Data-driven evaluation of SAR information extraction methods 

Get Complete Project Material File(s) Now! »

Fundamentals of SAR

RADAR is an acronym used for RAdar Detection And Ranging. It refers to a method developed for target acquisitions and distance measurements by means of radio or microwave measurements. A radar system has three primary functions: 1) It transmits microwave (radio) signals towards a scene; 2) It receives the portion of the transmitted energy backscattered from the scene; and 3) It observes the strength (detection) and the time delay (ranging) of the return signals. Radar provides its own energy source and, therefore, can be operated both during day or night and through cloud cover. Hence, images can be acquired completely independently of solar illumination, which is particularly important in high latitudes ( polar night ) or cloudy regions. This type of system is known as an active remote sensing system.
The modern uses of radar are highly diverse, including air traffic control, radar astronomy, military application such as : air-defense systems, antimissile systems, nautical radars to locate landmarks and other ships, guided-missile target-locating systems, etc. Moreover, during the last twenty years, radar has been widely used in remote sensing imagery and Earth observation. Imaging radar is an active illumination system, which is used for image generation. It has an antenna, mounted on a platform, which transmits a radar signal in a side-looking direction towards the Earth’s surface. The reflected signal, known as the echo, is backscat52 tered from the surface and received a fraction of a second later at the same antenna. Two types of imaging radar systems with side-looking view can be distinguished 1) systems with real aperture (RAR: Real Aperture Radar) and 2) systems using synthetic aperture radar (SAR: Synthetic Aperture Radar). In the radar context, aperture means the length of the antenna. In the following, we focus on the use of SAR images and point out only the most important characteristics of SAR systems. More details of SAR systems can be found on (Bamler and Sch¨attler, 1993), (Oliver and Quegan, 1998).

Geometrical and Radiometrical effects

SAR systems detect the distance between the sensor emitting a microwave pulse and the target reflecting the energy back to the receiving antenna. This range measurement principle leads to specific geometric distortions in the processed SAR image, which makes SAR images more difficult to interpret than optical images. In the following, the most common geometric distortions presented in the specific topography are described in section 2.2.1. A review of the radiometric and geometric corrections are detailed in section 2.2.2 and 2.2.3, respectively.

Statistical properties of the backscattered signal

An immediate conclusion from (2.12) is that the observed signals will be affected by inference effects as consequence of the phase differences between scatterers. Actually, speckle can be understood as an interference phenomenon in which the principal source of the noise-like quality of the observed data is the distribution of the phase terms Φk. To infer this distribution, note that the slant range resolution is typically many wavelength across. Hence, scatterers at different parts of the resolution cell will contribute very different phases to the return (2.12) even if their scattering behavior is identical. In practice, we can think of the phase Φk as being uniformly distributed in [−π, π] and independent of the amplitude Ak. The sum (2.12) then looks like a random walk (See Figure 2.11(b)) in the complex plane, where each step of length Ak is in a completely random direction. For a large number of statistically identical scatterers (N), several properties of the received signal can be derived (Goodman, 1975; Oliver and Quegan, 1998):

Table of contents :

1 Introduction 
1.1 Motivation
1.2 Goals of the thesis
1.3 Contribution of the thesis
1.4 Outline of the thesis
2 Synthetic Aperture Raadr 
2.1 Fundamentals of SAR
2.1.1 Principle of SAR
2.1.2 SAR image formation
2.1.2.1 Range resolution
2.1.2.2 Azimuth resolution
2.1.3 Basic scattering mechanisms
2.1.3.1 Surface and Volume Scattering
2.1.3.2 Single and Double Bounce
2.2 Geometrical and Radiometrical effects
2.2.1 Geometrical effects
2.2.1.1 Foreshortening
2.2.1.2 Layover
2.2.1.3 Shadows
2.2.2 Radiometric corrections
2.2.3 Geometric corrections
2.3 SAR image statistics
2.3.1 Speckle effect
2.3.2 Statistical properties of the backscattered signal
2.3.3 Statistical properties of Multi-Look data
2.3.3.1 Intensity images
2.3.3.2 Square-root intensity images
2.3.3.3 Amplitude images
2.3.4 The multiplicative noise model for Speckle
2.4 TerraSAR-X products and information content
2.4.1 TerraSAR-X Instrument
2.4.1.1 Imaging Modes
2.4.2 Basic TerraSAR-X products
2.4.2.1 Geometric resolution
2.4.2.2 Geometric projections and data representation
2.4.2.3 Product identification Scheme
2.4.3 Examples of TerraSAR-X image content
3 Image Information Mining systems 
3.1 Content-Based Image Retrieval
3.2 Image Information Mining Systems
3.2.1 Generic Architecture
3.2.1.1 Image database
3.2.1.2 Feature extraction
3.2.1.3 Data reduction and content index generation
3.2.1.4 User interaction or human-machine communication
3.2.2 Examples of IIM systems
3.3 Evaluation in Image Information Mining systems
3.3.1 Data-driven evaluation approach
3.3.1.1 Component by component evaluation
3.3.1.2 General evaluation of the performance
3.3.2 User-driven evaluation approach
3.4 Conclusions
4 SAR image analysis using Bayesian inference and Gibbs Random Field models 
4.1 Bayesian inference
4.1.1 Conditional probability and Bayes’ theorem
4.1.2 Parameter estimation
4.1.3 Model Selection
4.2 Markov and Gibbs Random Fields
4.2.1 Image description, neighbourhood systems and cliques
4.2.2 Markov-Gibbs Random Fields equivalence
4.2.2.1 Hammersley-Clifford theorem
4.2.3 Examples of Gibbs Random Field models
4.2.3.1 Gauss Markov Random Field model
4.2.3.2 Auto-binomial model
4.2.3.3 Examples of realizations of GMRF and ABM models
4.3 Bayesian SAR image analysis
4.3.1 SAR despeckling filters
4.3.1.1 Non-Bayesian despeckling filters
4.3.1.2 Bayesian despeckling filters
4.3.2 SAR information extraction
4.4 Conclusion
5 Data-driven evaluation of SAR information extraction methods 
5.1 Description of SAR information extraction model-based methods
5.1.1 Bayesian Approach
5.1.2 Maximum a Posteriori despeckling using Gauss Markov Random Field (MAP-GMRF) and evidence maximization
5.1.3 Maximum a Posteriori despeckling using Auto-binomial model (MAPABM) and evidence maximization
5.1.4 Overview of MAP-GMRF and MAP-ABM
5.2 Description of the test data set
5.2.1 Simulated SAR data
5.2.1.1 Synthetic textures
5.2.1.2 Speckled Brodatz textures
5.2.2 Real SAR data
5.2.2.1 Textures from TerraSAR-X images
5.2.2.2 TerraSAR-X sub-scene
5.3 Quantitative evaluation of despeckling
5.3.1 Quantitative quality measurements
5.3.1.1 Metrics for Simulated SAR data despeckling evaluation
5.3.1.2 Metrics for Real SAR data despeckling evaluation
5.3.2 Experimental results
5.3.2.1 Despeckling results based on Simulated SAR data
5.3.2.2 Despeckling results based on Real SAR data
5.3.2.3 The impact of window size and model order on despeckling performance
5.3.3 Conclusions
5.4 Quantitative evaluation of SAR information extraction
5.4.1 Quantitative Metrics
5.4.1.1 Classification and Confusion Matrix
5.4.2 Robustness of feature extraction evaluation in terms of classifications
5.4.2.1 Classification results based on Simulated SAR data
5.4.2.2 Classification results based on Real SAR data
5.4.3 The impact of window size and model order on feature extraction
5.4.4 Conclusions
5.5 Conclusions
6 User-driven evaluation of Image Information Mining systems 
6.1 Knowledge-based Image Information Mining system as an Image Information Mining tool
6.1.1 KIM Concept
6.1.2 KIM Architecture
6.1.2.1 GUI Components
6.1.3 Operation of KIM
6.1.3.1 The Server-side
6.1.3.2 The Client-side
6.2 Effectiveness evaluation metrics
6.2.1 Precision and Recall
6.2.2 Accuracy and F-measure
6.2.3 Probability to over-retrieve and to forget an image
6.2.4 Target and misclassified images
6.3 Evaluation Methodology
6.3.1 Role of the user
6.3.1.1 The user as interpreter of the ground-truth
6.3.1.2 The user as evaluator of the system
6.3.2 Creation of the ground-truth
6.3.2.1 Definition of the relevance criterion
6.3.2.2 Definition of the scheme for annotating images
6.3.2.3 Creation of the reference database
6.3.3 System evaluation methods
6.3.3.1 Configuration of the scenario
6.3.3.2 Training performed by the user and definition of semantic labels
6.3.3.3 Assessment of the probabilistic search and interactive learning
6.4 Examples of validation scenarios: Disaster monitoring
6.4.1 Study case 1: Oil Spill in Gulf of Mexico
6.4.1.1 Description of data set, parameters and ground-truth
6.4.1.2 Training performed by the end-user
6.4.1.3 Evaluation results of the probabilistic search
6.4.1.4 Classification results
6.4.1.5 User satisfaction degree
6.4.2 Study case 2: Nepal-Embankment breach
6.4.2.1 Description of the study area, data set and ground truth
6.4.2.2 Training performed by the end-user
6.4.2.3 Evaluation results of the probabilistic search
6.4.2.4 Classification results
6.4.2.5 User satisfaction degree
6.5 Conclusions
7 Conclusions 
7.1 Summary of the Thesis
7.2 Future work
A TerraSAR-X image content
B Evidence maximization approach
C Validation scenarios examples
Bibliography

READ  Connections between the stochastic heat equation(s) and the directed polymers in random environments

GET THE COMPLETE PROJECT

Related Posts