Download presentation
Presentation is loading. Please wait.
Published byJayson Johns Modified over 6 years ago
1
Amin Zehtabian, Hassan Ghassemian CONCLUSION & FUTURE WORKS
A FULLY ADAPTIVE OBJECT EXTRACTION TECHNIQUE USED FOR SPECTRAL-SPATIAL CLASSIFICATION OF REMOTELY SENSED DATA Amin Zehtabian, Hassan Ghassemian ย Faculty of Electrical and Computer Engineering, Tarbiat Modares University, Tehran, Iran ABSTRACT Object-based image classification has been frequently addressed in literature, especially for remote sensing applications. Object-based classifiers often benefit from a segmentation step before the classification process in order to generate objects. In this paper, we propose to use the Pixon concept for image segmentation. Meanwhile, in order to form objects which are spectrally homogenous, spatial smoothing is applied as a preprocessing step through using regularized nonlinear partial differential equations (RegAPDE). The parameters of RegAPDE as well as important thresholds used in the Pixon extraction technique are adaptively tuned using four different adaptation algorithms. We also localize the smoothing process via separately applying the RegAPDE algorithm to individual partitions extracted from each layer of the Hyperspectral data using the Watershed transformation. ADAPTATION METHOD (one of the four proposed approaches) ๐ ๐๐ ๐ = ๐ ๐(๐) โ ๐ ๐ (๐) , โฆ ๐ ๐๐.๐ ๐ค ๐ = ๐ ๐๐(๐) โ ๐ ๐ ๐ค ๐ , โฆ ๐ฃ ๐ =๐ฃ๐๐( ๐ ๐ ) The methodology of adaptively finding proper Pixon extraction thresholds (PETs) for each band of data (there are still three other proposed methods to adaptively find the PETs as well as PDE parameters). EXPERIMENTAL RESULTS The gained results for Pavia University and Indian Pines datasets are respectively shown and reported in followings: Pixel-based GTM Object-based METHODOLOGY Fig.1. . Briefed block-diagram of the proposed fully-adaptive object-based classification framework Indian Pines Data (50 training samples per class) ๐๐ผ(๐ฅ,๐ฆ,๐ก) ๐๐ก =๐ป. (๐ ๐ป I ๐ ๐ป๐ผ ๐ฅ,๐ฆ,๐ก ) Fig.2. The merging direction and discipline of the spectral vectors (or pixels) to create Pixons [1] Regularized nonlinear partial differential equations, used for smoothing the data [2] Referring to [1], the proposed Pixon extraction algorithm is categorized as a hierarchical clustering technique in which the pixels are merged together regarding their locations, spectral features and a predefined merging order. In this process, the initial pixel (located in upper-left corner of image) is compared to its neighbours according to a clock-wise order. If the difference between intensities of the two adjacent pixels is lower than a Pixon extraction threshold (PET), the pixels join together. When the first two pixels are merged, their averaged intensity is compared to the intensities of the next pixels and the comparisons continue till all the pixels placed in the neighborhood of the initial pixel are analyzed. By this, the first Pixon is created. This process goes on to extract all the Pixons. CONCLUSION & FUTURE WORKS Applying complex PDEs instead of traditional PDEs, using Genetic Algorithm for optimization as well as proposing an innovative dist-ance metric lead us to better results (as briefly addressed in the appendix) REFERENCES A. Zehtabian and H. Ghassemian, "An adaptive Pixon extraction technique for multispectral/hyperspectral image classification," IEEE Geosci. Remote Sens. Lett., vol. 12, no. 4, pp , Apr.2015. J. Weickert, B.M.T.H. Romeny, M. Viergever, โEfficient and reliable schemes for nonlinear diffusion filtering,โ IEEE Trans. Image Proc., vol. 7, issue 3, pp , 1998. M. Fauvel, Y. Tarabalka, A. Benediktsson, J. Chanussot, and J. C. Tilton, โAdvances in spectral-spatial classification of Hyperspectral images,โ Proc. IEEE, vol. 101, no. 3, pp. 652โ675, Mar M. Khodadadzadeh, J. Li, A. Plaza, H. Ghassemian, J.M. Bioucas-Dias, X. Li, "Spectralโspatial classification of Hyperspectral data using local and global probabilities for mixed pixel characterization," IEEE Trans. on Geosci. and Remote Sens., vol. 52, no. 10, pp , 2014. H. Ghassemian, D.A. Landgrebe, โObject-oriented feature extraction method for image data compactionโ, IEEE Cont. Syst. Mag., vol.8, no.3, pp , 1988.
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.