Slides ali-icomet2018

Supervised Classification for Object Identification
in Urban Areas using Satellite Imagery
1
Presentation by
Dr. Hazrat Ali
COMSATS Institute of Information Technology Abbottabad
Adnan Ali Awan, Sanaullah Khan, Omer Shafique, Atiq ur Rahman, Shahid Khan
COMSATS Institute of Information Technology Abbottabad, Pakistan
Hamad Bin Khalifa University, Qatar
University of Lorraine, France
Dr. Hazrat Ali @ 2018 International Conference on Computing, Mathematics and Engineering Technologies – iCoMET 2018

Outline
1. Introduction
2. Objective
3. Generalized Procedure
4. Dataset
5. Proposed Method
6. GLCM Features
7. Classification
8. Results
2Dr. Hazrat Ali @ 2018 International Conference on Computing, Mathematics and Engineering Technologies – iCoMET 2018

3
Introduction
Aerial and satellite imagery consists of pictures taken from aircraft and
from man-made satellites orbiting the Earth.
These images are useful for weather prediction or navigation.

Objective
To identify objects in aerial imagery of urban areas.

Generalized Procedure
5
Image pre-processing
Feature Extraction
Model Training
Testing
Results

Dataset
 3D Semantic Labeling Contest
 The dataset is obtained from the International Society for Photogrammetry
and Remote Sensing (ISPRS)
 http://www2.isprs.org/
 A total of 20 images

Proposed Method
Feature Extraction Techniques
 GLCM (Textural Features)
 Local Spatial Statistical features
 Moran’s I (A kind of Local Spatial
Statistical feature)
 Color features
 Edge features
 Shape features
Classification Algorithms
• Naïve Bayes
• Support Vector Machine
• K Nearest Neighbor
• Neural Networks
• Decision Trees
• Linear Discriminant Analysis

Block Diagram
Window
Selection ResultsClassification
Image
Feature
Extraction
Entropy
Homogeneity
Energy
Contrast

GLCM Features
• Grey Level Co-occurrence Matrix (GLCM)
• Number of count of intensity value with
another intensity value in a specific
relationship.
• The GLCM is a tabulation of how often
different combinations of gray levels co-occur
in an image or image section.
• GLCM is based on second order statistics.

GLCM Features
• Grey Level Co-occurrence Matrix (GLCM)
• Number of count of intensity value with
another intensity value in a specific
relationship.
• The GLCM is a tabulation of how often
different combinations of gray levels co-occur
in an image or image section.
• Based on second order statistics.
• GLCM Features:
1. Homogeneity 𝑖 𝑗 𝑃𝑑,𝑟(𝑖, 𝑗)
2. Contrast 𝑖 𝑗[ 𝑖. 𝑗 2 𝑃𝑑,𝑟 𝑖. 𝑗 2]
3. Energy 𝑖 𝑗 𝑃𝑑,𝑟 𝑖. 𝑗 2
4. Entropy 𝑖 𝑗 𝑃𝑑,𝑟 𝑖, 𝑗 [−ln[ 𝑗 𝑃𝑑,𝑟 𝑖. 𝑗 2
]]

Classification

A family of simple probabilistic classifiers based on applying Bayes' theorem.
𝑃 𝐶 𝑋 =
𝑃 𝑋 𝐶 𝑃(𝐶)
𝑃(𝑋)
𝑃𝑜𝑠𝑡𝑒𝑟𝑖𝑜𝑟 =
𝐿𝑖𝑘𝑙𝑖ℎ𝑜𝑜𝑑 × 𝑃𝑟𝑖𝑜𝑟
𝐸𝑣𝑖𝑑𝑒𝑛𝑐𝑒
𝑃 𝐶 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑐𝑙𝑎𝑠𝑠
𝑃 𝑋 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑑𝑎𝑡𝑎
𝑃 𝐶 𝑋 = 𝑃𝑟𝑜𝑏𝑎𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑐𝑙𝑎𝑠𝑠 𝑔𝑖𝑣𝑒𝑛 𝑑𝑎𝑡𝑎
𝑃 𝑋 𝐶 = 𝑃𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑑𝑎𝑡𝑎 𝑔𝑖𝑣𝑒𝑛 𝑡ℎ𝑒 𝑐𝑙𝑎𝑠𝑠
Naïve Bayes

A Support Vector Machine (SVM) is a discriminative classifier formally defined by a
separating hyperplane.
The optimal separating hyperplane maximizes the margin of the training data.
Support Vector Machine

Pre-processing
Input Image

Pre-processing
RGB to Grey level

Segmentation
with 70×70 window size
Input textural image Resulting segmentation GLCM
Run time 27.6 sec

Segmentation
with 50×50 window size
Input textural image Resulting segmentation GLCM
Run time 45.2 sec

Experimental Results
S.No Homogeneity Contrast Energy Entropy Naïve Bayes SVM
50x50 70x70 50x50 70x70 50x50 70x70 50x50 70x70 50x50 70x70 50x50 70x70
1 0.9323 0.9310 0.1654 0.1650 0.4636 0.4114 0.7892 0.8413 72.79% 71.63% 67.94% 64.52%
2 0.9409 0.9365 0.7658 0.7632 0.5094 0.4426 0.7501 0.08637 72.79% 71.63% 67.94% 64.52%
3 0.9346 0.9355 0.1550 0.1566 0.4791 0.4145 0.7494 0.8604 70.47% 70.37% 66.92% 66.07%
4 0.9018 0.9317 0.1837 0.1701 0.3951 0.3751 0.8036 0.8630 70.37% 70.17% 67.07% 66.92%

60.00%
62.00%
64.00%
66.00%
68.00%
70.00%
72.00%
74.00%
Naïve Bayes SVM
50x50
70x70
Visualization

Some challenges
• Experiments were performed on a CPU. Thus, the available RAM resources (4
GB only) were not sufficient for larger dataset.
• Most recent datasets are paid.
• In future:
• GPU based computation
• Deep learning approaches particularly generative models.

References
• [Available online] http://www.isprs.org/ (Accessed 2016)
• Chih-Chung Chang and Chih-Jen Lin, LIBSVM : a library for support vector machines. ACM
Transactions on Intelligent Systems and Technology, 2:27:1--27:27, 2011. Software available
at http://www.csie.ntu.edu.tw/~cjlin/libsvm
• R. M. Haralick, "Statistical and structural approaches to texture," in Proceedings of the IEEE,
vol. 67, no. 5, pp. 786-804, May 1979.

Thank you
23
Contact: hazratali@ciit.net.pk

Slides ali-icomet2018

More Related Content

Similar to Slides ali-icomet2018 (20)

Recently uploaded (20)

Slides ali-icomet2018