INVESTIGATION OF THE IMPACT OF PRIMARY DATA PROCESSING ON THE RESULTS OF NEURAL NETWORK TRAINING FOR SATELLITE IMAGERY RECOGNITION

Authors

  • Dmytro Soldatenko
  • Viktoria Hnatushenko

DOI:

https://doi.org/10.34185/1991-7848.itmm.2022.01.046

Keywords:

neural network, image recognition, data processing, satellite imagery, artificial intelligence

Abstract

Proper preparation of materials for training a neural network is one of the key stages that directly affects all subsequent and results. The research provides two trials, the first with raw data and the second with specially prepared for training. The training of the neural network uses data for three different classes, water, trees and field, test results and information about potential flaws provided after each test as well as recommendations for improving results. The results were analyzed and concluded that the improvement of primary materials for teaching the neural network to recognize satellite images directly affects the result for the better, namely by 27.6% in the case of using prepared data in this study, they may differ depending on the conditions.

References

P.Viola, M.Jones. Rapid object detection using a boosted cascade of simple features

/ 2001 IEEE Computer Society Conference on Computer Vision and Pattern

Recognitions Vol. 1. 8-14 December 2001 / The institute of Electrical and Electronics Engineers, Inc. P. 511-518.

M.Heihhila, M. Pietikainen. C. Schmid. Description of Interest Regions with Center-Symmetric Local Binary Patterns // ICVGIP 2006, P.58-69.

Hnatushenko V., Hnatushenko Vik. Recognition of High Dimensional Multi-Sensor Remote Sensing Data of Various Spatial Resolution. 2020 IEEE Third International Conference on Data Stream Mining & Processing (DSMP), Lviv, Ukraine, 2020, August 21-25, pp. 262-265, doi: 10.1109/DSMP47368.2020.9204186.

Published

2022-05-18

Issue

Section

Статті