Deteção Automática Precoce de Incêndios Florestais Utilizando Redes Neuronais de Aprendizagem Residual (ResNetDetect)

Type: National Project Project

Duration: from 2020 Jan 01 to 2022 Dec 12

Financed by: INESC-ID

Prime Contractor: INESC Inovação (INOV) (Other) - Lisbon, Portugal

The project aims to develop new automatic smoke detection algorithms for the identification of forest fires and its integration in the CYCLOPE system, developed by INOV. To accomplish this task the following tasks are identified: 1) to create a large scale database of forest fire images to allow training with recent deep learning algorithms. This will includes gathering images from the CICLOPE system, from the Internet, by doing bonfire exercises, and by augmenting images with highly advanced generative modeling approaches. 2) Study and compare the efficiency of different deep learning models in automatic smoke pattern detection; 3) Incorporate this new knowledge into CICLOPE operational system and demonstrate the solution in real operating conditions. The key performance indicators will be to improve the state-of-the-art false alarm percentage of 0.6% and fire detection percentage of 96%.


  • INESC Inovação (INOV) (Other) - Lisbon, Portugal
  • INESC-ID (Other)

Principal Investigators