Neural network methods for detecting fires in forests

This work includes an analytical review, investigated, supplemented and tested actual neural network methods, algorithms and approaches for solving the problem of early detection of fires in forests using images and video streams from unmanned aerial vehicles. The proposed scheme for solving the problem is based on feature extraction and the use of machine learning for frame classification, selection of a rectangular region with target fire sources and accurate semantic segmentation of fires using convolutional neural networks. The performed modifications of the architectures of neural networks are described, which made it possible to improve the F1-measures achieved by them by 20%.