ESTiG - Publicações em Proceedings Indexadas à WoS/Scopus
URI permanente para esta coleção:
Navegar
Entradas recentes
- Nonperiodic pathologic voice signals classification using mel-spectrogram and VGGishPublication . Fernandes, Joana; Pinto, João; Moura, Carla; Vilarinho, Helena; Teixeira, Felipe; Freitas, D.; Teixeira, João PauloIn this work and the literature, voice signals can be classified as peri-odic (type 1) or either some periodicity (type 2) and chaos (type 3). This work aims to classify signs into types 1, 2 or 3 to be subsequently applied in a classifi-cation system for pathological/control signs. The original dataset is composed of 466 type 1 individuals, 900 type 2 individuals, and 84 type 3 individuals classi-fied by an otolaryngologist. 15% of the data was used for testing and the remain-ing 85% was used for training and validation. A data augmentation technique was applied to balance the data in training set. Therefore, for the test set, 3380 sounds were used, 1020 type 1, 1280 type 2 and 1080 type 3. Of these, 80% were used for training and 20% for validation. The Mel spectrograms of the signals were used in the input of a VGGish to retrain the model in classifying the 3 types of signals. Regarding test accuracy, this network obtained 71.2%.
- Graphical user interface to acquire ECG using MATLAB and BITalinoPublication . Guerreiro, Nathan; Ribeiro, Luiz; Chaabani, Mohamed Khalil; Luiz, Luiz E.; Dajer, Maria Eugenia; Teixeira, João PauloThis study presents a MATLAB-based graphical user interface (GUI) designed to acquire, process, and visualise electrocardiogram (ECG) signals using the BITalino platform. The interface allows users, including healthcare professionals and researchers, to interactively monitor ECG signals in real-time, providing an accessible and effective tool for cardiac assessment. The system ensures noise reduction and signal clarity using a three-electrode configuration, Butterworth, and notch filters. Results highlight the GUI’s ability to display detailed and comprehensive ECG visualizations, aiding in identifying key ECG characteristics. This work demonstrates the potential for extending the system with machine learning algorithms for automated ECG pattern recognition.
- AI schizophrenia diagnosis through speech features F0 and MFCCPublication . Teixeira, Felipe; Fernandes, Joana; Santos, Adriana; Abreu, J.; Soares, Salviano; Teixeira, João PauloSchizophrenia affects over 20 million people globally and is often undetected in its early stages. Speech has unique characteristics that can help identify mental illnesses, including schizophrenia, which usually manifests through slower, repetitive, or incoherent speech patterns. By extracting acoustic features like fundamental frequency (F0) and Mel Frequency Cepstral Coefficients (MFCCs) and applying machine learning, we can identify patterns that distinguish healthy individuals from those with schizophrenia. In this work, was achieved 95% accuracy to classify between schizophrenic and healthy people through speech.
- Congestive heart failure detection in ECG using LSTM and CNNPublication . Ribeiro, Luiz; Guerreiro, Nathan; Chaabani, Mohamed Khalil; Luiz, Luiz E.; Lazzaretti, André; Teixeira, João PauloCongestive Heart Failure (CHF) is a chronic condition inm which the heart does not pump blood efficiently. This pathology causes fatigue, dyspnea, oedema, nausea, and memory problems, affecting patients’ quality of life. The causes include coronary artery disease, cardiomyopathy, arterial hypertension, and myocarditis. The diagnosis is usually based on the patient’s medical history, physical exams, echocardiogram, electrocardiogram, and other methods. Aiming to improve diagnostic tools, this study proposes an artificial intelligence model based on deep learning to classify pathological ECG signals indicative of CHF. The selected models were LSTM and CNN. The training was conducted using a personalised dataset created from the public databases BIDMC Congestive Heart Failure and PTB Diagnostic ECG from Physionet. ECG data from 28 individuals aged 22 to 71 were selected, including 14 with severe CHF (NYHA class 3 and 4) and 14 control samples without ECG abnormalities. The database architecture was designed so that the input to the neural networks was raw ECG signals without filtering or feature extraction. The results showed an accuracy of 98.21% for the CNN model and 92.26% for the LSTM model.
- Speaker verification on small datasets with ResNet50Publication . Manfron, Enrico; Minetto, Rodrigo; Teixeira, João PauloIn this study, we explore the capabilities of speaker recognition technology for biometric authentication, developing speaker recognitionbased access control systems, and serving as a resource for future research. We focused on developing and evaluating the ResNet50 model for speaker verification. The model was trained and tested on private datasets with 32 speakers and public datasets with 1251 to 6112 speakers. The model ResNet50 achieved a good result on our private dataset by achieving the best EER of 0.7%.
- Binary classification of cardiac pathologies using deep learning: a PTB-XL dataset approachPublication . Chaabani, Mohamed; Guerreiro, Nathan; Ribeiro, Luiz; Luiz, Luiz E.; Slim, Mohamed; Teixeira, João PauloCardiovascular diseases, such as myocardial infarction, are among the leading causes of death worldwide. Accuracy and time are crucial for diagnosing these conditions and for effective treatment, usually requiring time-consuming manual analysis of clinical-grade electrocardiogram (ECG). This paper presents a novel deep learning-based method for binary classification of cardiac patholo-gies using the PTB-XL dataset. The model integrates EfficientNetB3 for spatial feature extraction and a Linformer block to capture long-range dependencies be-tween leads. Preprocessing involves converting RGBA ECG images to RGB for-mat and normalizing them to meet the requirements of the inputs of the layers. Initial experiments have shown promising results, achieving an AUC (Area Un-der the Curve) of 86.06%. Further work includes tests to optimize the model's performance based on different key metrics, including accuracy and precision.
- A comparative analysis of MATLAB and Python neural networks for diabetes predictionPublication . Pimentel, Gabriel; Dessanti, Augusto; Teixeira, João PauloIn recent years, artificial intelligence (AI) has become an integral part of many everyday applications. The growth of AI is bringing intellectual benefits to humans. It is revolutionizing discovery, learning, communication, and work. Machine learning, especially deep learning, is critical to this progress, providing complex models to process data more effectively. Neural networks, which emulate biological processes, find applications across diverse fields, notably in medicine, where they automate disease diagnosis and prediction tasks, such as diabetes. This paper proposes a comparative analysis between Python and MATLAB for diabetes prediction using a dataset with 100,000 individuals. The study conducts simulations on both platforms and validates the results using metrics such as precision, specificity, accuracy and F-measure. Additionally, the study emphasizes the importance of platform selection based on considerations of functionality and cost, offering insights into optimizing outcomes in healthcare applications.
- Cross-border tourism: a residents’ perspective of the iberian meseta reservePublication . Scalabrini, Elaine C.B.; Vaz, Márcia; Teixeira, João Paulo; Rojo, Carlos; Alonso, David; Mestre, Lucia; Fernandes, Paula OdeteTourism destinations depend deeply on the contribution of residents, who play a crucial role as stakeholders. Obtaining insight into their perspectives is crucial for the success of the tourism industry. Hence, this study aimed to understand how residents in the cross-border region of the Iberian Meseta Reserve perceive tourism in their land. From April to May 2023, a questionnaire was administered to residents aged 18 and above, selected from various locations such as shops, parks, restaurants, streets, and residences. A total of 470 valid questionnaires were collected and analysed descriptively, focusing on the means and standard deviations of the effects. An inferential analysis was conducted to determine the differences between the dependence on tourism, gender and the tourism effects. The findings revealed that residents generally hold a positive perspective on tourism in cross-border areas. The economic effects ranked highest in mean perception, followed by the sociocultural effects, and lastly, the environmental effects. However, it's important to acknowledge a limitation of this study, namely the uneven sample size between residents from Portugal and Spain.
- The impact of corporate governance on financial performance: study for portuguese hotel companiesPublication . Fonseca, Clara; Moutinho, Nuno; Alves, JorgeThe present work intends to analyze the relationship between the economic and financial performance of companies and the characteristics of their corporate governance. Based on Portuguese hotel companies this study analyses the importance on the performance of factors such as the size of the board of directors, women on the board, managing shareholder and the largest shareholder. Based on a sample of 3.199 observations relating to companies in the hotel sector, a Tobit regression is used to highlight the impact of characteristics of the company's governance in its performance. As far as is known, this is the first study to link economic and financial performance with the characteristics of Portuguese hotel companies. The results obtained allow us to conclude that the percentage of the largest shareholder is negatively related to the financial performance of hotel companies. In this way, the companies with the largest shareholders are the companies that show a worse financial performance. It should be noted that both the size and gender of the members on the board of directors and the fact that the shareholder is simultaneously a director are not relevant to determine the financial performance of companies in the hotel sector. The results of this study contribute to a greater understanding of how companies are governed and how it affects their performance in a specific sector with a great impact on the national economy, corporate governance affects companies in the hotel sector in Portugal.
- Smart carving of hard-shell fruit with CO2 laserPublication . Farrero, Bernardo; Babo, Pedro; Ribeiro, Luís FrölénThe application of CO₂ laser technology in food processing has gained significant attention due to its precision and adaptability. This study presents an intelligent system for carving hard-shell fruits, specifically acorns, to facilitate their shelling process. The proposed approach integrates real-time control and monitoring technologies to enhance precision and efficiency. A key challenge in acorn processing is the size and shell thickness variability, which complicates mechanical carving. The developed system employs a CO₂ laser to create precise incisions, ensuring optimal shell cracking during dehydration while preventing kernel damage. Experimental tests conducted at the Polytechnic Institute of Bragança identified optimal parameters—6 seconds of laser exposure at 40𝑊 power—for consistent and controlled carving. A thoughtful analysis system was implemented to assess pre- and post-carving conditions, enabling real-time adjustments to laser settings. This self-optimizing process improves the efficiency of the shell carving while reducing waste. The results demonstrate the feasibility of automated acorn carving using CO₂ laser technology, offering a scalable solution for industrial food processing with continuous control of the shell incision. Future research could explore advanced automation techniques to enhance system robustness and adaptability to different fruit types.