Publicações

2024

1. Associação Entre Uso e Ocupação Do Solo e as Temperaturas do Ar: Uma Análise do Período 1990-2020 da Região Geográfica Intermediária de Sousa - Cajazeiras, Estado Da Paraíba

   Teobaldo Gabriel Souza Júnior, Daisy Beserra Lucena, Leandro Honorato Souza Silva, and 2 more authors

   2024

2023

1. Evaluation of Contrastive Learning for Electronic Component Detection

   Leandro H De, S Silva, Agostinho Freire, and 3 more authors

   In ESANN proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning., 2023

   Abs

   The rapid growth of electronic waste (e-waste) has led to an urgent need for efficient recycling processes to recover valuable materials and reduce environmental impact. Waste Printed Circuit Boards (WPCBs) constitute significant e-waste and contain valuable components and precious metals. Computer vision systems can automate the classification , disassembly, and recycling of WPCBs. However, obtaining large annotated datasets for machine learning in this domain is costly and often unavailable. This paper investigates using few-shot and supervised contrastive learning in electronic component detection. We propose a model incorporating contrastive learning components for detecting electronic components in scenarios with limited training data or annotated labels. Our experimental results show that, in limited-data scenarios, con-trastive learning outperforms the original versions of Faster R-CNN object detector. This study contributes to developing efficient recycling solutions for e-waste management and resource recovery.

2. Enhancing Electronic Component Classification with Supervised Contrastive Learning: A Comparative Evaluation

   Leandro H.de S. Silva, Agostinho Freire, George O.A. Azevedo, and 2 more authors

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   Electronic component classification is crucial in manufacturing, testing, and recycling. This paper evaluates the effectiveness of supervised contrastive learning in enhancing electronic component classification. Traditional and deep learning network-based methods are commonly employed for this task, with convolutional neural networks (CNNs) showing promising results. The paper evaluates supervised contrastive learning that encourages similar instances to have closer representations in the feature space while pushing different instances apart. The results indicate that supervised contrastive learning does not significantly improve accuracy with abundant data but outperforms the standard CNN approach in few-shot scenarios, achieving an accuracy of 81% compared to 33% in the traditional approach. Supervised contrastive learning has the potential to enhance electronic component classification tasks with limited labeled datasets, which are often costly and difficult to obtain.

3. Assessing the Effect of Urban Expansion and Deforestation on Temperature Rise in Cajazeiras, Brazil: A Data-Driven Approach

   João Vinicius Ribeiro Andrade, Teobaldo Gabriel Souza, Leandro H.de S. Silva, and 2 more authors

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   Climate change is a task of profound academic and societal relevance that promises to shape our actions for the planet’s sustainable future. Understanding the complex relationship between land transformations and temperature changes is crucial for effective decision-making in the face of climate change. This study uses a data-driven approach to assess the effect of urban expansion and deforestation on temperature rise in Cajazeiras, Brazil. By integrating statistical modeling and machine learning techniques, we investigate the influence of urban areas and other factors on temperature trends. The statistical model (ARIMA) reveals a strong association between urbanization and temperature increase, emphasizing the need to address urban development and land use changes to mitigate rising temperatures. The SHAP values technique applied to a Random Forest model also uncovers additional drivers of temperature rise beyond urban expansion, such as the absence of savanna formation near the city. This comprehensive analysis provides valuable insights for policymakers and urban planners in mitigating the impact of climate change.

4. Fine-Tuning MultiFit for Enhanced Legal Sentence Basis Classification

   David J.R. Barrientos, Bruno J.T. Fernandes, Cleyton M.de O. Rodrigues, and 3 more authors

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   Deep learning algorithms have shown promise in effectively classifying legal texts, surpassing traditional methods. However, existing approaches are primarily designed for English text and lack suitability for other languages, mainly Portuguese. This study addresses the challenge of classifying legal basis in first-degree sentences within Brazilian law by fine-tuning the multilingual MultiFit model using a novel basis dataset, comprehensively training the model for accurate legal basis classification. The bidirectional deep-learning MultiFit model has been subjected to rigorous fine-tuning, resulting in exceptional performance while maintaining consistently high quality. Results obtained highlight the model’s remarkable proficiency in precisely categorizing legal bases in first-degree sentences, achieving an accuracy rate of 80.0%, precision of 83.3%, recall of 80.7%, and an F1 score of 82.0%. These results demonstrate the model’s adaptability, versatility, and suitability for legal applications. In addition, it exhibits high precision, accuracy, and efficiency in classifying legal bases in first-degree sentences. Moreover, successfully fine-tuning pre-trained models for new tasks, leveraging extensive datasets, highlights their significant potential in enhancing performance in legal applications.

5. Interpretable Diagnosis of Banana Leaf Sigatoka Disease using CNNs and Shapley Values

   Agostinho Freire, Rafaella Lima Roque, Leandro H.de S. Silva, and 1 more author

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   Sigatoka disease significantly threatens banana crops, causing yield losses and quality deterioration. Convolutional Neural Networks (CNNs) have shown promise in diagnosing plant diseases. However, their lack of interpretability limits their practical application because they do not answer the disease stage and quantification questions. The study aims to identify the presence of Sigatoka disease in banana leaves and estimate the severity beyond binary classification. To achieve this goal, we propose a methodology that combines CNNs and Shapley values to develop an interpretable diagnosis system for Sigatoka disease in banana leaves. The results demonstrate the effectiveness of the proposed methodology, with the SHAP VGG 16 model outperforming other models in accurately identifying the affected surface beside the disease classification. Furthermore, this CNN model, particularly with good image focus and framing, show promise in detecting and evaluating the disease. Nevertheless, challenges remain in distinguishing pixels in shadowed regions from infected areas, requiring further refinement and verification by phytopathologists.

6. Legal Judgement Prediction: A comparative analysis of legal documents in Brazil

   Cristian Millan-Arias, Marie Chantelle C. Medina, Jean Felipe C. Ferreira, and 8 more authors

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   Legal Judgement Prediction has become an essential task in several domains of law. It plays a significant role when studies are performed for a given case to assist lawyers and experts. Machine learning approaches have been extensively used among the methods employed to perform this task due to their flexibility and accuracy. However, machine learning methods alone can not be used directly in the raw data, which raises the need for preprocessing and feature engineering methodologies. Several features can influence the final decision on a case. Thus, a comparative analysis of prediction methods and preprocessing strategies was conducted on legal datasets in the Portuguese language. The results showed that ensemble methods achieved the best results in predicting legal cases, achieving an accuracy between 65% and 85% in the data.

7. Evaluation of Deep Learning Model for Detecting Electronic Components in Few-Shot Learning Scenarios

   Allana Rocha, Bruno Fernandes, and Leandro Honorato

   In 2023 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2023, 2023

   Abs

   The increasing waste of electronic equipment causes adverse environmental impacts and emphasizes the importance of component reuse for a sustainable circular economy. This study uses few-shot learning to explore variations of machine learning models, specifically deep learning, to detect toroidal inductors on circuit boards. A customized database with 50 labeled images of toroidal inductors was developed and made available online due to the limitation of available data. Three detection models were trained and evaluated using different amounts of training images (10, 20, and 30 images) and proportional epochs. The results show that models trained with 20 and 30 images achieved satisfactory performance, with an accuracy of 96.8% and 97.1%, and mAP50 of 98.2% and 99%, respectively. However, the model trained with only ten training images showed inferior results, with an accuracy of 6.2% and recall of 28.6%. The results indicate that using 20 to 30 images in training can be an efficient strategy for detecting components on circuit boards and that the number of training images correlates differently with the model’s performance. Additionally, image quality, proper hyperparameter selection, and choice of detection algorithm are crucial factors in obtaining reliable results in different detection tasks. This study contributes to developing more sustainable and responsible solutions for the reuse of electronic components while emphasizing the importance of considering multiple factors when training detection models to ensure efficient and reliable results.

2022

1. Event-Based Angular Speed Measurement and Movement Monitoring

   George Oliveira Araújo Azevedo, Bruno José Torres Fernandes, Leandro Honorato Souza Silva, and 3 more authors

   Sensors, Oct 2022

   Abs

   Computer vision techniques can monitor the rotational speed of rotating equipment or machines to understand their working conditions and prevent failures. Such techniques are highly precise, contactless, and potentially suitable for applications without massive setup changes. However, traditional vision sensors collect a significant amount of data to process and measure the rotation of high-speed systems, and they are susceptible to motion blur. This work proposes a new method for measuring rotational speed processing event-based data applied to high-speed systems using a neuromorphic sensor. This sensor produces event-based data and is designed to work with high temporal resolution and high dynamic range. The main advantages of the Event-based Angular Speed Measurement (EB-ASM) method are the high dynamic range, the absence of motion blurring, and the possibility of measuring multiple rotations simultaneously with a single device. The proposed method uses the time difference between spikes in a Kernel or Window selected in the sensor frame range. It is evaluated in two experimental scenarios by measuring a fan rotational speed and a Router Computer Numerical Control (CNC) spindle. The results compare measurements with a calibrated digital photo-tachometer. Based on the performed tests, the EB-ASM can measure the rotational speed with a mean absolute error of less than 0.2% for both scenarios.

2. Learning from pseudo-labels: Self-training Electronic Components Detector for Waste Printed Circuit Boards

   Agostinho A.F. Junior, Leandro H.De S. Silva, Bruno J.T. Fernandes, and 2 more authors

   In Proceedings - 2022 35th Conference on Graphics, Patterns, and Images, SIBGRAPI 2022, Oct 2022

   Abs

   Electronic components (EC) detection is relevant for Printed Circuit Board (PCB) manufacturing, quality inspection, and assurance (cyber-security). Moreover, at end-of-life electronic devices, the PCBs become Waste PCBs (WPCBs), the primary source of high-value materials in electronic waste. However, the recycling process is challenging because of the WPCB’s high composition diversity. EC detection in WPCBs can reduce the uncertainty about WPCBs composition and help select a better recycling process. Nevertheless, large fully-annotated datasets for the PCB domain are available at a high cost. For WPCBs, there is only one publicly available dataset, the PCB-DSLR, and it is partially labeled. As for PCBs, the FICS-PCB dataset is available. However, it contains images of modern and clean PCBs, all in good condition, while the first has broken and dirty parts. Thus, we propose a self-training strategy for object detection algorithms without needing a fully labeled dataset. In the strategy, the YOLOv5 model is used to detect the eight types of electronic components present in FICS-PCB in PCB-DSLR, which initially has labels only for integrated circuits. The process occurs through the interaction between a teacher model, trained with the FICSPCB data, and a student model, which is trained by joining the FICS-PCB data and the teacher-generated labels for PCB-DSLR. The results show that the student model performs better in the electronic components detection task, obtaining greater precision and sensitivity when compared to the teacher model. Since we are dealing with partially labeled data, we provide a low-dimensional representation of the detections of both models using t-SNE.

3. Classification of Legal Documents in Portuguese Language Based on Summarization

   Marie Chantelle Cruz Medina, Lucas Matheus Da Silva Oliveira, Jean Felipe Coelho Ferreira, and 7 more authors

   In Proceedings - 2022 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2022, Oct 2022

   Abs

   Legal document classification in Portuguese language is a research area highly benefited by computational intelligence techniques as the availability of better processing with the easiness of digital text recording of juridic processes. Different techniques have been explored to achieve reliable results in real-world conditions; however, the most suitable configuration of methods remains to be an open problem. This study proposes a model consisting of four stages: preprocessing, extractive summarization using page rank algorithm, feature extraction with bag-of-words, and classification with Support Vector Classifier. Testing sessions were conducted using three versions of the model as a mean for comparison and evaluation. The first one was a basic classifier without preprocessing nor summarization stages, the second included preprocessing but not summarization, and the third one was an implementation of the complete proposed model. All three were evaluated using a separated set of examples falling into six different labeled categories and their performance was recorded calculating weighted average precision, recall, F1-score and accuracy values. The best performance obtained was the one presented by the proposed model with precision, recall and F-1 score values of 96% each, which represents a 2% improvement for all of them in comparison to the first version and a 1 % improvement for precision and recall in comparison to the second version. Specially F1-score pointing to the most balanced performance, the proposed model outperformed the versions of it itself excluding some stages, allowing to infer that preprocessing and extractive summarization have positive impacts in the text classification task for Portuguese-written legal documents.

4. A Legal Information System for Intelligent Sentence Mining Applied to Civil Law

   David Barrientos, Bruno J.T. Fernandes, Cleyton Mario O. Rodrigues, and 7 more authors

   In Proceedings - 2022 IEEE Latin American Conference on Computational Intelligence, LA-CCI 2022, Oct 2022

   Abs

   Manual search and evaluation of relevant information in legal documents usually are slow, repetitive, and liable to error. On the other hand, intelligent tools for text mining have been successfully employed in different areas such as Law. The interpretation and extraction of information from lawsuits are non-trivial. Even if more experienced lawyers, when analyzing each case individually, can indicate the most likely results and the best-associated strategies, there is no clear methodology for achieving this analysis. We describe in this study the Intelligent Sentence Mining (MIS) tool, an information system for prospecting and automatic extraction of information in Brazilian civil sentences, such as decision, name of the judge, sentence value, and obligations to do. The proposed system was conceived based on a hybrid architecture addressing ontologies and machine learning strategies. Regarding data prospecting from civil proceedings, it was observed from the analysis presenting an accuracy between 86% and 96% (in general) when extracting the judges name, values, and decision from the sentence.

2021

1. Estimating Recycling Return of Integrated Circuits Using Computer Vision on Printed Circuit Boards

   Leandro H. S. Silva, Agostinho A. F. Júnior, George O. A. Azevedo, and 2 more authors

   Applied Sciences, Mar 2021

   Abs

   The technological growth of the last decades has brought many improvements in daily life, but also concerns on how to deal with electronic waste. Electrical and electronic equipment waste is the fastest-growing rate in the industrialized world. One of the elements of electronic equipment is the printed circuit board (PCB) and almost every electronic equipment has a PCB inside it. While waste PCB (WPCB) recycling may result in the recovery of potentially precious materials and the reuse of some components, it is a challenging task because its composition diversity requires a cautious pre-processing stage to achieve optimal recycling outcomes. Our research focused on proposing a method to evaluate the economic feasibility of recycling integrated circuits (ICs) from WPCB. The proposed method can help decide whether to dismantle a separate WPCB before the physical or mechanical recycling process and consists of estimating the IC area from a WPCB, calculating the IC’s weight using surface density, and estimating how much metal can be recovered by recycling those ICs. To estimate the IC area in a WPCB, we used a state-of-the-art object detection deep learning model (YOLO) and the PCB DSLR image dataset to detect the WPCB’s ICs. Regarding IC detection, the best result was obtained with the partitioned analysis of each image through a sliding window, thus creating new images of smaller dimensions, reaching 86.77% mAP. As a final result, we estimate that the Deep PCB Dataset has a total of 1079.18 g of ICs, from which it would be possible to recover at least 909.94 g of metals and silicon elements from all WPCBs’ ICs. Since there is a high variability in the compositions of WPCBs, it is possible to calculate the gross income for each WPCB and use it as a decision criterion for the type of pre-processing.

2. Covid-19 diagnosis by combining RT-PCR and pseudo-convolutional machines to characterize virus sequences

   Juliana Carneiro Gomes, Aras Ismael Masood, Leandro Honorato S. Silva, and 7 more authors

   Scientific Reports, Mar 2021

   Abs

   The Covid-19 pandemic, a disease transmitted by the SARS-CoV-2 virus, has already caused the infection of more than 120 million people, of which 70 million have been recovered, while 3 million people have died. The high speed of infection has led to the rapid depletion of public health resources in most countries. RT-PCR is Covid-19’s reference diagnostic method. In this work we propose a new technique for representing DNA sequences: they are divided into smaller sequences with overlap in a pseudo-convolutional approach and represented by co-occurrence matrices. This technique eliminates multiple sequence alignment. Through the proposed method, it is possible to identify virus sequences from a large database: 347,363 virus DNA sequences from 24 virus families and SARS-CoV-2. When comparing SARS-CoV-2 with virus families with similar symptoms, we obtained 0.97 ± 0.03 for sensitivity and 0.9919 ± 0.0005 for specificity with MLP classifier and 30% overlap. When SARS-CoV-2 is compared to other coronaviruses and healthy human DNA sequences, we obtained 0.99 ± 0.01 for sensitivity and 0.9986 ± 0.0002 for specificity with MLP and 50% overlap. Therefore, the molecular diagnosis of Covid-19 can be optimized by combining RT-PCR and our pseudo-convolutional method to identify DNA sequences for SARS-CoV-2 with greater specificity and sensitivity.

3. Improved Smart Eye Tracker Communicator Using Low Resolution Webcam

   Allana L. S. Rocha, Leandro H. S. Silva, and Bruno J. T. Fernandes

   In Proceedings - 2021 35th Conference on Graphics, Patterns, and Images, SIBGRAPI 2022, Jun 2021

   Abs

   Eye tracking is a tool presented in many applications ranging from scientific research to commercial applications. One of them is assistive technologies that aim to help people with some disabilities, including communication. However, the applications usually require specific hardware components or a high computational cost. This work proposes the Smart Eye Communicator II (SEC-II), an evolution of a previously presented algorithm to detect the pupil center and the user's gaze direction in real-time, using a low-resolution webcam and a conventional computer without a need for calibration. In SEC-II, a face aligner, which gets a canonical face alignment based on translation, scale, and rotation, has been added to the system. Likewise, strategies using eye coordinates were implemented to find the dominant the algorithm eye. By implementing these new approaches, achieved 86% accuracy, even under variable and non-uniform environmental conditions. Moreover, a graphical interface was implemented connecting the SEC-II to the internet and allowing users to express their desires and watch online videos chosen by themselves.

4. MISLA²: A System to Information Retrieval in Labour Lawsuits using Legal Ontologies and Regular Expressions

   Cleyton M. O. Rodrigues, Bruno J. T. Fernandes, Leandro H. S. Silva, and 7 more authors

   In Anais do XVIII Encontro Nacional de Inteligência Artificial e Computacional (ENIAC 2021), Nov 2021

   Abs

   Electronic Legal Proceedings are a worldwide legal phenomena, allowing the use of computerized systems for the creation and monitoring of procedural acts in the most diverse legal bodies. On one hand, it allows greater transparency in the conduct of procedural acts, on the other, it has contributed to the bottleneck of open but unresolved lawsuits each year. Nowadays, Information Retrieval to automate the processing of these procedural objects is at the forefront of computer systems for Law. In this study, we present MISLA2, a system to retrieve orders and preliminaries from judicial labour sentences through ontological models built from previous cases. Instead of tied and difficult-to-maintain domain specification models, we demonstrate how light ontologies, in conjunction with regular expressions for extracting significant portions of the text, can achieve the desired results. In addition, empirical experiments carried out with real labour lawsuits evidence that results are quite promising.

2020

1. Binary and Multi-label Defect Classification of Printed Circuit Boards based on Transfer Learning

   Leandro H De, S Silva, Agostinho A F Júnior, and 3 more authors

   In ESANN 2020 proceedings, European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning., Nov 2020

   Abs

   Automatic optical inspection for printed circuit boards (PCB) is an important step to assure quality control in electronic manufacturing. Recently deep learning models have been used to detect and classify PCB defects. Since public PCB datasets usually are not large enough to train deep models from scratch, transfer learning has proved to be an effective strategy to overcome this limitation. In this paper we evaluate the influence of input image size for non-referential binary classification of PCB images from the DeepPCB dataset and moving further we evaluated a multi-label classification, both based on transfer learning. The best models achieved 99.5% accuracy for binary classification and mean accuracy of 95.16% for multi-label classification.