Loading...

Data Labelling Using Manifold-Based Semi-Supervised Learning in Multispectral Remote Sensing

Khajenezhad, Ahmad | 2013

1019 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 44125 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Rabiee, Hamid Reza; Safari, Mohammad Ali
  7. Abstract:
  8. Classification of hyperspectral remote sensing images is a challenging problem, because of the small number of labeled pixels, high dimensionality of the data and large number of pixels. In this context, semisupervised learning can improve the classification accuracy by extracting information form the distribution of all the labeled and unlabeled data. Among semi-supervised methods, manifold-based algorithms have been frequently used in recent years. In most of the previous works, manifolds are constructed according to spectral representation of data, while spatial dependency of pixel labels is an important property of the images in remote sensing applications. In this thesis, after analyzing the previous works, we propose a manifold-based method for using spatial information beside spectral information. This method uses both spectral and spatial graphs. The spectral graph utilizes spectral similarities between data points and the spatial graph utilizes spatial properties to improve the classification accuracy. The classification is based on optimizing an objective function, including a combination of the Laplacians of the two graphs. As our experimetal results over the famous Indian Pine data set show, by assigning proper values to the parameters of the combination of two Laplacians, this method has an accuracy improvement of about 37% against using the spectral graph alone, and an accuracy improvement of about 6% against using the spatial graph alone. According to the effect of the parameters on the accuracy of the proposed method, it is necessary to use an efficient parameter tuning method. We investigate the performance of two parameter tuning methods and propose a heuristic method, and a second method based on self-training to tune the parameters of the proposed classification method
  9. Keywords:
  10. Remote Sensing ; Semi-Supervised Learning ; Manifold-Based Learning ; Graph-Based Learning ; Hyperspectral Images ; Images Classification

 Digital Object List

 Bookmark

...see more