2026-06-112026-06-112025-09-15MELO, Ítalo José da Silva. Rede neural convolucional EfficientNet: uma proposta de classificação de fogo usando tranferência de aprendizado. Orientador: José Jaílton Henrique Ferreira Junior. 2025. 14 f. Trabalho de Curso (Bacharelado em Engenharia de Computação) – Faculdade de Engenharia da Computação, Campus Universitário de Castanhal, Universidade Federal do Pará, Castanhal, 2025. Disponível em: https://bdm.ufpa.br/handle/prefix/9597. Acesso em:.https://bdm.ufpa.br/handle/prefix/9597The alarming increase in wildfires in Brazil in 2024 reinforces the urgency of real-time automatic fire detection systems. This work proposes a binary classifier based on the EfficientNet architecture, pre-trained on ImageNet and fine-tuned using transfer learning and data augmentation techniques. The model was trained and evaluated on a public dataset, achieving 98% accuracy and 100% recall for the fire class, ensuring that no fire cases were missed. The results confirm the robustness and efficiency of the solution, which is suitable for integration into embedded devices and early warning applications. Future work includes expanding datasets, reducing false positives, and conducting field tests for practical validation.Acesso AbertoRede neuralTransferência de aprendizadoFine-tuningCNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAOTrabalho de Curso - Graduação - Artigo