COMPARISON OF INDIVIDUAL AND COMBINED CLASSIFICATION METHODS IN ASTER DATA – A CASE STUDY FROM LAHRUD – NW IRAN |
Paper ID : 1074-SMPR-FULL |
Authors: |
مجید محمدی اسکویی1, مهیار حقیقی *2 1استادیار دانشکده معدن، دانشگاه صنعتی سهند تبریز 2کرمانشاه-سنقرکلیایی-بلوار جانبازان -پایینتر از اداره آب-6751894141 |
Abstract: |
Detection and classification of earth features is the principle aim of remote sensing studies. Classification of remote sensing data is broadly used for mapping of different features on the earth, and mapping of alterations is a geological terrain can be considered as a classification task in remote sensing data processing. The prior information from the study scene plays a vital role in achieving accurate classification known as supervised methods. This information is usually collected from previous studies and makes up training dataset for classification. Training dataset is an important part of a classification process. The study aims to classify an ASTER scene with the use of Hyperion unmixing results of Lahrud, Iran. By using this method a couple of high accuracy Hyperion data for detection of training set and high coverage of Aster data in classification stage will be available. The detected minerals by Mixture Tuned Matched Filtering (MTMF) method on Hyperion image were used as training classes to classify the ASTER data. The integrated use of the ASTER and Hyperion datasets improves the accuracy of the classification. On the other hand, the discussed procedure is fast and cost effective. The capability of the ASTER data to spectrally discriminate between training classes is a challenging issue. The separability score was therefore computed between classes imported from Hyperion data analysis. This was satisfactory when a certain number (500) of pixels with highest probability were selected for each training class. In order to improve the accuracy of upcoming processes, classes with high similarity (low separability) were combined. The six classes with high separability are selected among ten minerals were remained as training set to the classification stage. In the next step, the ASTER scene was classified using six individual (maximum likelihood, minimum distance, mahalonobis distance, spectral angle mapper, support vector machine, neural network) classifiers. In order to obtain better results and improve the final maps, four combined classification methods (product rule, sum rule, max rule, median rule) are examined. The accuracy of the results of each classifier was evaluated by constructing the related confusion matrix. The matrix was computed based on all pixels of the respected training class. The max rule method was yielded most accurate results among the all individual and combined classifiers and the maximum likelihood showed a good performance among individual classification methods regarding to its highest overall accuracy. |
Keywords: |
Remote sensing, Hyperion, Aster, Individual classification, Combined classification, Training set, MTMF, Overall accuracy |
Status : Paper Accepted (Poster Presentation) |