| Abstract [eng] |
The goal of Master’s degree is the development and analysis of multispectral images classification system, which would be able to find differences between to images of the same place taken during the time span and then classify them. Firstly, it was researched how multispectral images are acquired and what are the fields of use for them. Then familiarization process with how spectral analysis works and what information you can obtain from different spectrums combination took place. Later, operations that are used for image preparation and processing were examined, together with operations which are doing positive or negative influence for the accuracy of the classification. According to the available materials and data, program was created in the MATLAB program packet. For the training of the program the images of Lithuanian seaside were chosen. At first, the two images taken of the same place during the different times were transformed, for easier comparison, then vegetation was accentuated by using NDVI index. Then both of the images indexes were summed up in order to find differences. Lastly, regions that differ were singled out, classes were assigned and databases created using expert knowledge. Because of the highest accuracy of 96 % KNN classificatory was chosen for the program. In order to evaluate the accuracy of the multispectral images classification, other location was chosen, in this case – Alytus region. The reason of that was that images here were taken during different time, than the images used to create database and teach program. After the classification was made it was found that the program assigned the same values to the pixels as the expert would have done, that is why classification should be considered successful. |
