This paper presents an algorithm for the detection of deforestation using satellite imagery. The proposed solution is based on the use of paired images, consisting of two images of the same area taken at different times, which are processed by neural network models to obtain a binary segmentation mask corresponding to the deforestation that has occurred. A set of 109 paired images was collected for the experiment. Three models with the architectures ResNet-34+U-Net, SegFormer_b5 and SegNeXt_l are considered as neural network models in this paper. Metrics such as Dice, F-score, presicion and recall were used to evaluate the performance of the models. The SegNeXt_l network performed best in Dice coefficient and recall with values of 0.84 and 0.80, while the ResNet-34+U-Net network performed best in presicion and F-measure with values of 0.71 and 0.74.

Change detection, image segmentation, satellite images, forest cuts, convolutional neural networks, computer vision

Larionov R.V.,  ,  orcid.org/0000-0002-7132-135X,  P.G. Demidov Yaroslavl State University

Sennikov A.V.,  ,  orcid.org/0009-0001-8771-6233,  P.G. Demidov Yaroslavl State University

Khryashchev V.V.,  ,  orcid.org/0000-0003-2978-2459,  P.G. Demidov Yaroslavl State University