https://repositorio.ufu.br/handle/123456789/18920 Fri, 28 Aug 2026 14:13:23 GMT 2026-08-28T14:13:23Z https://repositorio.ufu.br/handle/123456789/49734 Title: Avaliação temporal da detecção de deficiências nutricionais em folhas de café por deep learning Abstract: The diagnosis of nutritional deficiencies in coffee cultivation, a cornerstone of Brazilian agribusiness, is traditionally reactive and subjective. This work proposes a temporal analysis for the early detection of nutritional stress, with the objective of determining from which point in time symptoms become visually discriminative for deep learning models. To this end, three deep learning architectures were systematically compared, namely ResNet50, a Vision Transformer (ViT-B/16), and a hybrid CNN-ViT model, in the classification of six nutritional deficiencies (N, P, K, Ca, Mg, S) and a control class, using an RGB image dataset collected over five months. The results of 98 experiments demonstrate that the performance of the classifiers is strongly dependent on the symptom stage, with the best observed performance in the configuration that retains the final 40% of the experimental period. Under the standardized protocol adopted, the ResNet50 architecture outperformed the Hybrid CNN-ViT, which in turn outperformed the ViT-B/16, with the convolutional architecture reaching a maximum accuracy of 77.97% with 512 × 512 pixel images, against 71.68% for the Hybrid and 67.74% for the ViT-B/16. An exploratory interpretability analysis via Grad-CAM produced activation maps visually consistent with the expected agronomic symptomatology of each deficiency. Under the controlled experimental conditions, the inductive locality bias of CNNs proved more suitable for capturing foliar texture patterns in this domain, and temporal analysis is crucial for understanding the limits and potential of automated early detection. Fri, 17 Jul 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/49734 2026-07-17T00:00:00Z https://repositorio.ufu.br/handle/123456789/49722 Title: Avaliação de dor em camundongos com redes neurais e visão computacional Abstract: Mice are among the most widely used animal models in biomedical research, making the assessment of animal welfare a fundamental requirement for both the ethical conduct of experiments and the generation of reliable and reproducible scientific results. Therefore, pain assessment remains a significant challenge, as animals cannot directly communicate their sensations, requiring researchers to rely on behavioral and physiological indicators to infer the presence of pain. To reduce the subjectivity inherent to human evaluation, several assessment methods have been proposed over the years, among which the Mouse Grimace Scale has emerged as one of the most widely adopted approaches. This scale evaluates pain through changes in facial expressions associated with painful states in mice. Despite its widespread use, the method still relies heavily on human interpretation, making assessments susceptible to observer bias and inter-rater variability. Motivated by these limitations, this work investigates the application of deep learning and computer vision techniques for automated pain assessment in mice based on the facial action units defined by the Mouse Grimace Scale. To this end, a complete experimental pipeline was developed, encompassing image acquisition, dataset organization, image preprocessing, dataset construction, and supervised training of classification models. Experiments were conducted using well-established convolutional neural network architectures, including variants of the ResNet, MobileNet, and DenseNet families. In addition, the Vision Transformer (ViT) architecture was evaluated to investigate the effectiveness of attention-based mechanisms for modeling global image relationships and to compare its performance with traditional convolutional approaches. Overall, the findings demonstrate the feasibility of applying deep learning techniques to automated pain assessment in laboratory animals, contributing to the reduction of subjectivity in human evaluations and supporting the development of computational tools capable of improving standardization, scalability, and reproducibility in biomedical research. Beyond its contributions to computer vision and artificial intelligence, this work also presents potential benefits for animal welfare and the refinement of experimental protocols. Wed, 22 Jul 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/49722 2026-07-22T00:00:00Z https://repositorio.ufu.br/handle/123456789/49461 Title: Learning strategic poker decision-making with Large Language Models Abstract: Decision-making in the poker variant No-Limit Texas Hold’em involves integrating categorical, numerical, and sequential information in an imperfect-information environment. These elements can be represented as structured text, enabling the investigation of Large Language Models (LLMs) as predictors of strategic decisions. This dissertation examines that possibility in 6-max No-Limit Texas Hold’em, an imperfect-information setting in which each decision involves an action and, in aggressive cases, a bet or raise sizing. The task is formulated as supervised prediction over textualized states, separating its categorical and numerical components. A hybrid pipeline selects actions through log-probability scoring and obtains sizings through deterministic generation followed by numerical parsing, enabling separate analysis of action selection and numerical parameterization. PokerBench examples are evaluated in preflop and postflop settings. Six open LLMs are compared under few-shot prompting, and the best-performing model, Qwen3-14B, is specialized through supervised fine-tuning with LoRA and QLoRA for each game stage. In addition to Action Accuracy, a continuous formulation of Actionand-Sizing Accuracy (Ac-s) is developed to preserve action-first semantics while assigning proportional credit to sizing predictions when the aggressive action is correct. Few-shot prompting provides an informative baseline but remains below supervised adaptation. After fine-tuning, Qwen3-14B achieves 93.3% Action Accuracy preflop and 91.8% postflop, corresponding to gains of 17.2 and 40.3 percentage points, respectively. Analyses show consistent improvements across actions and contexts, while conditional diagnostics indicate high proportional sizing agreement in both stages after the aggressive action is correctly identified. The dissertation therefore contributes an operational formulation, a continuous metric, a reproducible evaluation pipeline, and an empirical analysis of open LLMs for structured strategic decision-making in an imperfect-information domain. Sat, 08 Aug 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/49461 2026-08-08T00:00:00Z https://repositorio.ufu.br/handle/123456789/49244 Title: Detecção de eventos em fóruns da Dark Web e da Surface Web usando algoritmos de aprendizado de máquina Abstract: The detection of malicious posts in Dark Web and Surface Web forums is widely utilized in the field of digital security. Analyzing these posts can aid in the development of better IDSs, the identification of cyber attacks, the discovery of new types of threats, and other benefits. Many forms of attacks can be proactively prevented by searching forum posts for topics related to digital security. To achieve this, it is necessary to identify emerging events in posts with new topics that may contain keywords of interest from a digital security perspective. This work aims to propose an algorithm for detecting novel events and analyzing publications from Dark Web and Surface Web forums over time. Novel events are defined as emerging topics of interest within the cybersecurity domain. To assess the significance of the detected events, a malicious post classifier was employed to categorize posts as having high, medium, or low relevance. The proposed algorithm demonstrated its potential by identifying 26.071 new events from a dataset of 30.584 analyzed posts, among which approximately 4% were classified as highly relevant. Thu, 26 Feb 2026 00:00:00 GMT https://repositorio.ufu.br/handle/123456789/49244 2026-02-26T00:00:00Z