Application of the YOLO framework in the classification of vigor levels in buffelgrass (Cenchrus ciliaris L.) seeds evaluated by tetrazolium testABSTRACT:
Buffelgrass stands out as a key forage species in the semi-arid regions of Brazil, particularly in northern Minas Gerais, due to its capacity to increase pasture productivity and enhance animal performance. Seed viability is traditionally assessed using the tetrazolium test, a reliable but time-consuming and manual method. In this context, computational techniques such as the YOLO framework - a state-of-the-art algorithm for object detection and classification - have shown great potential in seed analysis. This study proposes the classification of buffelgrass seed vigor levels through image-based analysis using version 11 of the YOLO framework. Four seed lots were evaluated through tests of germination, first count, seedling emergence, accelerated aging, and tetrazolium analysis. Mean values were compared using the Tukey test at 5% significance level. In addition, Pearson’s correlation coefficients were calculated among the tests. The seeds were categorized into three vigor levels based on the results obtained from the tetrazolium test, and four classes for training the classification model. For model training, the YOLOv11 algorithm was employed, using digitized laboratory images that underwent feature extraction, balancing, and data augmentation procedures. Model performance was evaluated using top-1 accuracy, as well as analyses of training and validation losses. The model showed excellent performance, achieving 100% accuracy in testing, confirming the potential of this computational approach.
Index Terms:
artificial intelligence; characterization; semi-arid; tetrazolium
RESUMO:
O capim-buffel destaca-se como espécie forrageira fundamental nas regiões semiáridas do Brasil, especialmente no norte de Minas Gerais, por aumentar a produtividade das pastagens e melhorar o desempenho animal. A viabilidade de sementes é tradicionalmente avaliada pelo teste de tetrazólio, método confiável, mas manual e demorado. Nesse contexto, métodos computacionais, como o framework YOLO, algoritmo de ponta em detecção e classificação de objetos, vêm se mostrando promissores na análise de sementes. Este trabalho propõe a classificação de níveis de vigor de sementes de capim-buffel por meio de imagens utilizando a versão 11 do framework YOLO. Foram utilizados quatro lotes de sementes, submetidos a testes de germinação, primeira contagem de germinação, emergência de plântulas, envelhecimento acelerado e tetrazólio. As médias foram comparadas pelo teste de Tukey (5% de significância). Além disso, foram calculados coeficientes de correlação de Pearson entre os testes. As sementes foram caracterizadas em três níveis de vigor, de acordo com os resultados obtidos no teste de Tetrazólio, e quatro classes para o treinamento do modelo de classificação. Para o treinamento, empregou-se o YOLOv11, com imagens digitalizadas em laboratório e submetidas a técnicas de extração, balanceamento e aumento de dados. O desempenho foi avaliado pela acurácia top-1, além da análise das perdas de treinamento e validação. O modelo apresentou excelente desempenho, alcançando 100% de acurácia nos testes, confirmando o potencial da abordagem.
Termos para Indexação:
inteligência artificial; caracterização; semiárido; tetrazólio
INTRODUCTION
Pastures represent an essential component of the livelihood and economic sustenance of farmers living in the Brazilian semi-arid region, which is marked by severe climatic limitations and restrictions on livestock production (Coêlho et al., 2021). In this scenario, buffelgrass (Cenchrus ciliaris L.) stands out as a strategic alternative to forage species less adapted to water scarcity, especially in the north of Minas Gerais, contributing to increased pasture productivity and better animal performance (Cunha et al., 2023; Alves and Ferreira, 2024).
To ensure the efficiency of pasture areas, it is essential to assess the quality of the seeds used in pasture formation, in order to guide grazing goals and improve decision-making in the production system. In the case of buffelgrass, this assessment becomes even more relevant, since its seeds have natural dormancy that can last from 6 to 12 months, making the initial establishment of the pasture difficult and requiring specific care regarding storage and sowing planning (Cunha et al., 2023).
Seed quality assessment is of utmost importance in the agricultural production program (Krzyzanowski et al., 2020). Classification of seed viability and vigor is essential to estimate crop performance. In this context, several studies have been conducted with the objective of developing methodologies for seed characterization in relation to these parameters. Some of these methodologies have been incorporated into the Rules for Seed Testing - RAS (Brasil, 2025). However, there are still species, such as buffelgrass, that lack standardized protocols, which makes it difficult to carry out consistent analyses of viability and vigor.
Among the tests that are conducted in the laboratory to assess the viability and vigor of seeds (Nucci et al., 2023), the tetrazolium test stands out for using the tetrazolium salt (2,3,5-triphenyl tetrazolium chloride) to determine the physiological potential of seeds based on the respiratory activity in the cells that make up their tissues (Brasil, 2025). One of the main critical points of the tetrazolium test is that the analysis of the results depends on human judgment, which makes the process time-consuming and highly dependent on the expertise of specialists (Krzyzanowski et al., 2020).
The use of computational methods in seed analysis has advanced consistently in recent decades. An example is the study conducted by Musaev et al. (2022), who used X-ray images associated with automated digital morphometric analysis to assess the quality of seeds of species of the Brassicaceae family, seeking to identify latent internal defects and their relationship with seed vigor and germination. Fernandes et al. (2023) demonstrated that convolutional neural networks can effectively assist sweet potato phenotyping, allowing the identification of qualitative attributes such as shape, peel color, and damage in digital images, with precision and reduction in the time of visual analyses.
Other studies point to the efficiency of using computational methods to analyze the results of the tetrazolium test, as is the case of the work with guavira seeds (Nucci et al., 2023) and Brachiaria brizantha (Custódio et al., 2012). Among the computational methods, object detection has advanced rapidly, with increasingly accurate and efficient algorithms. Among these methods, the YOLO (You Only Look Once) framework from Ultralytics stands out, which adopts a single-stage detection and classification approach, combining high accuracy and real-time performance. In its 11th version, YOLOv11 (Ultralytics, 2025a) improved on these principles with architectural advances and increased processing efficiency.
Some studies contemplate the efficiency of YOLO in seed image analysis, such as the use of the tool for detecting impurity content in machine-picked seed cotton (Zhang et al., 2021) and the efficiency of the tool in segregating healthy maize seeds from those with some type of disease (Kundu et al., 2021).
Thus, this study aims to characterize the vigor of buffelgrass seed lots and train a classification model of the YOLO framework for the automatic recognition of vigor levels, based on images of the tetrazolium test.
MATERIAL AND METHODS
The experiment was conducted in three stages, the first at the Seed Analysis Laboratory of the Universidade Estadual de Montes Claros (Unimontes), campus of Janaúba, MG, and the others at the Research Laboratory of the Federal Institute of Northern Minas Gerais (IFNMG), campus of Porteirinha, MG, Brazil. Seeds of buffelgrass cv. Aridus, 2023/2024 season, were acquired from four different companies in the local market of Janaúba, without identification of the producer or origin, composing four independent lots. The seeds were homogenized and packed in multiwall paper bags and stored under ambient conditions (~26 °C) for 7 days, until the experiments were conducted.
Characterization of the physiological potential of the lots
The first stage consisted of the characterization of the physiological potential of the lots, through tests of germination, first germination count, seedling emergence and accelerated aging, in a completely randomized design.
The germination test was performed with five replicates of 50 seeds per lot, sown in plastic gerbox boxes, containing Germitest® paper moistened with distilled water in a volume equivalent to 2.5 times the weight of the dry paper. The boxes were kept in a germination chamber at 30 °C, with a photoperiod of 12 hours, and the evaluations of normal seedlings were made weekly for up to 28 days, according to the criteria of the Rules for Seed Testing (Brasil, 2025). The first germination count was recorded on the seventh day, obtained by counting the number of normal seedlings.
The seedling emergence test was conducted under laboratory conditions (± 26 °C), with five replicates of 50 seeds sown in sterilized sand, at 2 mm depth measured with a millimeter ruler and irrigated daily to maintain 50% of the retention capacity. The evaluation was carried out at 28 days, by counting the number of emerged seedlings.
Accelerated aging test followed the method of Marcos-Filho (2020), in which the seeds were distributed in a single layer on an aluminum screen in gerbox boxes with 40 mL of distilled water, kept in a BOD-type incubator at 41 °C for 72 hours. After this period, five replicates of 50 seeds per lot were subjected to the germination test and evaluated on the seventh day.
Tetrazolium test
The second stage aimed to characterize the vigor levels of the seeds by means of the tetrazolium test. For this, a completely randomized design was adopted with four lots and four replicates of 50 seeds.
Initially, palea and lemma were removed, according to Agüero et al. (2017), and then the seeds were preconditioned between moistened Germitest® paper sheets (2.5 times the dry weight) and kept for 18 hours at 20 °C, as recommended for buffelgrass and other forages (Agüero, 2019; Brasil, 2025). Subsequently, the seeds were longitudinally sectioned, submerged in 0.25% tetrazolium solution and kept in BOD at 30 °C for 10 hours in the dark, a methodology defined from preliminary tests conducted during the research.
After staining, the seeds were washed in running water and kept in distilled water until evaluation, to prevent them from drying. The analysis was performed under a stereoscopic microscope, considering the integrity of radicle, coleorhiza, coleoptile and scutellum, essential structures for germination, nutrition and initial development of the seedling (Nakamura and Scatena, 2009; Silva, 2019). According to Krzyzanowski et al. (2020), the seeds were classified into three levels: viable and vigorous (Figure 1), characterized by uniform staining in all parts of the embryo, without lesions or discolored areas; viable and not vigorous (Figure 2), with partial necrosis or discoloration, but preserved essential structures; and non-viable (Figure 3), when they showed necrosis or absence of staining in the essential structures.
Representation of viable and vigorous Seeds: (A; B; C). Seeds fully stained with carmine red. Janaúba, MG, Brazil, 2025.
Representation of viable and non-vigorous seeds: (A) Seed with discolored areas in 1/3 of the radicle from the end, (B) Seed with discolored areas in 1/4 of the distal part of the scutellum; (C) Seed with plumule and radicle with intense red color. Janaúba, MG, Brazil, 2025.
Representation of non-viable seeds: (A) Fully necrotic seed; (B) Seed showing discolored plumule and coleoptile; (C) Fully discolored seed. Janaúba, MG, Brazil, 2025.
The results were subjected to analysis of variance and, after verifying the assumptions of normality and homogeneity, the means were compared by Tukey test at 5% significance level, using R software version 4.5.1 (R Core Team, 2025). Subsequently, Pearson’s correlation coefficients were calculated between the results of the tetrazolium test, germination, seedling emergence and vigor tests, specifically first germination count and accelerated aging.
Classification model
The third stage was aimed at developing a classification model based on the YOLO framework, version 11, with the objective of classifying the vigor levels of buffelgrass seeds based on images of seeds stained by the tetrazolium. To this end, image acquisition, pre-processing, training, validation and testing procedures were carried out.
The images were obtained by scanning the seeds subjected to the tetrazolium test, positioned in halves with the inner part facing the scanner glass, because evaluating the two halves of the seed is necessary for this species (Krzyzanowski et al., 2020), according to lot and replicate. Scanning was performed in an HP Deskjet Ink Advantage 2676 scanner, with a resolution of 1200 dpi, resulting in 16 images in JPG format, each containing 50 seeds, totaling 800 scanned seeds. Subsequently, each seed was manually extracted from the original images, cut out and stored as an individual file, forming a set of 800 images.
The Roboflow tool (Roboflow, 2025) was used to label the seeds according to vigor levels, according to Krzyzanowski et al. (2020), as follows: viable and vigorous, viable and not vigorous and non-viable. Considering the initial imbalance between classes, which was 169 viable and vigorous seeds, 118 viable and non-vigorous seeds, and 513 non-viable seeds, the non-viable category was subdivided into two: non-viable seeds partially stained or with necrosis in their structures considered essential (Non-viable A), and totally discolored non-viable seeds (Non-viable B); thus, all classes were balanced with 100 samples, obtaining representations of all the lots in each class. Subsequently, the dataset was divided into 70% for training, 20% for validation, and 10% for testing. Next, resizing and normalization techniques were used for compatibility with the model, in addition to augmentation techniques such as rotation, clipping, inversion, and segment combinations, to increase the diversity of data in the training base.
The model was trained in python language in Google Colab with a 12 GB NVIDIA Tesla GPU, using the AdamW optimizer, with an initial learning rate of 0.00125, momentum of 0.9, in addition to rate decay and early stop strategies to reduce overfitting. During the initial tests, the algorithm went through successive iterations of tuning and optimization, with the batch size being defined as 16 and 100 training epochs. Performance was evaluated by the Top-1 accuracy metric, in addition to monitoring training and validation losses, with:
where T1 corresponds to the number of correct predictions and N corresponds to the total number of samples (Pei et al., 2025; Ultralytics, 2025b).
RESULTS AND DISCUSSION
The germination, first germination count, seedling emergence and accelerated aging tests revealed significant differences in seed quality among the lots analyzed (Table 1). The germination percentages showed significant variations between the lots, and lot 4 stood out with the highest germination percentage (35%), followed by lot 1 (30%). On the other hand, lots 3 and 2 showed lower performances. These low values may be an indication of low physiological quality of the seeds (Souza and Smiderle, 2024). It is observed that only seeds from lot 4 had values above 30%, which is the minimum standard required for the commercialization of buffelgrass seeds, as established by the Ministry of Agriculture, Livestock and Food Supply (Brasil, 2008).
The results of the first germination count indicated superiority of lots 1 and 4, both with 22%, compared to lots 2 and 3, which had 2% and 9%, respectively. This trend was also observed in the seedling emergence test, in which lots 1 and 4 obtained the highest values (23% and 28%), while lots 2 and 3 had lower results. The first germination count and seedling emergence tests reinforced the differences in physiological potential between the analyzed lots, with the first count reflecting the most vigorous seeds (Krzyzanowski et al., 2020).
As observed in the other tests, in the accelerated aging test, the seeds of lot 4 stood out as more vigorous (8%), followed by those of lot 1 (4%). On the other hand, the seeds of lots 2 and 3 practically did not resist the conditions imposed by the test, with values of 0% and 1%, respectively. These results show the effectiveness of the accelerated aging test in detecting differences in vigor between seed lots (Mathias and Coelho, 2021), especially when distinguishing lots 1 and 4, which, despite having similar germination, showed different behaviors under stress conditions. The main objective of evaluating seed vigor is to detect variations in physiological potential between lots, so choosing the right method is essential for accurate evaluation. Thus, it is important to combine different approaches to obtain a more reliable diagnosis (Meneguzzo et al., 2021; Smiderle and Souza, 2021).
The second stage, aimed at characterizing the vigor levels of buffelgrass seeds through the tetrazolium test, revealed significant differences between the lots evaluated regarding the proportion of viable and vigorous, viable and non-vigorous and non-viable seeds (Table 2). In general, the results showed that the percentages of viable and vigorous seeds varied between the lots, with the seeds from lot 4 standing out with the highest values, while the seeds from lot 2 were classified as having lower physiological quality. For viable and non-vigorous seeds, a similar pattern was observed, with the highest percentage for lot 4, followed by lot 1, while lots 2 and 3 had the lowest values. These results corroborate those observed in the tests for initial characterization of the lots, and the differences are possibly attributed to factors such as the natural dormancy characteristic of the species, and/or the deterioration of tissues after physiological maturity, which is conditioned to genetic and environmental factors, such as the practices adopted in the management of seed lots (Krzyzanowski et al., 2020).
As for the non-viable seeds, it was possible to identify differences in the viability of the lots tested, with the lowest percentages of non-viable seeds being observed in lots 1 and 4 and the highest percentages in the seeds from lots 2 and 3, being characterized as the one with the worst performance (Table 2). This lower physiological quality in the seeds of lots 2 and 3 is possibly associated with failures in the production processes, such as inadequate harvesting, inefficient processing or inadequate storage conditions (Souza and Smiderle, 2024).
Pearson’s correlation between the results of germination, seedling emergence, first germination count, accelerated aging and the proportion of viable and vigorous seeds (Figure 4) showed a positive association between the tests, demonstrating coherence in the estimation of the vigor of the lots, reinforcing that the quality of the seed, which is determined by genetic, physical, physiological and sanitary attributes, directly influences its germination and performance in the field. High-vigor seeds have well-developed structures and production potential, ensuring vigorous plants in the shortest possible time under different conditions (Rossetti et al., 2023).
Pearson’s simple correlation between the variables germination (G), seedling emergence (SE), first germination count (FC), accelerated aging (AA) and viable and vigorous seeds (VV). Janaúba -MG, 2025.
The training of the classification model demonstrated good performance in the classification of vigor and viability levels for buffelgrass seeds. Figure 5 presents the confusion matrices of the validation (Figure 5A) and test (Figure 5B) datasets, both normalized. The validation and testing of the trained model showed 100% accuracy in the 4 classes of seeds, revealing that the model can differentiate the viability and vigor of the seeds. This high accuracy may indicate exceptional performance of the model or a possible overfitting (Santos et al., 2022), but the latter hypothesis can be ruled out because the model was trained until the last epoch, with no need to use the early stop feature during training. In addition, the results of loss during training and validation consistently decrease to approximately 0.2 and 0.4 (Figure 6), respectively, indicating that the model learned until the last epoch with a smoothing of the results on a stable trajectory of decline. The Top-1 accuracy metric reached the maximum score, showing excellent performance, when compared to recent studies that also applied architecture derived from YOLO. For instance, Ghate et al. (2025) obtained high performance in the classification of arecanuts, and Pei et al. (2025) improved the identification of hydrophobicity in electrical insulators through an optimized version of YOLO. The performance of the YOLO model showed high potential in identifying the vigor and viability levels determined by the tetrazolium test, suggesting that object classification techniques can speed up and standardize laboratory evaluations of seeds. Despite the promising results, the small size of the image database and the use of samples from a single cultivar may limit the generalization capacity of the model, indicating the need to expand the database and compare different neural network architectures in future studies. Even so, the results confirmed YOLO v11 as a promising tool for automating physiological analyses and supporting decisions related to seed certification and commercialization.
Graphs of training loss (A), validation loss (B), Top-1 accuracy (C) as a function of the epochs. The X-axis represents the training epochs, while the Y-axis represents the loss or accuracy values. Janaúba, MG, Brazil, 2025.
CONCLUSIONS
The YOLO machine learning model, trained to classify the vigor levels of Cenchrus ciliaris L., showed high performance in the validation and test data, proving to be a promising tool to support the analysis of the physiological quality of seeds.
ACKNOWLEDGMENTS
The authors thank the Universidade Estadual de Montes Claros (Unimontes) and the Instituto Federal do Norte de Minas Gerais (IFNMG) for providing the infrastructure and support essential for the development of this research. This study was also supported by the Fundação de Amparo à Pesquisa do Estado de Minas Gerais (FAPEMIG) and the Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq).
REFERENCES
-
AGÜERO, C.G. Ajuste de las especificaciones técnicas del test de viabilidad por tetrazolio en semillas de Cenchrus ciliaris L. (buffel grass). AgriScientia, v.36, n.2, p.81-88, 2019. https://revistas.unc.edu.ar/index.php/agris/article/view/24833
» https://revistas.unc.edu.ar/index.php/agris/article/view/24833 - AGÜERO, C.G.; PEREYRA, G.R.; ROLANDO, R.O. Método alternativo de germinación para determinar calidad de semillas en Buffel Grass (Cenchrus ciliaris L.). AgriScientia, v.34, n.1, p.47-58, 2017.
-
ALVES, R.F.; FERREIRA, G.H.C. Delimitação e diagnóstico de situação de áreas de preservação permanente no semiárido mineiro. Revista Geonorte, v.15, n.50, p.222-238, 2024. https://doi.org/10.21170/geonorte.2024.V.15.N.50.222.238.
» https://doi.org/https://doi.org/10.21170/geonorte.2024.V.15.N.50.222.238 -
BRASIL. Instrução Normativa nº 30 de 21.05.2008. Estabelece normas e padrões para produção e comercialização de sementes de espécies forrageiras de clima tropical. Diário Oficial da União, Brasília, 2008. https://www.gov.br/agricultura/pt-br/assuntos/insumos-agropecuarios/insumos-agricolas/sementes-e-mudas/publicacoes-sementes-e-mudas/INN30de21demaiode2008.pdf
» https://www.gov.br/agricultura/pt-br/assuntos/insumos-agropecuarios/insumos-agricolas/sementes-e-mudas/publicacoes-sementes-e-mudas/INN30de21demaiode2008.pdf -
BRASIL. Regras para Análise de Sementes (RAS) Ministério da Agricultura e Pecuária - MAPA, Brasília, 2025. https://wikisda.agricultura.gov.br/pt-br/Laborat%C3%B3rios/Metodologia/Sementes/RAS_2025/Amostragem
» https://wikisda.agricultura.gov.br/pt-br/Laborat%C3%B3rios/Metodologia/Sementes/RAS_2025/Amostragem -
COÊLHO, J.J.; COSTA, J.S.; SILVA, R.R.; SANTOS, E.R.S. Exotic forage grasses in the Brazilian semi-arid region: forage mass, competitiveness, and botanical similarities of non-native pastures. Journal of Arid Environments, v.193, p.104551, 2021. https://doi.org/10.1016/j.jaridenv.2021.104551.
» https://doi.org/https://doi.org/10.1016/j.jaridenv.2021.104551 - CUNHA, M.L.P.; SOUZA, D.C.F.S.; SALES, E.C.J.; MONÇÃO, F.P. Capim-buffel: forrageira para áreas secas. Montes Claros: Suprema Gráfica, 2023. 52p.
-
CUSTÓDIO, C.C.; DAMASCENO, R.L.; MACHADO NETO, N.B. Imagens digitalizadas na interpretação do teste de tetrazólio em sementes de Brachiaria brizantha Revista Brasileira de Sementes, v.34, n.2, p.334-341, 2012. https://doi.org/10.1590/S0101-31222012000200020.
» https://doi.org/https://doi.org/10.1590/S0101-31222012000200020 -
GHATE, D.D.; HEGDE, M.; PAI, R. Advancing arecanut quality grading: a comparative analysis of YOLO models with hyperparameter optimization. Research Square [Preprint], 2025. https://doi.org/10.21203/rs.3.rs-5755373/v1.
» https://doi.org/https://doi.org/10.21203/rs.3.rs-5755373/v1 -
FERNANDES, A.C.G.; BRITO, L.F.; SILVA, L.J.; NASCIMENTO, W.M. Convolutional neural networks in the qualitative improvement of sweet potato roots. Scientific Reports, v.13, n.8429, p.1-8, 2023. https://doi.org/10.1038/s41598-023-34375-6.
» https://doi.org/https://doi.org/10.1038/s41598-023-34375-6 - KRZYZANOWSKI, F.C.; FRANÇA-NETO, J.B.; GOMES-JUNIOR, F.G.; NAKAGAWA, J. Vigor de sementes: conceitos e testes Londrina: ABRATES, 2020. 601p.
-
KUNDU, N.; RANI, G.; DHAKA, V.S. Seeds classification and quality testing using deep learning and YOLO v5. In: INTERNATIONAL CONFERENCE ON COMPUTING CONCEPTS, 2021. Proceedings of the International Conference on Data Science, Machine Learning and Artificial Intelligence New York: ACM, 2021. p.1-5. https://doi.org/10.1145/3484824.3484913.
» https://doi.org/https://doi.org/10.1145/3484824.3484913 - MARCOS-FILHO, J. Teste de envelhecimento acelerado. In: KRZYZANOWSKI, F.C. et al. (Ed.). Vigor de sementes: conceitos e testes Londrina: ABRATES , 2020. p.185-246.
- MATHIAS, V.; COELHO, C.M.M. Correlation between vigor by accelerated aging at pre-sowing and soybean seedling emergence in the field. Semina: Ciências Agrárias, v.42, n.2, p.455-470, 2021.
-
MENEGUZZO, M.R.R.; MARTIN, T.N.; NUNES, U.R.; SAVIAN, T.V. Seedling length and soybean seed vigor. Ciência Rural, v.51, n.7, e20190495, 2021. https://doi.org/10.1590/0103-8478cr20190495.
» https://doi.org/https://doi.org/10.1590/0103-8478cr20190495 -
MUSAEV, F.; BUKHAROV, A.; BALEKHOV, O.; IVANOVA, M.; KASHLEVA, A. Assessment of Brassicaceae seeds quality by X-ray analysis. Horticulturae, v.8, n.1, p.29, 2021. https://doi.org/10.3390/horticulturae8010029.
» https://doi.org/https://doi.org/10.3390/horticulturae8010029 -
NAKAMURA, A.T.; SCATENA, V.L. Desenvolvimento pós-seminal de espécies de Poaceae (Poales). Acta Botanica Brasilica, v.23, n.1, p.212-222, 2009. https://doi.org/10.1590/S0102-33062009000100023.
» https://doi.org/https://doi.org/10.1590/S0102-33062009000100023 -
NUCCI, H.H.P.; SILVA, C.B.; CANDIDO, A.C.S.; ALVES, C.Z. Use of computer vision to verify the viability of guavira seeds treated with tetrazolium salt. Smart Agricultural Technology, v.5, 100239, 2023. https://doi.org/10.1016/j.atech.2023.100239.
» https://doi.org/https://doi.org/10.1016/j.atech.2023.100239 -
PEI, S.; ZHANG, Y.; LI, X.; WANG, Z. Detection of hydrophobicity grade of insulators based on AHC-YOLO algorithm. Scientific Reports , v.15, n.1, p.9673, 2025. https://doi.org/10.1038/s41598-025-92696-0.
» https://doi.org/https://doi.org/10.1038/s41598-025-92696-0 -
R CORE TEAM. R: A language and environment for statistical computing Vienna: R Foundation for Statistical Computing, 2025. Available at: https://www.R-project.org/
» https://www.R-project.org/ -
ROBOFLOW INC. Annotation tools. 2025. Online technical documentation. http://docs.roboflow.com/annotate/annotation-tools
» http://docs.roboflow.com/annotate/annotation-tools - ROSSETTI, C.; SOUSA, P.G.F.; SILVA, J.B. Gestão dos processos para produção de sementes: do campo a pós-colheita Vol. 2: controle de qualidade. Cuiabá: Pantanal Editora, 2023. 150p.
-
SANTOS, M.D.X.S.; CASTRO, P.A.; PAPA, J.P. Algoritmos de aprendizado de máquina para classificação de células nucleadas do sangue periférico: uma experiência do projeto Hemovision. In: ESCOLA REGIONAL DE INFORMÁTICA DE GOIÁS (ERI-GO), 10., 2022. Anais... Porto Alegre: Sociedade Brasileira de Computação, 2022. p.213-222. https://doi.org/10.5753/erigo.2022.227698.
» https://doi.org/https://doi.org/10.5753/erigo.2022.227698 - SILVA, G.R. Produção, tecnologia e armazenamento de sementes Londrina: Editora e Distribuidora Educacional S.A., 2019. 216p.
-
SMIDERLE, O.J.; SOUZA, A.G. Do scarification and seed soaking periods promote maximum vigor in seedlings of Hymenaea courbaril? Journal of Seed Science, v.43, e202143003, 2021. https://doi.org/10.1590/2317-1545v43243886.
» https://doi.org/https://doi.org/10.1590/2317-1545v43243886 -
SOUZA, A.G.; SMIDERLE, O.J. Deterioração e características de qualidade e vigor de sementes. In: DIAS, M. (Ed.). Ciências Agrárias: tecnologia, sustentabilidade e inovação Ponta Grossa: Editora Científica Digital, 2024. v.1, p.128-143. https://doi.org/10.1590/2317-1545v43254481.
» https://doi.org/https://doi.org/10.1590/2317-1545v43254481 -
ULTRALYTICS. Object detection 2025a. https://www.ultralytics.com/pt/glossary/object-detection
» https://www.ultralytics.com/pt/glossary/object-detection -
ULTRALYTICS. Accuracy 2025b. https://www.ultralytics.com/pt/glossary/accuracy Accessed on: Dec. 8, 2024.
» https://www.ultralytics.com/pt/glossary/accuracy -
YASMINE, G.; MAHA, G.; HICHAM, M. Overview of single-stage object detection models: from YOLO v1 to YOLOv7. In: INTERNATIONAL WIRELESS COMMUNICATIONS AND MOBILE COMPUTING (IWCMC), 2025b. Proceedings... IEEE, 2023. p.1-6. https://doi.org/10.1109/IWCMC58020.2023.10182423.
» https://doi.org/https://doi.org/10.1109/IWCMC58020.2023.10182423 -
ZHANG, C.; LI, T.; ZHANG, W. The detection of impurity content in machine-picked seed cotton based on image processing and improved YOLO V4. Agronomy, v.12, n.1, p.66, 2021. https://doi.org/10.3390/agronomy12010066.
» https://doi.org/https://doi.org/10.3390/agronomy12010066
Additional data will be made available by the authors upon reasonable request.
