Classification of meadow vegetation in the Krkonoše Mts. using aerial hyperspectral data and support vector machines classifier

• Název práce v češtině: Klasifikace lučních porostů v Krkonoších s využitím leteckých hyperspektrálních dat a s pomocí vector machines klasifikace
• Název v anglickém jazyce: Classification of meadow vegetation in the Krkonoše Mts. using aerial hyperspectral data and support vector machines classifier
• Klíčová slova: hyperspektrální data, AISA, Support Vector Machines, Neural Networks, trénovací dataset, horská luční vegetace
• Klíčová slova anglicky: hyperspectral data, AISA, Support Vector Machines, Neural Networks, training dataset, mountainous meadow vegetation
• Akademický rok vypsání: 2013/2014
• Typ práce: diplomová práce
• Jazyk práce: angličtina
• Ústav: Katedra aplikované geoinformatiky a kartografie (31-370)
• Vedoucí / školitel: doc. RNDr. Lucie Kupková, Ph.D.
• Řešitel: skrytý - zadáno vedoucím/školitelem
• Datum přihlášení: 30.01.2014
• Datum zadání: 30.01.2014
• Datum odevzdání elektronické podoby: 30.07.2015
• Datum proběhlé obhajoby: 15.09.2015
• Oponenti: Ing. Markéta Potůčková, Ph.D.
• Konzultanti: Mgr. Stanislav Březina, Ph.D.

Předběžná náplň práce

Meadow vegetation in The Krkonoše Mountains is very complex and heterogeneous environment with plenty of diverse vegetation communities. As this heterogenity makes it difficult to use conventional classifiers successfully, non-conventional classifier - support vector machines was chosen for this task. The thesis will focus on classification setup proposal of meadow vegetation in The Krkonoše Mountains using aerial hyperspectral data achieving as accurate classification as possible given the available data, methods and user requirements.
The first goal of classification will be to experimentally determine which kernel and specific values set for kernel parameters and also value of penalty parameter give the highest classification accuracyfor the area of interest.
Commonly, the result (accuracy) of classification is influenced by selection of training pixels and their sampling design, such selection is of even higher importance when complex environment is being classified. The size of training dataset is on the other side linked to classifier to be used and also to features of dataset to be classified. Unlike conventional classifiers, SVM classifier, as a boundary classifier, uses only pixels on the boundary of classes, because these pixels are the most informative for placing hyperplane which separates the classes. Thus the minimum size of training set is assumed to be smaller for SVM classifier than for conventional classifiers while the accuracy of classification stays high (Foody and Mathur, 2004).
Based on this hypothesis of Foody and Mathur we will experiment with size of training dataset for particular classes and also with distribution of training polygons in the imagery, related to the accuracy of classification. Based on this experimentthe second goal of the thesis will be to propose ideal design for training datasete. g. to propose 1) minimum size of the training set for each class (number of polygons, number of pixels), 2) ideal spatial distribution of training dataset (regular vs. irregular), 3) if for more abundant classes in the imagery more training pixels should be selected and how clustered they should be, 4) if there are some differences between different legend classes. The criterion of tests will be the result of classification accuracy. Results of these experiments could be also used in future as guideline for field mapping, to limit extensive whole-area mapping to only “hot spots” shown as possible sources of support vectors.
An additional goal of the thesis will be classification of the same training dataset using neural network method. This additional goal will be performed mostly in order to compare the result (accuracy) of classification using SVM to a result obtained by another, known, method. Also this experiment will either confirm or reject the hypothesis of Foody and Mathur, that SVM can achieve the same classification accuracy as other classifiers however using smaller training datasets (Foody and Mathur, 2004).

Rozsah grafických prací: dle potřeby

Rozsah průvodní zprávy: 50-60 stran textu, mapové přílohy – výstupy klasifikací

Seznam odborné literatury:

Belousov A.I., Verzakov S.A, von Frese J., 2002, A flexible classification approach with optimal generalisation performance: support vector machines; chemometrics and Intelligent laboratory systems, 64. 15 – 25;
Benediktsson J.A, Swain P.H., Ersoy O.K; 1990, Neural network approaches versus statistical methods in classification of multisource remote sensing data; IEEE Transactions on geoscience and remote sensing, 28/4, 540 – 551;
Borengasser M., Hungate W.S., Watkins R.; Hyperspectral remote sensing: principles and aplications; Taylor&Francis, 2008;
Camps-Valls, G., Gómez-Chova L., Calpe-Maravilla J., 2004, Robust support vector method for hyperspectral data classification and knowledge discovery, IEEE Transactions on geoscience and remote sensing, 20, 1 – 13;
Chan J.C-W., Beckers P., Spanhove T., Borre J.V.; 2012, An evaluation of ensemble classifiers for mapping Natura 2000 heathland in Belgium using spaceborne angular hyperspectral (CHRIS/Proba) imagery; International Journal of Applied Earth Observation and Geoinformation, 18, 13 – 22;
Foody G.M., Mathur, A., 2004; Toward intelligent training of supervised image classifications: directing training data acquisition for SVM classification, Remote sensing of environment, 93, 107 – 117;
Huang, C., Davis, L.S., Townshend, J.R.G, 2002, An assessment of support vector machines for land cover classification; International journal of remote sensing, 23/4; 725 – 749;
Jones H.G. Vaughan R.A.; Remote sensing of vegetation: Principles, techniques and applications; Oxford: Oxford university press, 2010;
Marcinokowska, A., Zagajewski, B., Ochtyra, A., Jarocińska, A., Raczko, E., Kupková, L., Štych, P., Meuleman, K., 2014, Mapping vegetation communities of the Karkonosze National Park using APEX hyperspectral data and Support Vector Machines; Miscellanea Geographica – Regional Studies on Development; 18/2, 23-29;
Melgani F., Bruzzone, L., 2004; Classification of Hyperspectral Remote sensing images with support vector machines; IEEE transactions on geoscience and remote sensing, 42/8; 1778 – 1790;
Mountrakis G., Im J., Ogole C., 2011; Support vector machines in remote sensing: A review; ISPRS Journal of Photogrammetry and Remote Sensing, 66, 247 – 259;
Pal M. and Mather P.M., 2005; Support vector machines for classification in remote sensing, International Journal of Remote Sensing, 26/5, 1007 – 1011;