Mag. Efrén Jiménez Delgado
Labor:
Es estudiante:
No
Programa en que estudia:
Proyectos
Detección automática de características de conservación, autenticidad y contexto histórico de obras y artefactos de patrimonio cultural costarricense mediante técnicas de visión computacional
Número de proyecto:
834-C6-515
De 15/Jun/2026 hasta 15/Jun/2029
Publicaciones
A Comparative Evaluation of ChatGPT, DeepSeek and Gemini in Automatic Unit Test Generation: a Success Rate Analysis
Sep
2025
Descripción:
The advancement of large-scale language models (LLMs) has opened up new possibilities for automating unit test generation, a traditionally manual and expensive task. This quantitative study evaluates the performance of three LLMs-ChatGPT 4o mini, DeepSeek v3, and Gemini 2.5 Flash Pro-in generating test cases for methods in C# developed in Unity. The execution success rate of the generated tests was measured using real and synthetic data. The synthetic data was intentionally created to represent common structures, while the real data came from existing project functions. The experimental design was controlled and included the factors LLM and data type and the blocks cyclomatic complexity and contextual memory with four replicates per combination, for a total of 96 experimental treatments. The results show that LLMs have a high potential to support the automatic generation of unit tests. Furthermore, it was evidenced that the choice of model has a significant effect on the success rate of the generated tests. These findings provide useful initial evidence to guide the selection and use of LLMs in test automation processes within software development environments
Tipo de publicación: Conference Paper
Publicado en: 2025 IEEE VIII Congreso Internacional en Inteligencia Ambiental, Ingenieria de Software y Salud Electronica y Movil (AmITIC)
A Human-Centered Approach for Tattoo Detection Using Convolutional Neural Networks: A Case Study in Forensic Applications
Ene
2026
Descripción:
This paper presents the design, development, and evaluation of a web-based tattoo detection system that integrates Convolutional Neural Networks (CNNs) with a Human-Centered Design (HCD) approach for forensic applications. Manual identification of tattoos in forensic investigations is often slow, error-prone, and subject to human bias, highlighting the need for automated solutions. To address this, we develop a system that combines deep learning with usability-driven design. The methodology involved expert and public surveys, iterative wireframe refinements, and model training using TensorFlow with a fine-tuned ResNet-50 network. Forensic professionals emphasized the importance of accuracy, privacy, and advanced search filters, while general users prioritized usability and transparency. Preliminary evaluations suggest that the system enhances forensic workflows by providing an intuitive interface and automated tattoo identification capabilities. Ethical considerations, such as fairness and bias mitigation, were also integrated into the design. These findings highlight the potential of AI-powered tattoo detection in forensic science, which offers both technical advancements and practical usability improvements.
Tipo de publicación: Book Chapter
Publicado en: Emerging Trends in Information Systems and Technologies
A Comparative Study of Transfer Learning Models for Industrial and Forensic Applications in Automated Tattoo Detection
Sep
2025
Descripción:
Tattoos have been used as biometric identifiers in forensic investigations and industrial security applications. Institutions such as the Costa Rican Judicial Investigation Organization (OIJ) use tattoo recognition as part of their identification and tracking processes. Manual methods for tattoo recognition can be time-consuming and may involve variability in interpretation. In response, this study evaluates multiple transfer learning (TL) approaches for automated tattoo detection. Seven pretrained Convolutional Neural Network (CNN) architectures were examined: MobileNet, DenseNet121, EfficientNetB0, InceptionV3, NASNetMobile, VGG16, and Xception. A dataset comprising 1,000 images was processed using standard data augmentation techniques to improve generalization. Model performance was assessed using established image classification metrics: precision, recall, F1 score, and computational efficiency. The results indicate that MobileNet and ResNet achieved 99.93% accuracy under the experimental conditions given. However, MobileNet required less inference time (3.92 min) compared to ResNet (6.35 min), suggesting potential advantages for applications with time constraints. The results highlight the behavior of the model in terms of classification accuracy and resource requirements. Further evaluation using larger and more diverse datasets is recommended, as well as testing under operational forensic conditions. This study provides a comparative analysis of pretrained CNNs and discusses their applicability in forensic and industrial contexts.
Tipo de publicación: Book Chapter
Publicado en: Intelligent Systems and Applications
A Comparative Study of Transfer Learning Models for Tattoo Detection
Ene
2026
Descripción:
This study explores the use of transfer learning (TL) with computational neural networks (CNNs) to address the challenge of forensic tattoo identification. Tattoos are distinctive markers that play a crucial role in personal identification in forensic investigations. Traditional manual methods for tattoo identification are often time consuming and prone to errors, underscoring the need for automated and reliable alternatives. Pre-trained CNN models, using TL, have shown potential in similar image classification tasks. To evaluate their applicability in tattoo detection, eight pre-trained models were tested: MobileNetV2, Xception, NASNetMobile, InceptionV3, VGG16, DenseNet121, ResNet50 and EfficientNetB0 on a balanced dataset of 1,000 tattoo images, enhanced with data augmentation techniques. The models were trained using a fixed 70%-15%-15% split for training, validation, and testing. The data set was randomly shuffled before splitting to minimize bias in the training process. Although cross-validation is a common approach in machine learning, we opted for a fixed split to maintain consistency in model evaluation. The results indicate that MobileNetV2 achieved the highest accuracy at 99.9%, followed by DenseNet121 (99.7%) and Xception (99.4%). NASNetMobile also performed well, with an accuracy of 99%. InceptionV3 and VGG16 demonstrated moderate precision levels (97.5% and 95.4%, respectively), while ResNet50 and EfficientNetB0 achieved lower precision levels of 83.1% and 70.3%, respectively. Based on these results, MobileNetV2, DenseNet121, and Xception emerged as the most effective models were evaluated based on precision, precision, recall, F1 score and computational efficiency. This evaluation provides a comparative analysis of TL-based CNN models, offering insights into their performance and resource requirements for forensic tattoo identification.
Tipo de publicación: Book Chapter
Publicado en: Emerging Trends in Information Systems and Technologies