JAJuan José Tapia Armenta

Juan José Tapia Armenta

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Juan José Tapia Armenta has a total of 37 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Spectral Metrics for Differentiating Mercury Surface Units from MASCS/VIRS VIS–NIR Data

    by Ana Victoria Ojeda Vera, Sonia Fornasier, Carlos Alberto Guerrero Peña, Juan J. Tapia Armenta, Luis Hector Sanchez Gamez

    We present a methodology for computing and evaluating spectral metrics from reflectance spectra of Mercury’s surface, demonstrated using NASA’s MESSENGER MASCS/VIRS data covering 300–1450 nm range. A set of spectral descriptors was derived from established formulations, including curvature metrics, spectral slopes, color indices, reflectance values, and diagnostic band depths. The analysis was applied to a dataset of 130 spectra sampling four representative surface units: Low Reflectance Material (LRM), High Reflectance Plains (HRP), Low Brightness Plains (LBP), and Hollows. Metric relevance was evaluated through variance-based statistical analysis, Principal Component Analysis (PCA), and supervised classification. Results show that curvature-related metrics dominate the spectral variability and provide the strongest discrimination between surface units, with ultraviolet curvature emerging as the most effective single parameter. Spectral slopes exhibit intermediate discriminative capability, while color indices show higher variability than reflectance but limited standalone separation. PCA confirms that the selected metrics provide strong separation in reduced-dimensional space, with Hollows forming a fully distinct cluster and only minor overlap among LRM, HRP, and LBP. A supervised multi-class classification test using the highest-ranked spectral descriptors further yielded near-complete separability, with a balanced accuracy of 0.9953. In contrast, diagnostic band-depth parameters show limited effectiveness for this dataset with no visually prominent absorptions. Overall, the results identify curvature-based metrics as the most reliable indicators for differentiating Mercury surface units in VIS–NIR spectra, while showing that slope, color-index, and selected near-infrared reflectance descriptors provide useful complementary information.

  • A comparative machine learning framework for stellar subdwarf classification using spectral metrics and feature engineering

    by Ana V. Ojeda, Christopher A. Theissen, C. A. Guerrero, Juan J. Tapia, T. Zhou

    Automated classification of spectra from low-mass stars and brown dwarfs requires methods that remain reliable not only for spectral-type assignment but also for gravity- and metallicity-sensitive peculiar classes. This study revisits the automated classification of M0–T9 dwarfs observed with the SpeX prism spectrograph (R ∼ 120, 0.8–2.5 μm), with emphasis on peculiar classes associated with surface gravity and metallicity. Rather than optimizing primarily for spectral type, the model-selection procedure prioritizes peculiar-class balanced accuracy (BAcc) to reduce the effect of the uneven representation of field, β, γ, d/sd, sd, and esd objects. A 90-dimensional input vector was constructed from binned spectral fluxes, targeted spectral slopes, the apparent J- and H-band magnitudes, and residual features measured relative to field standards of the same spectral type. Random forest (RF), support vector machine (SVM), K-nearest neighbors (KNN), and one-dimensional convolutional neural network (CNN) models were trained on 66,000 synthetic spectra, and their configurations were selected using a validation set of 396 spectra. Final performance was measured on an independent observed test set containing 1,300 spectra. KNN achieved the strongest overall performance, with 89.84% peculiar-class BAcc, 88.08% peculiar-class accuracy, and spectral-type accuracies of 96.85% and 99.54% within ±1 and ±3 spectral types, respectively. RF ranked second in peculiar-class BAcc, whereas CNN produced the lowest peculiar-class BAcc and SVM obtained the lowest spectral-type accuracy within ±3 spectral types. The RF feature-importance analysis identified the field-referenced features Di, Ai, and Ki, the apparent J- and H-band magnitudes, several near-infrared spectral slopes, and spectral bins near 0.89–0.99 μm as the most informative variables for peculiar-class classification. These findings show that KNN provides the most effective class-balanced solution for the proposed representation, while RF offers complementary interpretability of the spectral and photometric information driving the classification.

  • Traffic Sign Classification Using Real-Time GPU-Embedded Systems

    by Miguel Lopez-Montiel, Ulises Orozco-Rosas, Moisés Sánchez-Adame, Oscar Montiel, Kenia Picos, Juan J. Tapia


  • Artificial Intelligence in Astrophysics: Analytical Enhancements

    by Luis H. Sanchez, Carlos A. Guerrero, Juan J. Tapia, Ana V. Ojeda

  • Calibration of wavefront aberrations using supervised learning and deep learning for a Shack–Hartmann wavefront sensor

    by Mario Alberto Jara Reyna, Carlos Alberto Guerrero Peña, Joel Herrera Vázquez, Juan José Tapia Armenta, Eduardo Gerardo Pérez Tijerina

    We evaluate wavefront fitting methods in single-sample and multisample scenarios, focusing on least squares (LSQ), gradient descent, and dynamic hypernetwork approaches. LSQ is a well-established method, whereas gradient descent offers a low-cost, adaptable solution. By contrast, the dynamic hypernetwork achieves the highest precision, with a mean absolute error of 2.9559×10−5λ, significantly lower than the 2.1909×10−4λ obtained by LSQ. In addition, it reduces computation time to 0.0273 ms per sample, outperforming LSQ’s 0.4250 ms, and exhibits superior robustness in the presence of outliers. On the full testing dataset, the dynamic hypernetwork maintains consistent performance, demonstrating its ability to generalize effectively under varying levels of atmospheric turbulence. These results demonstrate that AI-based methods—particularly dynamic hypernetworks—not only address LSQ’s limitations but also enable fast and accurate wavefront fitting, making them promising tools for real-time optical applications.


  • Assessing data imputation for asteroid taxonomic classification

    by Ana V. Ojeda, Carlos A. Guerrero, Juan J. Tapia, Luis H. Sánchez

    Missing data are a prevalent issue in many astrophysics studies and pose a significant challenge when applying machine learning techniques. Specifically, missing data in asteroid databases directly affect the accuracy of taxonomic classification. We propose an imputation method, leveraging a weighted mean of nearest neighbors based on a combined error metric that balances global similarity and local shape preservation to address missing data issues. We evaluated its effectiveness across five machine learning models: Gaussian naive Bayes (GNB), multilayer perceptron, multinomial logistic regression, random forest (RF), and support vector machine (SVM). The models were tested on two datasets: one using only spectral data and another that included albedo as an additional feature. Accuracy (Acc), balanced accuracy, F1 score, and Matthews correlation coefficient were used to evaluate the models in 10-fold cross-validation runs, and Acc was used to evaluate the models in test Acc, with results ranging from 63.10% to 88.10%. The inclusion of albedo significantly improved model performance, particularly for the RF model, which showed a 3.4% increase in test Acc. SVM demonstrated the most robust performance, achieving up to 88.10% test Acc, when albedo was included, whereas GNB performed the worst. These findings demonstrate the potential of the proposed imputation method for achieving high classification accuracy of unclassified asteroids within asteroid databases.


  • A Variant of Parallel-Hybrid Genetic Algorithm for Large-Scale Traveling Salesman Problem

    by Juan J. Tapia, Anabel Martínez-Vargas, M. A. Cosío-León, Leonardo Trujillo

  • Developing a Quantum Genetic Algorithm in MATLAB Using a Quantum Device on AWS

    by Sandra S. Rosales-Alvarado, Oscar Montiel, Ulises Orozco-Rosas, Juan J. Tapia


  • A Variant of Parallel-Hybrid Genetic Algorithm for Large-Scale Traveling Salesman Problem

    by Juan J. Tapia, Anabel Martínez-Vargas, M. A. Cosío-León, Leonardo Trujillo

  • Open Access

    GSGP-CUDA — A CUDA framework for Geometric Semantic Genetic Programming

    by Leonardo Trujillo, Jose Manuel Muñoz Contreras, Daniel E. Hernandez, Mauro Castelli, Juan J. Tapia