractometric detector), using Aminex HPX-87C (300 mm × 7.5 mm) with a precolumn according to European Standard EN 12630. The isocratic flow was 0.6 mL·min −1 , column temperature was 80 • C, and mobile phase-edetate calcium disodium (Ca-EDTA). The samples were dissolved in redistilled water, homogenized, and purified on a Waters SepPak PLUS C18 filter. The sugars were quantified by calibration curve for sucrose, glucose and fructose, and the results were expressed as mg 100 g −1 fresh mass (f.m.).

Carotenoid Extraction
Carotenoid content was determined by the method according to Bohoyo-Gil et al. [36]. The sample was homogenized in the extraction solution (hexane:acetone 6:4) and filtered through a Büchner funnel under reduced pressure. Next, the extract was transferred to a separating funnel and shaken with the addition of water. After phase separation, the water-acetone phase was discarded. The acetone rinsing operation was repeated until the lower phase was free of acetone and the upper hexane phase containing carotenoids was filtered through a filter paper containing anhydrous sodium sulphate into an evaporation flask. Hexane was evaporated to dryness in a vacuum evaporator at 40 • C, the dry residue was quantitatively transferred to a 25 mL flask with a solution of acetonitrile:methanol:ethyl acetate 55:25:20 + 0.1% BHT + 1 mL TEA and 4 mL of hexane. The flask extract was filtered with a 45 µm PTFE filter into an amber bottle and analyzed by HPLC.

Statistical Analysis
The linear relationships between the chemical properties and image textures of pepper samples were determined using STATISTICA 13.1 (Dell Inc., Tulsa, OK, USA, StatSoft Polska, Kraków, Poland). The analysis was performed separately for samples of Sprinter F 1 and Devito F 1 and then for a dataset combining both cultivars Sprinter F 1 and Devito F 1 . In the case of Sprinter F 1 pepper, among the chemical properties, α-carotene, β-carotene, total carotenoids, and total sugars were considered. For Devito F 1 , α-carotene was not detected and the differences in the content of β-carotene were very small. Therefore, the analyses for Devito F 1 and a dataset combining Sprinter F 1 and Devito F 1 were performed only for total carotenoids and total sugars. In the first step, Pearson's correlation coefficients (R) were determined at a significance level of p < 0.05, and scatter plots were created. In the case of image textures from color channels L, a, b, R, G, B, X, Y, and Z for which the correlations occurred, the highest value of R for positive correlation and the highest value of R for negative correlation were selected to be presented in this paper. Then, regression equations for the estimation of chemical properties of pepper untreated and treated with natural fertilizers using image texture parameters were determined and coefficients of determination (R 2 ) were computed.

Relationship between Chemical Properties and Image Texture Parameters of Red Pepper Sprinter F 1
In the case of red pepper Sprinter F 1 , correlations between α-carotene, β-carotene, total carotenoids, and total sugars with selected image textures were observed (Table 1). Very strong positive and negative correlations were determined for each of the considered organic chemical compounds reaching 0.9999 and −0.9999 for α-carotene and total carotenoids. In the case of α-carotene, statistically significant correlation coefficients were found for textures of images in color channels a, R, G, B, Y, and Z. The correlation coefficient (R) of 0.9999 was found for one image texture from color channel B and −0.9999 for a texture from channel Y. None of the image textures from channels L, b, and X correlated with α-carotene. Therefore, these channels turned out to be useless for the estimation of α-carotene content. For total carotenoids, both positive and negative relationships with image textures from all color channels L, a, b, R, G, B, X, Y, and Z were observed, and the values of R equal to 0.9999 and −0.9999 were determined for color channels a and L, respectively. In the case of β-carotene, the positive correlations with image textures from color channels a, R, G, B, Y, and Z were observed, reaching 0.9994 for a texture from channel B. Whereas negative correlations with textures from color channels a, R, G, B, X, Y, and Z were observed, and the highest value was −0.9998 for a texture from channel a. It was found that the content of total sugars was positively correlated only with textures of images in color channels R, G, and X with the coefficient of up to 0.9998 in the case of color channel R. Negative correlations with textures from color channels L, a, R, G, X, Y, and Z reached the value of −0.9998 for channel a. Table 1. Correlation coefficients (R) between α-carotene (mg 100 g −1 f.m.), β-carotene (mg 100 g −1 f.m.), total carotenoids (mg 100 g −1 f.m.), and total sugars (%) with selected image texture parameters for control and treated with NaturalCrop ® , Bio-algeen S90, and nettle fertilizer red pepper Sprinter F 1 samples; p < 0.05.
The scatter plots for α-carotene, β-carotene, total carotenoids, total sugars with selected image texture parameters, for which the highest values coefficient R for positive and negative correlations were determined, confirmed the strong relationship between chemical properties and image textures of control red pepper Sprinter F 1 and samples treated with natural fertilizers (Figure 2). The positive relationships between α-carotene and the texture from color channel B, β-carotene and the texture from color channel B, total carotenoids, and the texture from color channel a, as well as total sugars and the texture from color channel R, are presented. The strongest negative correlations between α-carotene with the texture from color channel Y, β-carotene with the texture from color channel a, total carotenoids with the texture from color channel L, and total sugars with the texture from color channel a are shown.

Relationship between Chemical Properties and Image Texture Parameters of Yellow Pepper Devito F 1
The texture parameters of flesh images of yellow pepper Devito F 1 were correlated with the content of total carotenoids and total sugars ( Table 2). Positive correlations were found between total carotenoids and image textures from color channels L, a, b, B, X, and Y with the value of correlation coefficient reaching 0.9899 for the texture from channel X, as well as total sugars and image textures from color channels L, a, b, G, B, X, and Y reaching 0.9999 for the texture from channel Y; while the content of total carotenoids was negatively correlated with image textures from color channels L, b, R, B, and X with a correlation coefficient of up to −0.9993 for the texture from channel b. The negative correlations between the content of total sugars and image textures from channels L, a, G, B, X, and Y reaching −0.9971 for the texture from channel G were determined. The relationships between total carotenoids and total sugars with selected Image texture parameters of control and fertilized yellow pepper Devito F 1 are presented in Figure 3 in the form of scatter plots. The graphs for textures characterized by the strongest positive and negative correlations, such as textures from color channels X and b for total carotenoids, and textures from color channels Y and G for total sugars are shown.  DifEntrp-difference entropy; GNonZeros-percentage of pixels with nonzero gradient; Teta4 parametr θ4; Teta2-parametr θ2; DifVarnc-difference variance; Perc-percentile; Kurtosis-h togram's kurtosis; InvDfMom-inverse difference moment; LngREmph-long run emphas SumVarnc-sum variance; Entropy-entropy; SumEntrp-sum entropy; GMean-absolute gra ent mean; Contrast-contrast; and Correlat-correlation.

Relationship between Combined Chemical Properties and Image Texture Parameters of Red Pepper Sprinter F 1 and Yellow Pepper Devito F 1
In the next step of the analysis, correlations between a set of combined image texture parameters and chemical properties of red pepper Sprinter F 1 and yellow pepper Devito F 1 belonging to the unfertilized group and samples treated with NaturalCrop ® , Bio-algeen S90, and nettle fertilizer were determined ( Table 3). The content of total carotenoids was positively and negatively correlated with image texture parameters from color channels L, a, b, B, G, X, Y, and Z. The highest positive and negative correlation coefficients were 0.9956 and −0.9935 for textures from color channel Y. In the case of total sugars, the positive correlations with image textures from channels L, a, b, G, X, and Y were observed and the correlation coefficient reached 0.9768 for the texture from channel Y; whereas the negative correlations with image textures from channels L, a, b, G, B, X, and Y with the highest value of −0.9583 for the texture from color channel G were determined.
The scatter plots for total carotenoids and total sugars confirmed the strong relationships with selected image texture parameters of red pepper Sprinter F 1 and yellow pepper Devito F 1 belonging to control and samples treated with NaturalCrop ® , Bio-algeen S90, and nettle fertilizer (Figure 4). The strongest positive and negative correlations between total carotenoids with textures from color channel Y and total sugars with textures from color channels Y and G are presented. Table 3. Correlation coefficients (R) between total carotenoids (mg 100 g −1 f.m.) and total sugars (%) with selected image texture parameters for red pepper Sprinter F 1 and yellow pepper Devito F 1 belonging to control and samples treated with NaturalCrop ® , Bio-algeen S90, and nettle fertilizer; p < 0.05. (mg 100 g −1 f.
Additionally, the regression equations for estimating chemical properties of unfertilized and fertilized pepper samples, as well as coefficients of determination (R 2 ) separately for red pepper Sprinter F 1 , yellow pepper Devito F 1 , and a set combining parameters for Sprinter F 1 and Devito F 1 were determined ( Table 4). The higher R 2 , the better the data fit the regression model. Image textures whose values were the most different between groups of peppers were the most suitable for building regression models. In the case of each chemical parameter, one regression equation for a texture for which the strongest positive correlation and one equation for the strongest negative correlation were set up. In the case of pepper Sprinter F 1 , all values of R 2 were higher than 0.9980, reaching 0.9999 for α-carotene content and the texture YS4RVLngREmph from color channel Y. It indicated that the texture YS4RVLngREmph most differentiated all groups of pepper Sprinter F 1 , such as control and samples treated with NaturalCrop ® SL, Bio-algeen S90, and nettle fertilizer. Pepper Devito F 1 was also characterized by very high R 2 of up to 0.9998 for the content of total sugars and the texture YS5SH1Entropy from color channel Y. It meant that the control sample of pepper Devito F 1 , and samples treated with three different natural preparations were the most different in terms of the texture YS5SH1Entropy that was most useful for developing the regression model; whereas the values of R 2 reached 0.9912 for total carotenoids and the texture YS5SH5Contrast from color channel Y for a set combining parameters for both cultivars. This confirmed that the textures of color channel Y of images are the most successful for estimating chemical properties of untreated peppers and samples treated with natural fertilizers. Additionally, the regression equations for estimating chemical properties of unfertilized and fertilized pepper samples, as well as coefficients of determination (R 2 ) separately for red pepper Sprinter F1, yellow pepper Devito F1, and a set combining parameters for Sprinter F1 and Devito F1 were determined ( Table 4). The higher R 2 , the better the data fit the regression model. Image textures whose values were the most different between groups of peppers were the most suitable for building regression models. In the case of each chemical parameter, one regression equation for a texture for which the strongest positive correlation and one equation for the strongest negative correlation were set up. In the case of pepper Sprinter F1, all values of R 2 were higher than 0.9980, reaching 0.9999 for α-carotene content and the texture YS4RVLngREmph from color channel Y. It indicated that the texture YS4RVLngREmph most differentiated all groups of pepper Sprinter F1, such as control and samples treated with NaturalCrop ® SL, Bio-algeen S90, and nettle fertilizer. Pepper Devito F1 was also characterized by very high R 2 of up to 0.9998 for the content of total sugars and the texture YS5SH1Entropy from color channel Y. It meant that the control sample of pepper Devito F1, and samples treated with three different natural preparations were the most different in terms of the texture YS5SH1Entropy that was most useful for developing the regression model; whereas the values of R 2 reached 0.9912 for total carotenoids and the texture YS5SH5Contrast from color channel Y for a set combining parameters for both cultivars. This confirmed that the textures of color channel Y of     The performed studies revealed strong correlations between the chemical properties and selected image texture parameters of red and yellow pepper samples. Very correct regression equations turn out to be useful for the estimation of the content of carotenoids and sugars based on pepper image parameters. Additionally, available literature reported the relationships between image or spectral features and chemical characteristics. In our previous research, the correlations between lycopene content and image textures of tomato fruit were determined with the correlation coefficient (R) reaching −0.99 for selected texture parameters of images in channels G, b, and Y and coefficients of determination (R 2 ) of 0.99 for image texture from channel G [37]. Pace et al. [38] observed relationships of the visual appearance and color parameters of images of fresh-cut nectarines with R 2 of up to 0.7622. The sugar content of potatoes was predicted based on hyperspectral imaging data by Rady et al. [39] with R reaching 0.97 for glucose. The sugar content was also estimated for apples with a high R parameter (0.8861) using multispectral imaging [40] and wine grapes with R 2 > 0.8 based on visible and near infrared and the short-wave infrared point spectroscopic data [41]. Furthermore, the prediction of soluble solid content of apples with R > 0.8 was possible based on features extracted from laser light backscattering images [42]. Our results presented in this paper were a significantly novel compared to the available literature, and included the determination of correlations and models for predicting the content of chemical properties based on selected parameters from a very large set of over 1600 textures from nine color channels. It allowed for the selection of textures providing correlation coefficients of up to 0.9999 and −0.9999, as well as very high coefficients of determination reaching 0.9999. The successful results may prompt the use of the developed procedure for other species and varieties of fruit and vegetables and the testing of the approach for other chemical parameters.

Conclusions
The correlations between chemical properties and image textures from different color channels of red and yellow peppers untreated and treated with natural fertilizers were determined. In the case of red pepper Sprinter F 1 , statistically significant correlation coefficients were found between α-carotene, β-carotene, total carotenoids, and total sugars with selected image texture parameters. Yellow pepper Devito F 1 , as well as a combined dataset for both cultivars were characterized by the correlations between total carotenoids and total sugars with selected textures. Furthermore, the high coefficients of determination confirmed the possibility of estimation of chemical properties based on image texture parameters from different color channels for individual cultivars, as well as regardless of the cultivar. The highest values of R 2 were for regression equations set up based on selected textures from images in color channel Y.
Due to the achievement of promising results, research aimed at estimating chemical properties based on image features can be continued and further research directions can be set. Further research may include more pepper cultivars, as well as other species of fruit and vegetables. Furthermore, more natural fertilizers can be used to provide more groups