Digital processing of satellite images using neural network algorithms

Abstract

The article discusses the system, the basis of which can be divided into two parts, hardware and software. The hardware part is a single-board computer and software using, which includes the use of neural network algorithms. This system is able to extract the necessary information from the photo and/or video frame specified by the parameters when receiving images from the satellite. This possibility is relevant, because at the moment the trend is that large enterprises want to have at least a few small spacecraft on their account and in this regard they face a number of problems associated with limited mass and size of payload devices. Due to these certain limitations, the processing of satellite frames takes place on Earth, and not on the device itself. Image processing on the spacecraft makes it possible to send only individual fragments of data and not clog the transmission channel, which saves traffic and time. Also, it makes it possible to process data on the spacecraft itself 24/7. This becomes possible thanks to the Python programming language and the trained convolutional neural network. The neural network is connected through the free library TensorFlow. The purpose of this work is to describe the features for solving the problem of classification of objects (pattern recognition).