Abstract

Abstract
There is currently a lack of objective, quantifiable metrics to evaluate children's health and athletic performance during Physical Education classes. To address this gap, the TexP@ct Consortium is developing a Smart Vest for Physical Education (SV4PE)-a textile engineered wearable solution that integrates a single triaxial Inertial Measurement Unit (IMU) and electrocardiogram (ECG) sensors, embedded at the T8 spinal level. Designed for comfortable and unobtrusive use, the SV4PE enables recording and analysis of biomechanical and physiological data during physical activity. This paper presents the preliminary system validation and algorithm development for the SV4PE system, detailing the sensor fusion and signal processing methods used to extract metrics from live and recorded data, along with results from experimental and prototype datasets. The algorithms designed measure an athlete's heart rate, movement intensity, and effort, with additional post-exercise metrics to characterize fundamental movements such as walking, running, and jumping. Sensor fusion packages were implemented, combining acceleration and angular velocity, to correct sensor drifts and remove gravity components. Following filtering and resampling, walking and running metrics, such as cadence, distance and velocity, are extracted through gait event identification, using wavelet transforms. Jumping characteristics are derived from vertical acceleration using projectile motion equations to estimate jump height, take-off force, and power output. Lastly, heart rate is calculated from QRS peak detection in the ECG signal, and associated with subject metadata to evaluate exercise intensity and effort levels. Additional algorithms are under-development to assess fitness tests (e.g., mile run, shuttle run, push-ups, etc.), for team sport motion classification using machine learning, and for player localization within a playfield for detailed performance analysis. Ultimately, this work seeks to provide teachers and trainers with practical tools to objectively monitor and assess children's performance during sports and physical activities.

Abstract
Due to space and energy restrictions, lightweight autonomous underwater vehicles (AUVs) are usually fitted with low-power processing units, which limits the ability to run demanding applications in real time during the mission. However, several robotic perception tasks reveal a parallel nature, where the same processing routine is applied for multiple independent inputs. In such cases, leveraging parallel execution by offloading tasks to a GPU can greatly enhance processing speed. This article presents a collection of generic matrix manipulation kernels, which can be combined to develop parallelized perception applications. Taking advantage of those building blocks, we report a parallel implementation for the 3DupIC algorithm-a probabilistic scan matching method for sonar scan registration. Tests demonstrate the algorithm's real-time performance, enabling 3D sonar scan matching to be executed in real time onboard the EVA AUV.

Abstract
In this paper we demo the Temporal Game, a novel approach to temporal relation extraction that casts the task as an interactive game. Instead of directly annotating interval-level relations, our approach decomposes them into point-wise comparisons between the start and end points of temporal entities. At each step, players classify a single point relation, and the system applies temporal closure to infer additional relations and enforce consistency. This point-based strategy naturally supports both interval and instant entities, enabling more fine-grained and flexible annotation than any previous approach. The Temporal Game also lays the groundwork for training reinforcement learning agents, by treating temporal annotation as a sequential decision-making task. To showcase this potential, the demo presented in this paper includes a Game mode, in which users annotate texts from the TempEval-3 dataset and receive feedback based on a scoring system, and an Annotation mode, that allows custom documents to be annotated and resulting timeline to be exported. Therefore, this demo serves both as a research tool and an annotation interface. The demo is publicly available at https://temporal-game.inesctec.pt, and the source code is open-sourced to foster further research and community-driven development in temporal reasoning and annotation.

Abstract
Breast cancer treatments often affect patients' body image, making aesthetic outcome predictions vital. This study introduces a Deep Learning (DL) multimodal retrieval pipeline using a dataset of 2,193 instances combining clinical attributes and RGB images of patients' upper torsos. We evaluate four retrieval techniques: Weighted Euclidean Distance (WED) with various configurations and shallow Artificial Neural Network (ANN) for tabular data, pre-trained and fine-tuned Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), and a multimodal approach combining both data types. The dataset, categorised into Excellent/Good and Fair/Poor outcomes, is organised into over 20K triplets for training and testing. Results show fine-tuned multimodal ViTs notably enhance performance, achieving up to 73.85% accuracy and 80.62% Adjusted Discounted Cumulative Gain (ADCG). This framework not only aids in managing patient expectations by retrieving the most relevant post-surgical images but also promises broad applications in medical image analysis and retrieval. The main contributions of this paper are the development of a multimodal retrieval system for breast cancer patients based on post-surgery aesthetic outcome and the evaluation of different models on a new dataset annotated by clinicians for image retrieval.

Abstract
Accurate estimation of soil moisture is vital for sustainable water management in agriculture, particularly in olive orchards where precise irrigation strategies are crucial for maintaining productivity and crop quality. Climate change intensifies water scarcity, intensifying the need for advanced methodologies to optimize agricultural water use. Remote sensing technologies, such as Synthetic Aperture Radar (SAR), have emerged as promising tools for monitoring soil moisture over large areas. When combined with in situ measurements and data-driven models like Artificial Neural Networks (ANNs), these technologies offer scalable solutions for addressing the challenges of soil moisture estimation in heterogeneous agricultural landscapes.This study integrates Sentinel-1 SAR data with ANN models to estimate soil moisture in olive orchards located in the Vilariça Valley, northeastern Portugal. Soil moisture measurements were recorded at a depth of 10 cm every 30 minutes from July 2020 to December 2021. Sentinel-1 SAR images were acquired in dual polarizations (VV and VH), and synthetic bands were generated through arithmetic operations combining polarization and calibration metrics (Beta, Sigma, Gamma, Gamma TF), yielding 24 features per image. Two datasets were constructed to evaluate the impact of orbit geometry: (1) D1, containing 161 images from ascending orbits, and (2) D2, comprising 246 images from ascending and descending orbits.The ANN regression model, comprising six hidden layers and K-fold cross-validation (20 splits), demonstrated greater performance with the D1 dataset, achieving a Root Mean Square Error (RMSE) of 2.78, a coefficient of determination (R²) of 0.69, and a Mean Absolute Percentage Error (MAPE) of 8.26%. In contrast, the D2 dataset showed reduced accuracy (RMSE: 3.96, R²: 0.59, MAPE: 12.41%), likely due to variability introduced by combining ascending and descending orbits. These findings underscore the importance of dataset homogeneity in SAR-based soil moisture modeling.This study highlights the potential of integrating Sentinel-1 SAR data with ANN models for soil moisture estimation in olive orchards, contributing to the development of sustainable agricultural practices. Future work should focus on addressing dataset imbalances by expanding the range of observed conditions, incorporating topographic features, and exploring advanced data augmentation techniques to enhance model robustness and scalability. AcknowledgmentsThis work is financed by National Funds through the Portuguese funding agency, FCT - Fundação para a Ciência e a Tecnologia, within project LA/P/0063/2020. DOI 10.54499/LA/P/0063/2020 https://doi.org/10.54499/LA/P/0063/2020 and a doctoral scholarship in a non-academic environment at Fundação Côa Parque (PRT/BD/154871/2023).