توصيفگر ها :
مدل آميخته فرآيند ديريكله , خوشه بندي مدل مبنا , زون هاي دگرساني , سنجش از دور , هايپريون , سيستم¬هاي مس پورفيري , SAM , SVM
چكيده انگليسي :
In this research, the Dirichlet Process Mixture Model (DPMM) clustering algorithm based on the Stick-Breaking process (DPSB) is used for clustering. Although the focus of this research was on hyperspectral and multispectral satellite images for the clustering of porphyry copper deposits, other images were also used to check the performance of this method. Also, to check the effectiveness of the method, its results were compared with several clustering algorithms.
The DPSB algorithm was applied on the synthesized image, AVIRIS, Hyperion, and ASTER data. Then the results obtained from DPSB were compared with the results of K-means clustering, finite Gaussian mixed model (GFMM) clustering, Gaussian model for high dimensional data (GMHD), CLARA clustering, hierarchical clustering (HC), and clustering spectra (SC). In choosing these methods, two criteria have been considered. One is the similarity in their performance with the DPSB method, the GFMM and GMHD based models were used, and secondly, the effectiveness of the methods in high-dimensional data was also considered.
In order to validate the results, three indices, including the normalized information distance (NID), the normalized mutual information index (NMI) and the modified Rand index (MRI) were examined for the method proposed in this thesis. The results obtained on Hyperion images are are 0.821, 0.1799, and 0.2253, respectively. The DP method shows better results than other clustering methods. Among the various data clustering methods used in this study in terms of performance, the SC method has almost similar results to the DP method compared to other clustering methods. K-means ranks third in terms of clustering performance, and the CLARA method is poor than other clustering methods. After field survey of the Kuh-Panj area and analysis of the samples taken from it, the total accuracy and kappa coefficient were 88.6% and 0.85, respectively, which shows more coincidence with filed data compared to other methods. Fieldwork, petrographic study, XRD analysis and ASD field portable spectroradiometer were verified the presence of alteration minerals and zones in the study area. The fusion of DP clustering map and SAM spectral classification map helped to generate an accurate and updated alteration map for the study area.
Next, initially, using ASTER data, different alteration zones of the Zefreh porphyry copper deposit were detected by Band Ratio, Relative Band Depth (RBD), Linear Spectral Unmixing (LSU), Spectral Feature Fitting (SFF), and Orthogonal Subspace Projection (OSP) techniques. These factors were used as input to the DP method. By clustering these, the expansion of alterations was determined and used as training data for use in classification algorithms. The results of DP clustering were consistent with field study and laboratory analysis. By performing the SVM and SAM methods on the ASTER data, areas including phyllic, argillic, propylitic, and iron oxide alterations in the ASTER scene were identified. By field study of these areas, a good agreement was observed between the results obtained from the SVM method and field observations. Alteration regions similar to the SVM results were observed in most of the study areas. By SAM method, most iron oxides and propylitic alterations were detected and in some areas, the observed alterations were less compatible with field studies than SVM method. Field surveys and laboratory studies show that the results obtained from the combination of the unsupervised method and the proposed method have led to better identification of alterations. In this study, the results of the SVM method showed more adaptation to field surveys and laboratory studies than the SAM method. The overall accuracy values for SVM and SAM are 84.4% and 67.2%, respectively. The kappa coefficient calculated for these two methods was 0.74 and 0.52%, respectively.
