Analyze the Effect of Fusion of Different Remote Sensing Datasets on Image Classification

1
Analyze the Effect of Fusion of Different Remote
Sensing Datasets on Image Classification

• Prof madya Dr Juazer Rizal
• Arnisuhaila bt Rahmat
• 2003113246
• Noorita bt Sahriman
• 2003113239

2
Vision

• Apply remote sensing technique in making fusion
  image between SPOT-4 Panchromatic image and
  SPOT-4 Multispectral image.

3
Objective

• To visualize and interpret more clearly of Spot-
  4 Multispectral or Spot- 4 Panchromatic remote
  sensing data based on image transformation
  technique.
• To observe the effect of the fusion of this data
  through the use of HSV approach
• To analyze the result of classification before
  and after the fusion.
• To study the technique in making a data fusion
  that includes a data from spot-4 Multispectral
  and spot-4 Panchromatic at the same study area.

4
Study Area-Kuala Muda
5
Fusion

• Wald (2002) describes fusion as a formal frame
  work in which are expressed means and tools for
  the alliance of data originating from different
  sources. It aims at obtaining information of
  greater quality the exact definition of greater
  quality will depend upon application.
• Data fusion is the combination of multi source
  data which have different characteristics such
  as, temporal, spatial, spectral and radiometric
  to acquire high quality image. The fusion of
  different sensor images is crucial method for
  many remote sensing applications such as land
  cover/ land use mapping.
• The fusion technique can be categorized according
  to the processing level at which the fusion takes
  place-pixel, feature and decision level. Fusing
  data at pixel level requires co-registered images
  at sub-pixel accuracy. (H.H. Yoo, 2002,)
• The pixel based fusion technique applied to
  remote sensing data is grouped into two major
  categories that are colour transformations and
  statistical and numerical methods.

6

• For all the fusion process, some pre-requisite
  are needed
• Images shall have different spatial and spectral
  resolutions.
• Images to be merged shall represent the same
  area.
• Images shall be registered accurately.
• No major changes shall occur in the area during
  the interval between time acquisitions of the
  source images.
• Among fusion methods for fusing multi-spectral
  images of low resolution and high resolution, the
  most common methods are HSV (Hue Saturation
  Value), HPF (High Pass Filter), PCA (Principal
  Component Analysis) Wavelet Fusion and Brovey
  Transform

7
HSV Method

• A color space in terms of three constituent
  components (hue, saturation,value)
• The HSV transform separates spatial (V) and
  spectral (H, S) information from a standard RGB
  image
• The ranges of hue is from 0-360, saturation is
  from 0-100 , value is 0-100 .
• The method is limited to three bands for the
  lowest spatial resolution dataset. It cannot be
  applied to all bands that exist in the dataset.
• We prefer to use the HSV color model over
  alternative models such as RGB or CYMK because of
  its similarities to the way humans tend to
  perceive color
• HSV method is easy to apply, it has the
  disadvantages because it allows only three bands
  to be applied. And if replaces intensities
  corresponding to images sizes, so its sharpness
  tends to be degraded

8
Problem Statement

• Remote sensing image in Spot-4 Multispectral are
  prone to cloud problem. This cloud pours some
  difficulty on image interpretation. This will
  also affect the accuracy of interpretation in
  that image.
• Spot-4 Multispectral has 20 meter spatial
  resolution but have a good spectral resolution.
  While, Spot-4 Panchromatic have 10 meter spatial
  resolution but have low spectral resolution. By
  fusing from spot 4 panchromatic we believe that
  information extracted from that image will become
  clearer and easily interpreted. The image will
  have good spatial resolution (10m from Spot-4
  Panchromatic) and good spectral resolution (from
  Spot-4 Multispectral).

9
(No Transcript)
10
Data

• SPOT-4 Multispectral (20m)
• Path/Raw 266/339
• Date of Acquisition 18/07/2004
• Number of Bands 4
• SPOT-4 Panchromatic (10m)
• Path/Raw 266/339
• Date of Acquisition 18/07/2004
• Number of Bands 1

11
Methodology
12
Land cover data From SPOT-4-Multispectra
l
Land cover data From SPOT-4 Panchromatic
Geometric Correction (Image to Image)
Resample (Nearest Neighbour)
Geocoded Image
Subset
Spot 4 Multispectral
Spot 4 Panchromatic
Fusion Image
Classification
Unsupervised
Supervised

• Clump
• Eliminate
• Recode

Accuracy Assessments
Compare the accuracy assessment between
multispectral image and fusion image. Choose the
best supervised and unsupervised.
13
Classification Result
14
Supervised (SPOT-4 Multispectral)
15
Supervised (Fusion Image)
16
Unsupervised (SPOT-4 Multispectral)
17
Unsupervised (Fusion Image)
18
Analysis
19
Analysis 1 Image Brightness

• SPOT-4 Multispectral has low spatial resolution.
  The features and boundaries can not be seen
  clearly. The types of paddy field are quite same
  with other plant area such as forest and
  mangrove. However, SPOT-4 Multispectral has high
  spectral resolution. These help in determine the
  area that have high-density and low-density based
  on the brightness on that area. Sediment
  transportation from river to sea through estuary
  can be clearly seen on this image.

20

• Fusion image has high spatial resolution from
  SPOT-4 Pancromatic and has high spectral
  resolution from SPOT-4 Multispectral. This image
  show more clearly feature on the image such as
  partition of paddy field area. The urban area can
  be obviously seen especially their shape and the
  structures of the buildings. The small river
  tracks are sharpening, thus the deltas around the
  river area can be show. The roads in the image
  are shown and this is better than SPOT-4
  Multispectral.

21
Analysis 2 Classification Accuracy

• Supervised classification

22

• Unsupervised classification

23
(No Transcript)
24
Conclusion

• Based on our result in this special project, we
  have several conclusions
• to verify that fusion image will give a better
  interpretation in study area
• Image before and after fusion are obviously
  different especially in their brightness and
  sharpness of the features in the image.
• The characteristic of fusion image are containing
  10m spatial resolution from spot-4 panchromatic
  and 3 spectral resolution from SPOT-4
  Multispectral.
• From two types of classification, unsupervised
  classification is suitable in fusion image
  because the classes of area are decided by the
  user.
• The total accuracy from two classification shows
  that fusion image have the highest accuracy
  better than SPOT-4 Multispectral.
• The fusion process disadvantages are distorting
  the spectral information and only allow three
  bands at one time.
• Finally, we have success in produce fusion image
  that fulfill every requirement in our objective.

25
Opinion

• We suggest that fusion image can be applied if
  multispectral and panchromatic image cannot give
  the accurate features interpretation since fusion
  image are proved to have better accuracy.
• The fusion method can be improve by add more than
  3 band so that more band can be put in the
  process.
• Fusion image can also be used in produced land
  use and land cover map.

26
Thank You