Aerial Fire Image Synthesis and Detection

• Type: Article in International Conference Proceedings Book
• Date: 2022

Authors

Sandro Campos (Author) Other; Daniel Castro Silva (Author) FEUP

Conference proceedings International

Title: Proceedings of the 14th International Conference on Agents and Artificial Intelligence, ICAART 2022, Volume 3, Online Streaming, February 3-5, 2022

Pages: 273-284

Indexing

ISI Web of Knowledge - 0 Citations

Other information

Authenticus ID: P-00W-9AV

DOI: 10.5220/0010829400003116

Abstract (EN): Unmanned Aerial Vehicles appear as efficient platforms for fire detection and monitoring due to their low cost and flexibility features. Detecting flames and smoke from above is performed visually or by employing onboard temperature and gas concentration sensors. However, approaches based on computer vision and machine learning techniques have identified a pertinent problem of class imbalance in the fire image domain, which hinders detection performance. To represent fires visually and in an automated fashion, a residual neural network generator based on CycleGAN is implemented to perform unpaired image-to-image translation of non-fire images obtained from Bing Maps to the fire domain. Additionally, the adaptation of ERNet, a lightweight disaster classification network trained on the real fire domain, enables simulated aircraft to carry out fire detection along their trajectories. We do so under an environment comprised of a multi-agent distributed platform for aircraft and environmental disturbances, which helps tackle the previous inconvenience by accelerating artificial aerial fire imagery acquisition. The generator was tested using the metric of Frechet Inception Distance, and qualitatively, resorting to the opinion of 122 subjects. The images were considered diverse and of good quality, particularly for the forest and urban scenarios, and their anomalies were highlighted to identify further improvements. The detector performance was evaluated in interaction with the simulation platform. It was proven to be compatible with real-time requirements, processing detection requests at around 100 ms, reaching an accuracy of 90.2% and a false positive rate of 4.5%.

Language: English

Type (Professor's evaluation): Scientific

No. of pages: 12