IEEE 2015 MATLAB SPATIAL COHERENCE-BASED BATCH-MODE.pptx

1
SPATIAL COHERENCE-BASED BATCH-MODEACTIVE
LEARNING FOR REMOTE SENSINGIMAGE CLASSIFICATION
2
 ABSTRACT

• Batch-mode active learning (AL) approaches are
  dedicated to the training sample set selection
  for classification, regression, and retrieval
  problems, where a batch of unlabeled samples is
  queried at each iteration by considering both the
  uncertainty and diversity criteria. However, for
  remote sensing applications, the conventional
  methods do not consider the spatial coherence
  between the training samples, which will lead to
  the unnecessary cost. Based on the above two
  points, this paper proposes a spatial
  coherence-based batch-mode AL method. First, mean
  shift clustering is used for the diversity
  criterion, and thus the number of new queries can
  be varied in the different iterations.

3

• Second, the spatial coherence is represented by a
  two-level segmentation map which is used to
  automatically label part of the new queries. To
  get a stable and correct second-level
  segmentation map, a new merging strategy is
  proposed for the mean shift segmentation. The
  experimental results with two real remote sensing
  image data sets confirm the effectiveness of the
  proposed techniques, compared with the other
  state-of-the-art methods.

4
EXISTING SYSTEM

• For hyper spectral image
  classification, the geometric properties of
  different objects may not always be visibly
  distinguishable, and they usually cannot be
  recognized with high reliability by a human
  being. In such cases, a ground survey is
  necessary for the sample labeling. Thus under the
  limitation of labeling costs, the available
  training samples are often not enough for
  adequate learning of the classifier. However, in
  practical applications, manually selecting the
  region of interest in the scene as the training
  set is the common approach this procedure often
  introduces much redundancy into the training
  sample set.

5

• As a result, the labeling costs are increased and
  the corresponding trained classifier performance
  is considerably reduced. In order to reduce the
  labeling costs and optimize the classification
  performance, the training set should be as small
  as possible, to avoid redundancy, and should
  include the samples that are most relevant, to
  improve the performance of the classifier. In
  this context, active learning (AL) approaches
  have been introduced into the classification
  field, which can enrich the information of the
  training sample set and improve the statistic of
  the classes. AL approaches repeatedly ask the
  user to attribute the labels to the most
  informative unlabeled samples, according to a
  function of their class membership uncertainty,
  and then add them to the training set. Through
  this sampling procedure, the classification
  models performance can be iteratively improved.

6
PROPOSED SYSTEM

• The proposed method is based on the batch mode AL
  framework. In the uncertainty procedure, we use
  the two classical criteria to select a batch of
  samples lying close to the decision function. And
  in the diversity procedure, we use mean shift
  clustering to remove the spectral redundancy in
  these samples, and over-segmentation map is used
  to remove the spatial redundancy. Based on the
  refined segmentation map, new selected
  informative samples are tested whether can be
  labeled by the over-segmentation map. If these
  are samples located in a same homogenous patch
  with samples have been assigned with a label,
  these samples are free to be labeled.

7
SOFTWARE REQUIREMENTS

• Mat Lab R 2015a
• Image processing Toolbox 7.1