Meeting 02/03/2012

New results, using the seasonal feature of polygon area per quadrant

Typology based on MOD12Q1 (2004-2005), using 11267 instances (Water, Evergreen Broadleaf forest, Deciduous Broadleaf forest, Open shrublands, Woody savannas, Croplands)

Typology (Single crop, Double crop, other classes)

Using a combination of features (Basic + Polar), 90% of correct classified points (2074 instances).

Results presented in paper

Basic features X Polar Features. Typology (Single crop, Double crop, other classes)

TerraClass (2008) X GeoDMA (growing season 2007-2008)

scene 22469 with 58110 points, and 72% of correct classified points.
scene 22670 with 13629 points, and 85% of correct classified points.
scene 22768 with 83861 points, and 81% of correct classified points.
scene 22969 with 10890 points, and 15% of correct classified points.
scene 22769 with 60286 points, and 77% of correct classified points.
scene 22668 with 65784 points, and 63% of correct classified points.
scene 22770 with 29331 points, and 31% of correct classified points.
scene 22669 with 97752 points, and 77% of correct classified points.

Thesis

First Version

1 Introduction

1.1 Scientific Question and Hypothesis
1.2 Contributions
1.3 Document Organization

2 Review

2.1 Landscape Ecology
2.2 Object-Based Image Analysis
2.3 Data Mining
2.4 Multitemporal analysis

3 Method

3.1 Defining the input data
- 3.1.1 Segmentation
- 3.1.2 Cycles detection
3.2 Feature Extraction
- 3.2.1 Segmentation-based
- 3.2.2 Landscape-based features
- 3.2.3 Multitemporal
  - 3.2.3.1 Basic features
  - 3.2.3.2 Linearity features
  - 3.2.3.3 Features based on polar representation
3.3 Data mining to detect land change
- 3.3.1 Building a reference set
- 3.3.2 Classification using decision trees
3.4 Output
- 3.4.1 Measuring change

4 Results

4.1 Land cover using intra-urban imagery
- 4.3.1 Input dataset
- 4.3.2 Experiment
4.2 Detection of patterns of the urban landscape
4.3 Classification of agriculture cycles
4.4 Evaluating the new multitemporal features