The article discusses transfer learning methods of convolutional neural network VGG16 for solving the problem of object recognition in images from UAVs (unmanned aerial vehicles). In the absence of the required amount of initial information, it is proposed to work on the augmented dataset. The article presents the architecture of a neural network and considered its action on a specific example. When developing a service, loading the image and displaying the results of the model, was used Flask framework, training of models took place using a cloud service Google Colab based on Jupyter Notebook.

Keywords: deep learning, neural networks, object recognition, data augmentation