Pattern Recognition in AVHRR Images by means of Hibryd NeuroFuzzy Systems and Fuzzy Lattice Neurocom

1
Pattern Recognition in AVHRR Images by means of
Hibryd Neuro-Fuzzy Systems and Fuzzy Lattice
Neurocomputing Model

• J. A. Piedra, M. Canton and F. Guindos

2
TABLE OF CONTENTS

• Problem.
• Automatic Interpretation System.
• Feature Selection.
• Classifiers Neuro-Fuzzy Systems.
• Results.
• Conclusions.
• Future works.

3
1. Problem.

• The impossibility of interpreting all the data
  received from earth observation satellites.
• A series of factors combine to make the
  identification of oceanic structures.
• The degree of thermal and morphological
  variability present in oceanic structures.
• The identification process is influenced by a
  very significant contextual factor

4
1. Problem.

• Study zone.
• Advanced Very High Resolution Radiometer (AVHRR)

5
1. Problem.

• Ocean structures.

10/08/1993
6
2. Automatic Interpretation System
7
3. Methodology of Feature Selection.
8
3. Methodology of Feature Selection.

• Experiment
• Upwellings 2,3 y 4
• Wakes 5,6,7,8,9 y 10
• Anticyclonic eddy 11
• Cyclonic eddy 12

9
3. Methodology of Feature Selection.

• Experiment

10
3. Methodology of Feature Selection.

• Results

Equal Frequency 100
K-Means 100
Expectation Maximization (K-Means 9)
11
3. Methodology of Feature Selection.

• Results

Naive Bayes
TAN
12
4. Classifiers.

• Graphic Expert System (GES).
• ANN-based Symbolic Processing Element (SPE).
• Bayesian Network Naïve Bayes and TAN.
• Neuro Fuzzy Hybrid System.

13
4. Classifiers.
Linguistic expressions IF ... THEN ...
Fuzzy Logic
Hybrid System RBF Sugeno Neuro Fuzzy
Systems ANFIS NEFCLASS NEFPROX Fuzzy Lattice
Neurocomputing Model FLNMAP
Linguistic data
Neural Network
NeuroFuzzy Hybrid
Numeric data
Learning Parallel Computation
14
5. Results.
15
5. Results.
20/05/1990
19/05/1990
10/08/1993
27/04/1988
16
6. Conclusions.

• Finally, the structure of the automatic
  interpretation system that has been introduced,
  is a complex and independent structure for the
  following tasks
• Iterative segmentation-recognition cycle is made
  by means of graphic expert system.
• Feature selection and validation of knowledge is
  done by bayesian learning.
• Multiple classification allows the discovery of
  different interpretations of the existing
  knowledge to validate the knowledge of the
  classifiers.
• Neuro fuzzy systems improve the results in the
  classification and allows the construction of
  knowledge by means of fuzzy rules.

17
7. Future works.

• To increase the interpretability of neuro fuzzy
  system maintaining an acceptable approximation
  results.
• The application of algorithms for the learning of
  mixed fuzzy rules with symbolic and numerical
  data in neuro fuzzy systems.
• The combination of multiple classifiers to
  improve the balance between understandability
  (the user), performance (accuracy rates) and cost
  (building) by means of sFLNMAP.

18
THANKS FOR YOUR ATTENTION
19
Pattern Recognition in AVHRR Images by means of
Hibryd Neuro-Fuzzy Systems and Fuzzy Lattice
Neurocomputing Model

• J. A. Piedra, M. Canton and F. Guindos