Himanshu Govil

1

Comparative evaluation of fuzzy based
object-oriented image classification method with
parametric and non-parametric classifiers
Himanshu Govil A.M.U.Aligarh
2
Objectives

• Up to what level of classification can we perform
  on LISSIII/LISSIV data?
• Is any advantage of high spectral resolution of
  LISSIII over LISSIV. If yes than how can we use
  it for classification ?
• Would object based classification method work on
  LISS III/LISSIV. If yes than what would be the
  level of accuracy?
• Would knowledge based classification give the
  appropriate result for low and medium resolution
  images?
• Could we increase the accuracy of these
  classification methods?

3

• Maximum Likelihood (ML)
• (Parametric Classifier)
• Object Based (OB)
• (Fuzzy classifier)
• Knowledge Based (KB)
• (Non-parametric Classifier)

4
DATA AND STUDY AREA

• Satellite images of the area
• IRS-P6 LISS IV
• IRS-P6 LISS III
• Toposheet of the area (150,000)
• Field data (training sites, test sites, GPS
  locations)

5
Sahaspur, Rampur and adjoining area
(Dehradun dist.)
6
LISS IV LISS III
Methodology Flowchart
Images
Preprocessing stages
Separability analysis
Training Sites
Ground Truth
Maximum Likelihood Knowledge Based Object Based
Classification Methods
Accuracy Analysis
Maximum Likelihood Knowledge Based Object Based
Comparison
Prepare land use /land cover map
Final results
7
NRSA LANDUSE/ LANDCOVER CLASSIFICATION SCHEME
APPLIED ON STUDY AREA
No. First Level Second Level Third Level
1. Built up land Residential
Industrial
2. Agriculture land Cropland
Fallow land
3. Forest Evergreen Dense/Open
4. Water bodies River Dry/Perennial
Water
8
LISS III
FEATURE SPACE FOR LISS III AND LISS IV (MLC)
LISS IV
9
SEPARABILITY ANALYSIS FOR LISS III AND LISS IV
10
CLASSIFIED IMAGE OF LISS III AND IV (MLC)
LISS III
LISS IV
11
SEGMENTATION PARAMETERS FOR OBJECT-ORIENTED METHOD
LISS III
LISS IV
12
CLASS DESCRIPTION (OBJECT BASED)
Water (LISSIII)
Urban (LISSIII)
Agriculture (LISSIII)
Urban (LISSIV)
Water (LISSIV)
Agriculture (LISSIV)
13
FEATURE SPACE FOR LISS III AND LISS IV
(OBJECT-ORIENTED)
LISS III
LISS IV
14
FEATURE SPACE OF SPECTRALLY MIXED CLASSES (LISS
III OBJECT BASED CLASSIFICATION)
Dry river/Industrial
Urban/Agriculture
15
FEATURE SPACE OF SPECTRALLY MIXED CLASSES (LISS
IV OBJECT BASED CLASSIFICATION)

Dry river/Industrial
Residential/Dry river
Industrial/Urban
16
LISS III, IV CLASSIFIED IMAGE (OBJECT BASED)
17
RULES FOR EXPERT CLASSIFIER
18
LISS IV CLASSIFIED IMAGE (EXPERT CLASSIFIER)
Before Rule base classification
After Rule base classification
19
Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods. Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods. Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods. Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods. Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods. Table 6 Overall accuracies (OA) Kappa (K) achieved through various classification methods.
Dataset Pixel based Classification approach(MLC) Object based Expert classifier Increase in accuracy from MLC to Object Based Increase in accuracy from MLC to Expert classifier
LISS IV (OA) 71.59 89.26 80.94 17.67 9.35
LISS III (OA) 84.00 89.15 - 5.15 -
LISS IV (K) 62.57 86.04 74.88 23.47 12.31
LISS III (K) 80.33 86.66 - 6.33  
20
CONCLUSION

• On LISS III and LISS IV images up to second and
  third level of classification is possible but
  consideration of accuracy is needed.
• High spectral resolution of LISS III can provide
  some good results to separate classes as
  compare to LISS IV.
• Object based classification can also be
  applicable on LISS III and LISS IV images. But in
  LISS III it needs more parameters as compare to
  LISS IV.
• By the help of expert classifier the accuracy of
  maximum likelihood results can be improved by the
  help of some additional layers.

21
Thanking you for your kind attention