The paper proposes an improvement of the MobileNetV2 artificial neural network model for detecting and analyzing forest fires in satellite images using attention mechanisms. The main focus is on increasing the accuracy and efficiency of the model by identifying key spatial and channel features. A comparative analysis of the base and improved models in terms of accuracy, loss, computational complexity, and inference time is conducted. The results show that the integration of attention mechanisms improves the quality of fire detection, minimizes false positives, and allows optimizing resources for tasks with limited computing capabilities.

artificial neural networks, attention mechanism, satellite images, classification quality metrics

Mathematical modelling in actual problems of science and technics