Quantifying naturally colored cotton fiber color intensity using RGB image-based phenotyping

Abstract

Image-based phenotyping has become an innovative approach in plant breeding. This study quantified fiber color intensity in Gossypium hirsutum L. genotypes and evaluated the effect of boll position in the upper, middle, and lower plant canopy thirds. Five cultivars and twelve F _2:3 lines were assessed by sampling five plants per plot, collecting three bolls per third, and photographing them with a standard color card. Images were processed using ImageJ and analyzed in R with the FIELDimageR package to calculate the Brightness Index (BI), Saturation Index (SI), and Normalized Green-Red Difference Index (NGRDI). Color card calibration increased mean values without changing genotype groupings. The upper third showed 6.25% lower color intensity, with no significant genotype × third interaction. SI efficiently discriminated genotypes, while BI strongly correlated with fiber Uniformity Index, Strength, and Upper Half Mean Length (R² = 0.77, 0.80, and 0.88, respectively), supporting its use in breeding programs.

Keywords:
Plant breeding; phenomics; high-throughput phenotyping

INTRODUCTION

Upland cotton (Gossypium hirsutum L.) is the most important fiber crop worldwide, cultivated in over 60 countries. Brazil is the third largest producer and second largest exporter, accounting for 13% of global production. In the 2023/24 season, Brazilian output reached 3.6 million tons, with an average yield of 1,889 kg ha⁻¹ (ABRAPA 2025, USDA 2025, CONAB 2025).

Besides fiber, cotton is a multipurpose crop used in food, biofuel, cosmetics, animal feed, and other industries (Zonta et al. 2015). Historically, Brazilian breeding programs have focused on white fiber cultivars with high yield and fiber quality, while naturally colored cotton has largely remained restricted to artisanal and niche markets (Carvalho et al. 2015).

Colored cotton represents a sustainable alternative to conventional dyeing, reducing water consumption, toxic waste, and manufacturing costs (Naoumkina et al. 2024). Nevertheless, its large-scale adoption is still constrained by lower yield, inferior fiber quality, color non-uniformity, and the absence of standardized color-based classification systems (Penna 2005). Although the High-Volume Instrument (HVI) systems are widely used for fiber classification, they remain limited by high cost, infrastructure demands, and low mobility (Gonçalves et al. 2015).

Identifying colored cotton genotypes with stable coloration and fiber properties compatible with industrial standards is essential. However, traditional visual color assessment is subjective and inefficient, particularly because commercially important cultivars may exhibit very similar shades, making visual distinction difficult (Gonçalves et al. 2015). In this context, image-based phenotyping has emerged as an accurate, scalable, and promising alternative, with successful applications already reported in several crops (Bac et al. 2013, Zhao et al. 2016).

Thus, this study aimed to: i) quantify fiber color intensity using image-based phenotyping in 17 colored cotton genotypes, including commercial cultivars and F_2:3 breeding lines; ii) assess the effect of boll position within the plant canopy; and iii) evaluate the use of color cards in extracting image-based color indices.

MATERIAL AND METHODS

Plant material and field trials

The experiment was conducted over two growing seasons (2023/24 and 2024/25) at Fazenda Glória, Federal University of Uberlândia (lat 18° 56′ 52.77″ S, long 48° 12′ 51.97″ W, and alt 865 m asl). Planting was carried out on January 5, 2024, and December 17, 2024, with harvests in July 2024 and between July and August 2025, respectively. The regional climate is classified as tropical savanna (Aw), characterized by rainy summers and dry winters. During the experimental periods, the mean air temperature was approximately 22 °C in both growing seasons, while cumulative rainfall was approximately 680 mm in 2023/24 and 930 mm in 2024/25, with precipitation concentrated mainly between December and March.

In the first growing season, 17 naturally colored cotton genotypes and one white cotton genotype were evaluated using an augmented block design (Federer 1961) with four replicates. The commercial cultivars BRS Rubi, BRS Jade, BRS Verde, BRS Topázio (naturally colored cotton), and Suregrow 821 (white cotton) served as the check genotypes. The remaining entries consisted of 11 F₂:₃ breeding lines (UFUJP-01, UFUJP-02, UFUJP-03, UFUJP-04, UFUJP-10, UFUJP-11, UFUJP-13, UFUJP-14, UFUJP-15, UFUJP-17, and UFUJP-18) developed by the Cotton Genetic Breeding Program (PROMALG) at the Federal University of Uberlândia (UFU).

In the second growing season, only the five commercial check genotypes (BRS Rubi, BRS Jade, BRS Verde, BRS Topázio, and Suregrow 821) were evaluated in a randomized complete block design (RCBD) with four replicates, resulting in a total of 20 experimental plots. The experiment conducted solely with the five fixed genotypes was designed to provide a common environmental reference, allowing the separation of environmental and genetic effects across seasons.

The experimental plots consisted of four rows, each five meters long and spaced 0.9 m apart, totaling 18 m², with the two central rows considered as the usable area. Prior to sowing, soil acidity was corrected by the manual application of 2.2 t ha⁻¹ of dolomitic limestone, calculated by the base saturation method to increase V% from 53 to 70%, as well as supplying calcium and magnesium according to soil analysis recommendations (Table S1). At sowing, 20 kg ha⁻¹ of nitrogen and 60 kg ha⁻¹ of phosphorus were applied. Topdressing consisted of 60 kg ha⁻¹ of nitrogen and 30 kg ha⁻¹ of potassium, split into two applications at 25 and 45 days after emergence (DAE). Additionally, boron was supplied at a total rate of 1 kg ha⁻¹ through four foliar applications.

Plant growth was managed using the growth regulator chlormequat chloride (2-chloroethyltrimethylammonium chloride). The first application was performed at 45 DAE, and subsequent applications were carried out according to the plant growth rate, targeting a final plant height of approximately 1.3 m at harvest. For boll evaluation, five bolls were collected from each plant third (upper, middle, and lower), totaling 15 bolls per plot.

Image acquisition and analysis

After harvest, cotton bolls were carefully processed to separate lint from the seeds. Individual locules were isolated, and the lint was arranged to highlight the natural gradient of fiber coloration for photographic documentation. Bolls were grouped according to their original plant canopy thirds (upper, middle, and lower), and three representative bolls from each position were photographed together to ensure consistent comparative analysis.

Image acquisition was performed using a low-cost custom imaging camera (30 cm wide × 50 cm high) designed to provide a controlled environment and reduce interference from external light. Two neutral white LED lamps (9 W, ~800 lm each), symmetrically positioned on the sides and covered with two layers of tracing paper, provided homogeneous diffuse illumination. Samples were centralized in a blue background with a Pixel Perfect Camera color correction card (5" × 7") (Schneider et al. 2012) at the base of the camera and photographed at a fixed vertical distance of approximately 50 cm using an iPhone 14 Pro Max™ camera (2× telephoto, 12 MP, 48 mm, f/1.78 aperture; Apple, Zhengzhou, China). The use of this device was justified by its high image resolution, optical stability, and portability, allowing for fast and standardized image acquisition. All images were captured with manual settings, with automatic exposure and white balance disabled to ensure consistent acquisition parameters throughout the experiment. After acquisition, color calibration was performed using ImageJ (Figure 1).

Figure 1
Experimental workflow: i) image acquisition and color calibration, ii) color thresholding, iii) background segmentation and removal, iv) color index extraction, and v) statistical analysis.

The analysis of the image dataset was performed using the R software (R Core Team 2023). For all images, the background was removed using the Saturation Index (SI) with a value of 0.0 (Figure 1), and each photo was visually inspected to ensure accuracy with the aid of the FIELDimageR package (Matias et al. 2020). The fieldIndex function of the FIELDimageR package was used to measure the average red (R), green (G), and blue (B) values ​​of the selected area on a scale of 0-255. These values ​​were used to measure the following indices and standard formulas:

The following color indices were used to characterize fiber color properties. The Brightness Index (BI) quantifies color intensity based on the root mean square of the RGB values, with higher values indicating greater intensity (Richardson and Wiegand 1977):

$BI=\sqrt{\left(\frac{{\mathrm{R}}^2+{\mathrm{G}}^2+{\mathrm{B}}^2}{3}\right)}$

b. Normalized green-red difference index (NGRDI) assesses color variation based on the difference between the green and red bands, and a higher value may indicate greater uniformity (Tucker 1979).

$NGRDI=\frac{G-R}{G+R}$

c. Spectral slope saturation index (SI) evaluates spectral saturation in relation to the red and blue bands, and a higher value may indicate a more saturated color (Escadafal et al. 1994).

$SI=\frac{R-B}{R+B}$

Evaluation of the intrinsic characteristics of the fiber

Bolls harvested from each canopy third were collected and ginned separately using a 20-saw laboratory gin to obtain lint samples. The lint was subsequently packaged and submitted to the Minas Cotton Laboratory of the Minas Gerais Cotton Producers Association (AMIPA) for fiber quality analysis. Fiber technological properties were determined using a High Volume Instrument (HVI), including upper half mean length (UHML, mm), length uniformity (UI, %), fiber strength (Str, gf tex⁻¹), micronaire (Mic), elongation (Elg, %), yellowness (+b), fiber maturity (Mat, %), spinning consistency index (SCI), and short fiber index (SF).

Statistical analysis

Using R (R Core Team 2023), individual analyses were first performed separately for each growing season and for the indices obtained with and without image calibration via ColorCard, considering the bolls collected in each developmental third. For the 2023/24 season, conducted under an augmented block design according to Federer, the repeated check genotypes distributed across blocks were used to estimate the experimental error and to adjust the means of unreplicated entries for block effects through linear modeling. Thus, adjusted genotype means (least-square means) were obtained for each evaluated trait within this environment. For the 2024/25 season, established in a randomized complete block design, individual analyses of variance were performed using the conventional model including genotype and block effects, and adjusted means were similarly extracted.

Subsequently, a two-stage joint analysis of variance was carried out using the adjusted means obtained from both environments in order to compare the five genotypes common to both growing seasons and to evaluate genotype stability across environments (p ≤ 0.05). The combined linear model adopted was:

$Y_{ijk}=\mu +B_{j\left(k\right)}+G+E_k+{\left(GE\right)}_{ik}+{\varepsilon }_{ijk}$

where $\mu$ is the overall mean; $B_{j\left(k\right)}$ is the effect of block $j$ nested within environment $k$; $G_i$ is the fixed effect of genotype $i$; $E_k$ is the fixed effect of environment $k$; ${\left(GE\right)}_{ik}$ is the genotype × environment interaction; and ${\epsilon }_{ijk}$ is the residual error. In the augmented block environment, residual variance was estimated exclusively from the replicated checks and incorporated into the least-square means, allowing the inclusion of unreplicated genotypes in the pooled analysis without inflating the residual degrees of freedom.

An analogous joint analytical procedure was also adopted to evaluate the effect of boll developmental thirds, replacing genotype means with adjusted means corresponding to each plant canopy third within each environment.

In addition, the genetic parameters of broad-sense heritability ($H²$), genetic variability (${\sigma ²}_g$), and environmental variability (${\sigma ²}_e$) were estimated. The means obtained were compared via the Tukey test (p ≤ 0.05), and Pearson correlation coefficients were estimated to quantify linear associations among continuous variables. Furthermore, the Euclidean distance was calculated, and subsequently, a dendrogram was developed with the conglomeration of the 17 genotypes evaluated.

RESULTS AND DISCUSSION

The use of a color card significantly affected the extraction of RGB-based indices (p ≤ 0.01), although no interaction was observed between genotype and calibration (Table S2). After brightness and color correction, BI and SI values increased, while NGRDI decreased, demonstrating that calibration alters index magnitudes. Therefore, the use of color cards is important to accurately estimate the characteristics and variability of color in colored cotton fiber, since calibration interfered with the intensity of the color shown in the images.

This effect occurs because RGB vegetation indices are highly sensitive to spectral and radiometric variations (Kefauver et al. 2020). Radiometric calibration corrects illumination and sensor distortions, improving data reliability (Galvíncio and Naue 2020), but it also modifies the relative intensity among RGB channels. For the present dataset, BI and NGRDI exhibited an inverse relationship relative to SI. BI was associated with white and lighter fiber genotypes, indicating greater sensitivity to higher luminosity, whereas SI was correlated with brown and reddish-brown fibers, reflecting higher chromatic intensity. NGRDI showed predominantly negative values in brown fibers due to the predominance of the red spectral component, while genotypes closer to white had values near 0.0, indicating lower contrast between green and red channels.

Significant differences were observed among genotypes for all evaluated indices (p ≤ 0.01), indicating substantial phenotypic variation in fiber color (Table S3). Additionally, a significant genotype × developmental stage interaction was detected for the SI, highlighting the potential of high-throughput phenotyping to identify genotypes with more saturated fiber color across different growth stages (Table 1).

Cotton fiber pigmentation is strongly influenced by environmental conditions such as solar radiation, drought, salinity, and irrigation, which affect fiber development and pigment accumulation (Soares et al. 2018). Light intensity and spectral composition also regulate cotton metabolism, altering the timing and magnitude of flavonoid biosynthesis (Fait et al. 2025, Rahmonova et al. 2025). Early light exposure may suppress flavonoid accumulation, whereas exposure at later stages can lead to enhanced pigment production (Zhang et al. 2022), contributing to spatial variability in fiber color within the same plant.

In naturally colored cotton, brown fiber is partially dominant over white, while other phenotypes may show white dominance (Carvalho et al. 2014). Brown coloration is mainly associated with tannins and proanthocyanidins, whose synthesis is influenced by light quality, especially red wavelengths, through regulation of the phenylpropanoid pathway (Shao-hua et al. 2005, Shao-hua et al. 2006).

Because fiber color depends on flavonoids, anthocyanins, and proanthocyanidins, RGB-based indices such as NGRDI, BI, and SI can indirectly estimate pigment dynamics (Clemente et al. 2023, Ma et al. 2024). Studies in other crops show strong associations between spectral indices and biochemical pigment content, supporting that RGB index variation in cotton likely reflects underlying shifts in pigment biosynthesis.

Analysis of fiber quality and color index across two seasons revealed genotype × environment (GxE) interactions for multiple traits including SCI, Mic, UHML, SF, SI, BI and NGRDI. This interaction indicates differential genotype performance across seasons.

Significant genetic variance was observed for all traits except for Mat. Environmental variance significantly affected Mic, SF, Elg, SI, BI, and NGRDI (Table S4). These fiber characteristics are known to be climate-dependent, making them susceptible to environmental modifications.

High broad-sense genotypic determination coefficient ($H²$ > 0.70) was observed for key fiber traits including Mic, UHML, UI, SF, Str, and Elg (Table S4). These high H² values indicate that genetic factors predominantly control the observed phenotypic variation. Such highly heritable traits are particularly valuable for selection in early-generation breeding populations (Bonifácio et al. 2015). Additionally, both BI and NGRDI showed moderately high genotypic determination coefficient values of 0.66 and 0.65, respectively. High $H²$ is associated with the dominant characteristic of fiber coloration, as fiber color is governed by a gene, GhTT2-3A, located on chromosome A07 (Zheng et al. 2024).

Wide variability was observed among the evaluated genotypes regarding the technological characteristics of the fibers. For UHML, Mic, SCI, and Str, BRS Topázio, SG 821, and BRS Jade were superior when compared to the other genotypes (Figures S1, S2, S3, and S6). UI was higher in UFU-13, SG 821, and BRS Jade, and lower in BRS Rubi, UFU-05, and UFU-11 (Figure S4). SF was higher for BRS Rubi and lower for SG 821 (Figure S5). Regarding elongation (Elg), UFU-10 and UFU-18 were superior, in contrast to BRS Jade (Figure S7). BI was higher in SG 821 and BRS Topázio and lower in BRS Rubi (Figure S8). As for SI and NGRDI, the highest values ​​were observed in BRS Rubi, UFU-18, and UFU-14, and the lowest in SG 821 (Figure S9 and S10).

The lower color intensity observed in the upper canopy third may be explained by differences in sunlight exposure. Considering that the bolls in the upper third are the last to open, they experience a shorter period of solar radiation exposure and consequently exhibit lower fiber color intensity (Carvalho et al. 2011). Differences in color intensity among canopy thirds were more evident when assessed using SI.

Principal Component Analysis (PCA) explained 72% of the total variation across two dimensions. SG 821 showed superior fiber quality, while BRS Jade and BRS Topázio clustered together with similar characteristics. Although BRS Rubi had the highest SI value, it exhibited lower overall fiber quality. Genotypes UFU-13, UFU-14, and UFU-18 showed traits similar to BRS Rubi but with improved fiber quality. Most remaining UFU genotypes were positioned near the center of the PCA plot, whereas BRS Verde appeared isolated due to shorter fiber length and distinct color traits (Figure S11).

Multivariate approaches such as PCA, canonical variable analysis, and clustering are commonly used to assess genetic divergence, with method selection depending on research objectives and data characteristics (Cruz et al. 1994). In breeding programs, dissimilarity estimates are essential for identifying suitable parents and maximizing variability in new populations (Cruz et al. 2011). Similar approaches using digital imaging and multivariate analysis successfully discriminated naturally colored cotton cultivars (Gonçalves et al. 2015).

Using half of the maximum genetic distance as a dendrogram cutoff (Figure S12), two groups were identified: (i) SG 821, BRS Jade, and BRS Topázio; and (ii) BRS Rubi and UFU 10, UFU 13, UFU 04, and UFU 18, characterized by more intense fiber coloration. Inter-group crosses may generate segregating populations with diverse fiber color intensity and quality, indicating substantial variability and promising selection potential.

Pearson's correlation analysis revealed strong associations between the tested indices and fiber quality parameters, including UHML, UI, SF, and Str. BI exhibited the highest positive correlation coefficient with these fiber quality traits, suggesting a potential relationship between fiber color and quality traits (Figure 2). It is important to note that colored cotton breeding programs are relatively recent compared to white cotton breeding efforts. Nevertheless, genotypes with lower SI, particularly SG 821, BRS Jade, and BRS Topázio, demonstrated superior fiber quality characteristics compared to BRS Rubi. These findings indicate the need for further research to better understand the underlying factors and correlations between fiber color and quality traits.

Figure 2
Correlation matrix between the brightness index (BI), saturation index (SI), and normalized green-red difference index (NGRDI) and the technological properties of naturally colored brown cotton fibers. SCI: spinning consistency index; Mic: micronaire; Mat: fiber maturity (%); UHML: upper half mean length (mm); UI: length uniformity (%); SF: short fiber index; Str: fiber strength (gf tex⁻¹); and Elg: fiber elongation (%).

Regression analyses between color indices and intrinsic fiber characteristics showed that brown fiber genotypes exhibited moderate coefficients of determination ($R²$ > 0.50) for the relationships between BI and SCI, UHML, SF, and Str. These strong linear associations indicate that spectral responses related to brightness are closely linked to important technological attributes of the fiber, supporting the use of BI as a promising non-destructive auxiliary estimator for the indirect phenotyping of fiber quality (Figure 3).

Figure 3
Linear regression of the spinning consistency index (SCI). (A) upper half mean length (UHML) (B), fiber strength (Str) (C), short fiber index (SF) (D), length uniformity (UI) (E), and micronaire (Mic) (F) as a function of the brightness index (BI) derived from RGB images for naturally colored brown cotton genotypes. Asterisks indicate statistical significance according to the F-test (* p ≤ 0.05; *** p ≤ 0.01).

High-throughput phenotyping has become an important resource for supporting decision-making in plant breeding programs by enabling rapid, accurate, and reproducible data acquisition. The findings of this study highlight the potential of RGB image-based phenotyping as an objective approach for characterizing naturally colored cotton fibers, representing an important step toward increasing the scalability of colored cotton production and commercialization. At the same time, continued efforts to improve the technological quality of naturally colored fibers remain essential to expand their acceptance and application within the textile industry.

Recent advances in image analysis, remote sensing, and machine learning have accelerated the development of predictive models for cotton fiber quality. Studies integrating RGB and multispectral imagery acquired from unmanned aerial vehicles with machine learning algorithms have demonstrated the feasibility of estimating key technological traits, such as upper half mean length, length uniformity, and micronaire, rapidly and nondestructively (Xu et al. 2023). These approaches reinforce the growing role of digital phenotyping as a complementary tool for cotton breeding.

RGB image-based phenotyping also provides a more objective alternative to conventional visual assessment of fiber color. Traditional evaluations rely on trained personnel and are inherently subject to evaluator variability and fatigue, which can reduce consistency over time. In contrast, standardized RGB image acquisition enables reliable color assessment with minimal operator expertise, reducing subjectivity while increasing reproducibility and operational efficiency.

Despite these advantages, successful implementation requires careful standardization of image acquisition conditions. Controlled and diffuse illumination, consistent image capture procedures, and the separate sampling of bolls from different canopy positions are critical to minimize environmental variation and ensure accurate color measurements.

Differences in fiber color intensity were observed among canopy positions. Fibers obtained from bolls located in the upper third of the plant exhibited approximately 6.25% lower color intensity than those collected from the middle and lower thirds. This pattern is likely associated with the shorter period of solar exposure between boll opening and harvest, which may reduce pigment development.

Among the evaluated color indices, BI exhibited strong associations with important technological fiber traits in brown fiber genotypes, with $R²$ of 0.61 for SCI, 0.56 for UHML, 0.53 for SF, and 0.55 for Str. These relationships suggest that BI has potential as an indirect indicator of fiber quality, although additional validation across larger populations and environments is needed before its routine application in breeding programs.

Future research should focus on developing prediction models specific to each fiber color class (white, green, and brown), as genotype grouping according to color may improve model accuracy and increase the efficiency of selecting superior breeding materials. Expanding the number of evaluated genotypes and environmental conditions will also contribute to improving the robustness and general applicability of image-based prediction models.

Overall, the results demonstrate that RGB image-based phenotyping is a promising, rapid, non-destructive, and scalable approach for the objective characterization of naturally colored cotton fibers. Among the evaluated indices, SI was the most effective for discriminating genotypes according to fiber color intensity, whereas BI showed the strongest associations with key technological fiber traits. Together, these indices provide valuable tools for supporting selection decisions in colored cotton breeding programs and establish a foundation for future advances in digital fiber phenotyping.

ACKNOWLEDGMENTS

This study was supported by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) - Finance Code 001. The authors thank the members of PROMALG for their support and collaboration, as well as AMIPA and Minas Cotton Laboratory for technical assistance in fiber analysis. Supplementary materials are available from the corresponding author upon request.

Data Availability Statement

The supplementary file and datasets used and analyzed during the current study are available from the corresponding author upon reasonable request.

REFERENCES

• ABRAPA - Associação Brasileira dos Produtores de Algodão2025 BI cotton Brazil. Available at <Available at https://cottonbi.com.br/ >. Accessed on February 18, 2026.
  » https://cottonbi.com.br/
• Bac C, Hemming J, Van Henten E2013 Robust pixel-based classification of obstacles for robotic harvesting of sweet-pepperComputers and Electronics in Agriculture 96:148-162
• Bonifácio DOC, Mundim FM, Sousa LB2015 Variabilidade genética e coeficiente de determinação em genótipos de algodoeiro quanto a qualidade de fibraRevista Verde 10:66-71
• Carvalho LP, Andrade FP, Silva Filho JL2011 Cultivares de algodão colorido no BrasilRevista Brasileira de Oleaginosas e Fibrosas 15:37-44
• Carvalho LP, Farias FJC, Rodrigues JIS2015 Selection for increased fiber length in cotton progenies from acala and nonacala typesCrop Science 55:985-991
• Carvalho LP, Farias FJC, Lima Lima, MMA MMA, Rodrigues JIS2014 Inheritance of different fiber color in cotton (Gossypium barbadense L.)Crop Breeding and Applied Biotechnology 14:256-260
• Clemente AA, Maciel GM, Siquierolli ACS, Gallis RBA, Luz JMQ, Sala FC, Pereira LC, Yada RY2023 Nutritional characterization based on vegetation indices to detect anthocyanins, carotenoids, and chlorophylls in mini-lettuceAgronomy 13:1403
• CONAB - Companhia Nacional de Abastecimento2025 AgroConab. Available at Available at https://www.conab.gov.br/info-agro/safras/graos Accessed on February 18, 2026.
  » https://www.conab.gov.br/info-agro/safras/graos
• Cruz CD, Carvalho SP, Vencovsky R1994 Estudos sobre divergência genética. II Eficiência da predição do comportamento de híbridos com base na divergência de progenitoresRevista Ceres 41:183-190
• Cruz CD, Ferreira FM, Pessoni LA2011 Biometria aplicada ao estudo da diversidade genética. Editora UFV, Viçosa, 620p.
• Escadafal R, Belghith A, Ben-Moussa H1994 Indices spectraux pour la télédétection de la dégradation des milieux naturels en Tunisie aride. In Symposium international sur les mesures physiques et signatures en télédétection. Centre National d’Étude Spatiale, Paris, p. 253-259
• Fait MER, Marani IKH, Ferro RCAM, Cabral JSR, Rosa RR, Silva FB, Ordoñez GAP, Oliveira NR, Miola VF, Cantão VCG2025 Espectros de luz fazem a diferença: o algodoeiro sob LED e luz naturalCaderno Pedagógico 22:e18476
• Federer WT1961 Augmented designs with one-way elimination of heterogeneityBiometrics 17:447-473
• Galvíncio JD, Naue CR2020 Estimation of NDVI with visible images (RGB) obtained with dronesJournal of Hyperspectral Remote Sensing 9:407-420
• Gonçalves MIS, Vilar WTS, Medeiros EP, Pontes MJC2015 A novel strategy for the classification of naturally colored cotton fibers based on digital imaging and pattern recognition techniquesAnalytical Methods 7:5869-5875
• Kefauver SC, Romero AG, Buchaillot ML, Vergara-Díaz O, Fernandez-Gallego JA, El-Haddad G2020 Open-source software for crop physiological assessments using high resolution RGB images. In IGARSS 2020 IEEE international geoscience and remote sensing symposium. IEEE, New York, p. 4359-4362
• Ma Z, Wei C, Wang W, Lin W, Nie H, Duan Z, Liu K, Xiao X2024 Non-destructive prediction of anthocyanin concentration in whole eggplant peel using hyperspectral imagingPeerJ 12:e17379
• Matias F, Caraza-Harter M, Endelman J2020 FIELDimageR: An R package to analyze orthomosaic images from agricultural field trialsThe Plant Phenome Journal 3:1-6
• Naoumkina M, Hinchliffe DJ, Thyssen GN2024 Naturally colored cotton for wearable applicationsFrontiers in Plant Science 15:1-7
• Penna JCV2005 Melhoramento do algodão. In Borém A (ed) Melhoramento de espécies cultivadas. Editora UFV, Viçosa , p. 15-53
• R Core Team2023 R: A language and environment for statistical computing. R Foundation for Statistical Computing, Vienna. <https://www.R-project.org/>.
• Rahmonova RB, Ravshanova MX, Shomurodov NP, Raxmatullayeva MA, Kenjayeva OF, Narzullayev ZO, Sadriddinova MA2025 Formation of the structure of naturally colored cotton fiberAIP Conference Proceedings 3268:040038
• Richardson AJ, Wiegand CL1977 Distinguishing vegetation from soil background informationPhotogrammetric Engineering and Remote Sensing 43:1541-1552
• Schneider C, Rasband W, Eliceiri K2012 NIH Image to ImageJ: 25 years of image analysisNature Methods 9:671-675
• Shao-hua Z, Yi L, Yong-ping C2005 A study on the stability of the pigment of natural brown cotton and its mechanismActa Laser Biology Sinica 15:1-6
• Shao-hua Z, Yi L, Yong-ping C, Zhengpeng L2006 Study on the stability and mechanism of brown pigment in natural colored cottonJournal of Laser Biology 15:354-358
• Soares LAA, Fernandes PD, Lima GS, Suassuna JF, Brito MER, Sá FVS2018 Growth and fiber quality of colored cotton under salinity management strategiesRevista Brasileira de Engenharia Agrícola e Ambiental 22:332-337
• Tucker CJ1979 Red and photographic infrared linear combinations for monitoring vegetationRemote Sensing of Environment 8:127-150
• USDA - United States Department of Agriculture2025 Production - Cotton. Available at Available at https://fas.usda.gov/data/production/commodity/2631000 Accessed on February 18, 2026.
  » https://fas.usda.gov/data/production/commodity/2631000
• Xu W, Yang W, Chen P, Zhan Y, Zhang L, Lan Y2023 Cotton fiber quality estimation based on machine learning using time series UAV remote sensing dataRemote Sensing 15:1-19
• Zhang Q, Ruan J, Mumm R, de Vos RCH, Liu MY2022 Dynamic changes in the antioxidative defense system in the tea plant reveal the photoprotection-mediated temporal accumulation of flavonoids under full sunlight exposurePlant and Cell Physiology 63:1695-1708
• Zhao Y, Gong L, Zhou B, Huang Y, Liu C2016 Detecting tomatoes in greenhouse scenes by combining AdaBoost classifier and color analysisBiosystems Engineering 148:127-137
• Zheng L, Zhang J, He H, Meng Z, Wang Y, Guo S, Liang S2024 Anthocyanin gene enrichment in the distal region of cotton chromosome A07: mechanisms of reproductive organ colorationFrontiers in Plant Science 15:1-12
• Zonta JH, Bezerra JRC, Sofiatti V, Farias FJC, Carvalho LP2015 Efeito da irrigação no rendimento e qualidade de fibras em cultivares de algodoeiro herbáceoCaatinga 28:43-52

Correspondence

Corresponding author: [e-mail]

SCIENTIFIC EDITOR:

Luiz Antônio dos Santos Dias

Publication Dates

• Publication in this collection
  05 Oct 2026
• Date of issue
  2026

History

• Received
  06 Mar 2026
• Accepted
  22 June 2026
• Published
  16 July 2026