Presentation is loading. Please wait.

Presentation is loading. Please wait.

Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Five sampling types with P=8, R=1: (a) X=0, (b) X=1, (c) X=2, (d) X=3, (e) X=4.

Similar presentations


Presentation on theme: "Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Five sampling types with P=8, R=1: (a) X=0, (b) X=1, (c) X=2, (d) X=3, (e) X=4."— Presentation transcript:

1 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Five sampling types with P=8, R=1: (a) X=0, (b) X=1, (c) X=2, (d) X=3, (e) X=4. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

2 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Flow chart of the proposed approach. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

3 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Images considered in the experiments. Leftmost two columns first line: Indian Pines hyperspectral data (three-band color composite and corresponding ground survey). Middle two columns first line: Kennedy Space Center (KSC) hyperspectral data (three-band color composite and corresponding ground survey). Rightmost two columns first line: University of Pavia image (three-band color composite and corresponding ground survey). Leftmost two columns second line: Botswana hyperspectral data (three-band color composite and corresponding ground survey). Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

4 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Data distribution of the four images considered. First row: Indian Pines airborne visible/infrared imaging spectrometer (AVIRIS): (a) mean spectral profiles; (b) example of data manifold in bands 52, 102, and 208. Second row: KSC AVIRIS: (c) mean spectral profiles; (d) example of data manifold in bands 52, 102, and 208. Third row: Pavia ROSIS: (e) mean spectral profiles; (f) example of data manifold in bands 77 (near-infrared) and 55 (red). Bottom row: Botswana AVIRIS: (g) mean spectral profiles; (h) example of data manifold in bands 52, 102, and 208. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

5 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Different class patches, feature histograms by circular/elliptical sampling and Euclidean distance (ED). (a) to (d) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches corn, respectively, and the ED of two histograms is 1.49. (e) to (h) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches soybeans, respectively, and the ED of two histograms is 2.05. (i) to (l) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches woods, respectively, and the ED of two histograms is 0.52. (m) to (p) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches grass-trees, respectively, and the ED of two histograms is 0.53. (q) to (t) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches oak-broadleaf, respectively, and the ED of two histograms is 1.24. (u) to (x) Original sample, rotated sample, two feature histograms obtained before and after sample rotation for class patches water, respectively, and the ED of two histograms is 0.11. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

6 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Classification maps for different features on Indian Pines image. (a) Ground survey data. (b) to (f) Classification maps for RGB, local binary pattern (LBP), gray-level cooccurrence matrix (GLCM), MsLBPriu2 features, and entropy-based query-by-bagging (EQB) sampling, respectively. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

7 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Classification maps for different features on KSC image. (a) Ground survey data. (b) to (f) Classification maps for RGB, LBP, GLCM, MsLBPriu2 features, and EQB sampling, respectively. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

8 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Classification maps for different features on University of Pavia image. (a) Ground survey data. (b) to (f) Classification maps for RGB, LBP, GLCM, MsLBPriu2 features, and EQB sampling, respectively. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

9 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Classification maps for different features on Botswana image. (a) Ground survey data. (b) to (f) Classification maps for RGB, LBP, GLCM, MsLBPriu2 features, and EQB sampling, respectively. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014

10 Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. (Left) Overall accuracy (OA) (%) and (right) kappa statistic as a function of different training rates for different (in markers) features/sampling and different (in colors) classifiers in the considered images (in rows). Features/sampling: RGB (circle), LBP (square), GLCM (diamond), MsLBPriu2 (pentagram), and EQB (cross), respectively. Classifiers: support vector machine (red), K- nearest neighbor (blue), and linear discriminant analysis (black). (a) and (b) OA and kappa statistic for Indian Pines, respectively. (c) and (d) OA and kappa statistic for KSC, respectively. (e) and (f) OA and kappa statistic for University of Pavia, respectively. (g) and (h) OA and kappa statistic for Botswana, respectively. Figure Legend: From: Adaptive classification of hyperspectral images using local consistency J. Electron. Imaging. 2014;23(6):063014. doi:10.1117/1.JEI.23.6.063014


Download ppt "Date of download: 6/23/2016 Copyright © 2016 SPIE. All rights reserved. Five sampling types with P=8, R=1: (a) X=0, (b) X=1, (c) X=2, (d) X=3, (e) X=4."

Similar presentations


Ads by Google