This article presents a literature review on the identification and mitigation of non-line-of-sight conditions in UWBnetworks, with a focus on edge computing. The difference from previous studies lies in the emphasis on solutionsthat can be executed directly on embedded devices, considering computational resource constraints. The search wasconducted in March 2025 in the IEEE Xplore, ScienceDirect, Scopus, Taylor & Francis, Web of Science, and Wiley OnlineLibrary databases. Initially, 1,050 articles were identified. After removing duplicates, screening abstracts, and readingthe full text, 17 studies aligned with the scope of the review remained. The results indicate the predominance of researchaimed at simplifying features and reduced models. Several studies mention suitability for edge computing, but do notpresent evaluation on restricted hardware devices. The overview obtained highlights the need to move towards solutionsthat close the loop between detection and action, evaluating latency, consumption, and memory directly at the edge.
This study develops an enhanced anomaly detection framework that integrates state-of-the-art deep learning techniques. The system detects hate speech, malicious websites, vulnerabilities in open ports, and suspicious processes. It optimizes detection accuracy and reliability by leveraging tailored datasets, advanced model architectures, and refined preprocessing strategies. To ensure robustness, it employs comprehensive validation metrics, including the Polygon Area Metric (PAM). Empirical results show significant improvements in generalization and prediction accuracy over previous methodologies. The hate speech detection model achieved an accuracy of 88%, with precision of 75%, recall of 62%, F1 score of 62%, and ROC AUC of 92%. These results deliver a proactive monitoring solution suited for modern enterprise security challenges.
Pre-trained contextual language models have demonstrated excellent performance in training with data from new languages and tasks. For this purpose, additional pre-training is necessary, since the lack of vocabulary of the language always tends to degrade the results. In this article, we present a procedure to deal with and treat unknown languages or languages without available resources such as Kicongo, a Bantu matrix language, commonly spoken in the northern regions of Angola, with greater incidence in rural regions and in the countries of central Africa (Democratic Republic of Congo, Republic of Congo and Gabon) and throughout the world. With the extension of Natural Language Processing (NLP) models based on the Bidirectional Encoder Representations from Transformers (BERT) architecture (BERT, RoBERTa and DistilBERT), the performance evaluation of the models was carried out using metrics such as accuracy, precision, recall and f1-score, achieving a performance rate higher than 97%.
A agricultura digital consiste no uso integrado de tecnologias da informação e de técnicas de inteligência artificialpara otimizar a produção agrícola, reduzir custos e promover práticas sustentáveis. Porém, a literatura atual carece deum mapeamento sobre os avanços e resultados de pesquisa publicados nos últimos 5 anos. Este artigo apresenta umarevisão sistemática da literatura sobre sustentabilidade aplicada à agricultura digital, relacionando tecnologias voltadasà produção agrícola e à atenção dada às questões de sustentabilidade. A revisão foi conduzida em bases acadêmicas queapresentaram estudos sobre sustentabilidade aplicadas à agricultura digital. Foram analisados artigos publicados entrejaneiro de 2020 e dezembro de 2025 nas bases acadêmicas ACM, IEEE, Scopus, Springer, Taylor & Francis eWiley. A buscainicial resultou em 13.971 trabalhos. Destes, foram obtidos 40 estudos para a condução da análise. Os resultados obtidosapontam que, a maioria dos artigos selecionados focam na utilização de tecnologias, como sensores e Internet das Coisas,Big Data, análise de dados e inteligência artificial, para o aumento da produtividade. Nota-se que a sustentabilidade éabordada de forma secundária, não sendo o foco principal dos estudos. A análise indica que, dada a atenção necessária, aagricultura digital, com sistemas baseados em sensores, redes IoT e algoritmos de aprendizado de máquina, contribuipara reduzir desperdícios, garantindo sustentabilidade por meio do cuidado coma fauna e flora.
Background: In viticulture, producers sometimes have a decision-making problem in determining the use of land to produce table grapes (for juice production, for instance, the Isabella grape) or wine grapes (for fine wine production). Considering this situation, the aim of this work is to develop and apply a computational code to solve a bi-objective optimization problem in order to minimize irrigation water consumption and maximize the economic benefit from viticulture. To achieve the objective, an interactive algorithm was used based on the solution of multi-objective stochastic linear programming problems. Results: After applying the numerical methodology to solve the optimization problem, the fraction of the production area to be allocated to grapes for juice production and grapes for fine wine production was determined, according to a relaxation parameter involving the different objectives and scenarios of the problem. Conclusions: The results obtained indicate that the choice of the type of grape to be planted depends on different variables, both economic (sales prices and production costs for grape juice and wine) and inherent to the crop itself (yields and crop coefficients). In addition, the code developed is a flexible tool for different decision-making situations.
Visual impairment, ranging from partial limitations to total blindness, remains one of the most critical challenges to digital and social inclusion in Brazil and worldwide. Despite legal advances in the country, such as the Brazilian Inclusion Law, significant barriers still hinder access to digital technologies for visually impaired individuals. This paper presents the development of an accessible laptop featuring a tactile interface and voice commands, aiming to promote autonomy and digital inclusion for visually impaired users. The initiative was carried out through a partnership between two Brazilian companies and applied methodologies such as the decision funnel model, Design Thinking, and a structured product development flow. The project included the construction of a physical prototype with an ABS casing, Braille-labeled keyboard, proximity and location sensors, as well as software with voice command functionality and native OS integration. The proposed solution proved effective in overcoming technological challenges, contributing to a more inclusive digital environment in line with legal frameworks and current social demands.
The classification of rock art symbols, based on the similarity between their various shapes, is essential for the analysis of these ancient symbols. However, the manual classification process is time-consuming, laborious, and prone to errors. Thus, this study investigates the use of the Orange Canvas platform and its feasibility to automate this task, using Machine Learning models. Thus, 3,137 images of rock art symbols were analyzed, divided into three categories: anthropomorphs, circles, and hands. Before training and classification, the images underwent preprocessing involving resizing, cropping, and conversion to compatible formats, in addition to the data augmentation process. For feature extraction, convolutional neural network models, such as Inception V3, SqueezeNet, VGG-16, and VGG-19, were tested on the Orange Canvas platform, and the classification was performed using the algorithms: Neural Network, SVM, and Logistic Regression. The results indicated that the Inception V3 model achieved the best performance, reaching 97\% accuracy, F1 Score, Precision and Recall, with Neural Network. With these metrics, the approach demonstrated to be efficient in the classification of rock art symbols, contributing to the automated analysis of these symbols. It is concluded that the use of the aforementioned approach in the classification of rock art symbols is promising.
This work addresses the development of a classifier for police reports from the city of Marabá-PA, employing data mining techniques and fine-tuning of pre-trained Large Language Models (LLMs), such as Bidirectional Encoder Representations from Transformers (BERT) and its Portuguese-adapted version, BERTimbau. The evaluation of the models indicates that the BERT base and BERTimbau transformers achieved overall accuracies of approximately 90% and 92%, respectively, in experiments conducted with test data. These results demonstrate the feasibility of using LLMs for the automatic classification of police reports, offering potential to enhance criminal data analysis and contribute to more efficient, data-driven public security policies.
Online social networks play a central role in modern communication, serving as platforms for information dissemination, debates, and individual opinion expressions. In this context, stance detection emerges as a promising tool for public opinion analysis, enabling the identification of patterns and trends in controversial discussions. This task can be performed at two levels: statement level or user level. However, the literature has predominantly focused on stance detection at the statement level, leaving gaps regarding the potential of user-level approaches. In light of this, this study presents a systematic literature review aimed at mapping and analyzing the methods and strategies employed in the automatic detection of user stance in online social networks. The findings provide a structured overview of the key practices adopted in the field, as well as the challenges and opportunities for future research.
This work addresses the development of a classifier for police reports from the city of Marab & aacute;-PA, employing data mining techniques and fine-tuning of pre-trained Large Language Models (LLMs), such as Bidirectional Encoder Representations from Transformers (BERT) and its Portuguese-adapted version, BERTimbau. The evaluation of the models indicates that the BERT base and BERTimbau transformers achieved overall accuracies of approximately 90% and 92%, respectively, in experiments conducted with test data. These results demonstrate the feasibility of using LLMs for the automatic classification of police reports, offering potential to enhance criminal data analysis and contribute to more efficient, data-driven public security policies.
Visual impairment, ranging from partial limitations to total blindness, remains one of the most critical challenges to digital and social inclusion in Brazil and worldwide. Despite legal advances in the country, such as the Brazilian Inclusion Law, significant barriers still hinder access to digital technologies for visually impaired individuals. This paper presents the development of an accessible laptop featuring a tactile interface and voice commands, aiming to promote autonomy and digital inclusion for visually impaired users. The initiative was carried out through a partnership between two Brazilian companies and applied methodologies such as the decision funnel model, Design Thinking, and a structured product development flow. The project included the construction of a physical prototype with an ABS casing, Braille-labeled keyboard, proximity and location sensors, as well as software with voice command functionality and native OS integration. The proposed solution proved effective in overcoming technological challenges, contributing to a more inclusive digital environment in line with legal frameworks and current social demands.
Background: In viticulture, producers sometimes have a decision-making problem in determining the use of land to produce table grapes (for juice production, for instance, the Isabella grape) or wine grapes (for fine wine production). Considering this situation, the aim of this work is to develop and apply a computational code to solve a bi-objective optimization problem in order to minimize irrigation water consumption and maximize the economic benefit from viticulture. To achieve the objective, an interactive algorithm was used based on the solution of multi-objective stochastic linear programming problems. Results: After applying the numerical methodology to solve the optimization problem, the fraction of the production area to be allocated to grapes for juice production and grapes for fine wine production was determined, according to a relaxation parameter involving the different objectives and scenarios of the problem. Conclusions: The results obtained indicate that the choice of the type of grape to be planted depends on different variables, both economic (sales prices and production costs for grape juice and wine) and inherent to the crop itself (yields and crop coefficients). In addition, the code developed is a flexible tool for different decision-making situations.
This study aimed to develop an accessible Quantitative Structure-Activity Relationship model based on Machine Learning techniques for a set of analgesic cannabinoid compounds. It represents a cheminformatics contribution to the promising field of endocannabinoid system modulation. Three-dimensional molecular structures and biological activity data were retrieved from PubChem. Molecular descriptors were calculated using Dragons software and subjected to variable selection through two complementary feature selection strategies: Wrapper and Filter methods. The final predictive model was constructed using Support Vector Machines and validated through both K-fold cross-validation and leave-one-out approaches. A robust and predictive model comprising 29 molecular descriptors was obtained, enabling the prediction of novel thiophenyl-acetamide analogs. The combined use of two feature selection techniques proved effective in capturing relevant molecular information and enhancing model performance. This model offers valuable support for the synthetic optimization and design of more potent cannabinoid-based analgesics.
Information systems are seen as strategic tools for managing smart cities. However, the implementation bottlenecks and essential opportunity variables are unknown. The main goal of this paper is to present the challenges and success factors mapped in the literature and the main government information systems in the field of smart cities. The methodological procedures comprises a systematic mapping study of the literature and the application of a survey research with stakeholders interested in the subject in order to evaluate the literature findings from the Brazilian perspective. At the end, 20 government information systems were identified in the literature. The results highlighted geographic expansion as the main challenge, and the main success factor was data availability. According to the participants, in the Brazilian context, the biggest challenge was public/social policies and the main success factor was governance. The potential implications highlight the advancement of research in information systems to support the management of smart cities and the assistance to researchers, public managers and information technology professionals to anticipate risks and opportunities in the implementation of information systems for smart cities.
The classification of rock art symbols, based on the similarity between their various forms, is fundamental for the analysis of these ancient symbols; however, the manual classification process is time-consuming, laborious, and prone to errors. Therefore, this study investigates the use of the Orange Canvas platform and its viability for automating this task using Machine Learning models. Thus, 3,137 images of rock art symbols were analyzed, divided into three categories: anthropomorphic figures, circles, and hands. Before training and classification, the images underwent pre-processing involving resizing, cropping, and conversion to compatible formats, in addition to data augmentation. For feature extraction, convolutional neural network models such as Inception V3, SqueezeNet, VGG-16, and VGG-19 were tested on the Orange Canvas platform, and classification was performed using the Neural Network, SVM, and Logistic Regression algorithms. The results indicated that the Inception V3 model achieved the best performance, reaching 97% Accuracy, F1-Score, Precision, and Recall with a Neural Network. With these metrics, the approach proved to be efficient in classifying rock art symbols, contributing to the automated analysis of these symbols. It is concluded that the aforementioned approach shows promise for classifying rock art symbols.
This paper proposes a lean computational cognitive modeling framework for representing care pathways in primary health care and its operational use. The framework is based on logical decision trees that can model any feed-forward boolean decision process using widely accessible technologies. A user interface and API allow users to define pathways as trees by specifying nodes, rules, and outputs. The framework was experimentally implemented for care pathways from the Brazilian Ministry of Health and from a major Brazilian hospital. Load testing showed responses below 40ms for 1000 simulated users. Integration with an electronic health record displayed pathway recommendations to users in real-time. Analysis of time spent on consultations before and after deployment found a statistically significant 2-minute reduction, suggesting improved efficiency. The proposed framework provides a simple yet effective approach to automating care pathways using only open-source tools, with potential to support primary care delivery and decision-making at scale.
The upkeep of roadside vegetation is essential for ensuring the safety of both motorists and pedestrians. However, identifying the necessity for such maintenance is frequently a time-consuming and costly process, with a high potential for errors in annotation. This work therefore proposes the development of a solution capable of estimating the height of vegetation along roadsides in an automated manner. A machine learning model was developed and evaluated using a dataset of manually annotated data. The model employs a convolutional neural network architecture, adapted for the task of classifying vegetation heights. The results demonstrate that the model is able to detect the different vegetation height classes with low error rates, indicating its potential for automating the decision-making process for mowing vegetation. This study contributes to the advancement of road monitoring techniques, providing greater operational efficiency and reducing costs for road maintenance.
The present work is a systematic literature review (SLR) based on the PICOC framework (Population, Intervention, Comparison, Outcome, and Context), with the objective of mapping and analyzing studies on the application of software engineering methodologies to game development. We analyzed scientific papers up to 2024, retrievable from the ACM, IEEE, and SOL databases using the search string ("game development model"OR "game development models") OR(("software development model"OR "software development models") AND (games OR game)). As a result, 9 articles that met the inclusion criteria were found, and we present the main methodologies applied, their characteristics, applicationcontext, and performance. Overall, the analysis of the selected articles reveals that, although there are significant parallels with traditional software development, game development presents unique challenges and demands thatrequire adapted and specific methodologies. This distinction becomes evident in how software engineering models are applied and modified to meet the specific needs of the gaming industry, which intrinsically combines technical andcreative aspects.
As an optimization problem, the job sequencing and tool switching problem has been the subject of several studies in operations research on its different variations, emphasizing its academic and industrial relevance. Although current methods approaching this problem yield extremely high-quality solutions, the computational time required has proven prohibitive when considering the practical aspects of the problem. Thus, in this paper, a method is presented for generating valid, high-quality solutions in low computational time, which can be used as initial solutions by more robust methods, aiming to accelerate them and contribute to the final quality of the solutions. The proposed approach consists of a new implementation of the random variable-neighborhood descent method using traditional and tailored local searches. Five traveling salesman problem heuristics were considered to generate the initial exploration point for the proposed method. The results obtained were compared with a recent strategy in the literature for generating initial solutions, which demonstrated significant improvement. Additionally, the proposed method was compared to the current state-of-the-art method for the addressed problem, and an average gap of only 5.36% was reported, evidencing the high quality of the solutions achieved for the proposed objective.
A implementação eficiente de redes neurais em hardware, especialmente usando redes-em-chip para interconexão de redes neurais pulsantes em chips neuromórficos, é um avanço significativo em sistemas computacionais que imitam o cérebro humano. Este estudo realiza uma revisão sistemática sobre a integração dessas tecnologias em chips neuromórficos, usando a metodologia PRISMA para analisar diversos estudos e comparar diferentes abordagens em hardware. Os resultados destacam a importância de soluções eficientes em energia e desempenho para inteligência artificial. Apesar dos desafios técnicos, a computação neuromórfica está evoluindo rapidamente, com potencial para impulsionar várias tecnologias emergentes. Os desafios incluem explorar modelos de neurônios, técnicas de aprendizagem e estratégias de interconexão e roteamento para melhorar a eficiência e desempenho.
