NEURAL NETWORK MODEL OF FOREST CLASSIFICATION BASED ON SATELLITE IMAGERY DATA

Evdokimova Inga Sergeevna, Tulokhonova Inna Stepanovna, Galdanov Gesar Zhambalovich

The article examines neural network models for classifying forests based on satellite images. Various neural network architectures, including convolutional, segmenting and residual networks, are considered, and the most appropriate ResNet-34 architecture is selected. Algorithms for cutting and assembling satellite images have been developed. The neural network model of forest classification was developed using the TensorFlow framework and the error back propagation method. The model was trained on a training sample using NVIDIA RTX 2060 Super GPUs.

Keywords:сlassification, forest, satellite images, loss function, neural network, model, Jacquard coefficient, markup, image slicing algorithm, result evaluation.