Aristotle University Of Thessaloniki (AUTH)
Επεξεργασία και ανάλυση χρονοσειρών δορυφορικών εικόνων για την ταξινόμηση χρήσεων/καλύψεων γης και τον εντοπισμό διαχρονικών μεταβολών
Abstract
dc:descriptionCurrently, space missions have reached the point of providing images at different spatial and spectral resolutions and long-term, regular and continuous data acquisition has become their core aspect. At the same time, rapid changes in land-use/land-cover resulting from human activities at global and finer scales, have expanded to a critical degree. It is clear that the development of a methodological framework of mapping and monitoring land-cover changes by integrating remote sensing observations, has become a priority. Although there are significant classification techniques processing individual satellite images or pairs of images, the dynamics of land-cover types demonstrates the need for a different approach. In the field of processing time-series of satellite observations, research is still at an early stage and the development of automated techniques is relatively limited. The main reason can be found in the lack of free, full and open data access to frequent high resolution satellite images. This obstacle has been recently overcome with the missions of Sentinel 2 (2015) and Landsat 8 (2013). Furthermore, the heterogeneity of terrestrial ecosystems and the uneven rate of change constitute a rather complex environment, the systematic monitoring of which requires innovative approaches. In this study, a reliable and precise methodological framework for analyzing and extracting information from high-resolution multitemporal images is developed. The thesis focuses on designing and developing algorithms based on Hidden Markov Models (HMMs), while few studies can be found exploring HMMs’ theory in remote sensing problems. Thus, a methodology is developed that takes into account HMMs’ assets so as to simulate the evolution of classes. In this way it is possible to distinguish and classify them with the utmost accuracy. Research is expanded in investigating the different factors that influence the accuracy of classification. Parameters such as the time frame of the image acquisition, the correction and enhancement of the original images and separability analysis techniques are thoroughly examined and analyzed. In addition, a set of decision fusion functions is proposed according to which the advantages of the classified data are combined. Moreover, a methodology of change detection is presented which involves the exploitation of spectral information and the multitemporal correlation of pixels. The aim of the attempt is to improve the monitoring of artificial or natural phenomena by taking into account the dimension of time. It should be noted that the proposed methodology does not require the manual definition of thresholds to determine the “real” changes and error propagation resulting from individual classifications is avoided. Unlike prior techniques, the original images which are used as inputs in an HMM, are processed simultaneously and the extracted information defines also the type of change. The proposed classification and change detection models is tested over different areas and images from different sensors. Through the evaluation of the results different observations are made considering the characteristics of each study area as well as ways to reduce the volume and extract the “useful” information from the input data.
Degree
thesis:*- Grantor dc:publisher
- Aristotle University Of Thessaloniki (AUTH)
- Year dc:date
- 2016
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Siachalou, Sofia
- Σιάχαλου, Σοφία
Subjects
dc:subject × 14- Τηλεπισκόπηση
- Διαχρονικές εικόνες
- Φαινολογικός κύκλος
- Μοντέλα Hidden Markov
- Remote sensing
- Multitemporal images
- Phenology
- Hidden Markov models
- Επιστήμες Μηχανικού και Τεχνολογία
- Περιβαλλοντική Μηχανική
- Άλλες Επιστήμες Μηχανικού και Τεχνολογίες
- Engineering and Technology
- Environmental Engineering
- Other Engineering and Technologies
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/37518