Entradas recentes
Centuriação e reciclagem das formas cadastrais no território de Bracara Augusta
( 2017 ) Carvalho, Helena Paula Abreu de
A partir da análise da centuriação romana no território próximo da cidade de Bracara Augusta, discute‑se as pistas, as modalidades e os ritmos eventuais da sua reciclagem em contextos de reconfiguração da paisagem fundiária e periurbana, com novos enquadramentos políticos e sociais. Nesse sentido, procede‑se a um alinhamento de várias pistas, tendo em vista a investigação sobre a historicidade longa dos objetos cadastrais.
Using deep learning models for image classification: a case study in water pollution
( 2025 ) Pereira, João Delfim Da Cruz; Novais, Paulo
The increasing availability of satellite imagery from missions like Sentinel-2, combined with advances
in Deep Learning (DL), presents a compelling opportunity to develop scalable, automated, and low-cost
systems for monitoring water quality over large areas.
Water pollution poses a significant threat to ecosystems and public health. Yet, traditional in-situ monitoring
methods are often costly, infrequent, and limited in spatial coverage. Therefore, this dissertation
proposes and validates a DL framework for classifying Lake Water Quality using a multimodal approach
that integrates in-situ data with Sentinel-2 satellite imagery for two Azorean Lakes. The framework systematically
evaluates state-of-the-art architectures, including Convolutional Neural Networks (CNNs) and
Transformers. A robust preprocessing pipeline was developed to handle cloud cover and calculate spectral
indices, while a Water Quality Index (WQI) was engineered from ground-truth measurements to serve
as labels. The study rigorously compares model performance under 3 distinct paradigms: training from
scratch, training with extensive data augmentation, and applying Transfer Learning (TL) with pre-trained
models, all optimised through a systematic hyperparameter search.
The results demonstrate that data augmentation is a critical strategy, consistently yielding the highest
classification performance across all models, with F1 scores frequently exceeding 0.979. While TL
offers a compelling trade-off between performance and computational cost, its success is highly contextdependent,
presenting a ”high-risk, high-reward” scenario. Ultimately, this dissertation confirms the feasibility
of using DL and Remote Sensing (RS) to build a robust, scalable, and automated framework for
environmental monitoring. This solution provides a valuable tool for sustainable water resource management,
transforming raw satellite data into actionable insights to preserve aquatic ecosystems.
Machine learning for alkali-activated concrete: feature attribution, strength–carbon relationships, and the limits of out-of-campaign generalisation
( 2026 ) Pacheco-Torgal, F.; Iqbal, Saqib
Machine learning (ML) models for alkali-activated concrete (AAC) are almost universally evaluated with random train–test splits, yet the literature-compiled datasets are strongly clustered by source study, and the reliability of such evaluations has rarely been quantified. The novelty of this study is a systematic quantification of out-of-campaign generalisation—via Leave-One-Study-Out (LOSO) cross-validation—for ML models trained on the largest curated public AAC dataset (1630 mixtures compiled from 106 published sources), together with model interpretation and an exploratory strength–carbon analysis. Four models (Linear Regression, Random Forest, Gradient Boosting, and optimised extreme gradient boosting, XGBoost) were benchmarked for predicting 28-day compressive strength (CS28). XGBoost performed best under conventional random splitting, with test-set coefficient of determination R2 = 0.801 and root-mean-square error (RMSE) = 7.21 MPa (5-fold cross-validation R2 = 0.758 ± 0.050). Under LOSO validation across 85 study folds, however, the median R2 collapsed to −0.328, with 49 of 85 folds negative: random-split metrics on literature-compiled AAC datasets are substantially inflated by within-study clustering, and study-stratified evaluation should become standard practice in this field. Within these limits, SHapley Additive exPlanations (SHAP) identified ground granulated blast-furnace slag (GGBFS) content, specimen geometry, CaO fraction, curing time, and sodium silicate (Na2SiO3) content as the five most influential predictors; because the oxide descriptors are derived from the declared binder proportions and the carbon-footprint values are inherited estimates from the source dataset, these attributions are associational rather than causal. No practically meaningful overall linear association was observed between estimated CO2 footprint and CS28 (Pearson r = −0.113, 95% CI [−0.175, −0.050], R2 = 0.013), and a Pareto analysis identified 14 candidate low-carbon, high-strength formulations for further experimental and life-cycle assessment. The developed models are suitable for within-dataset feature attribution and exploratory screening restricted to the represented feature domain; they should not be used as external mix-design tools without validation on independent experimental campaigns.