Dr. Iman Khosravi
Home Publication Projects Software Contact Me
Monitoring Urban Change Maps Using Object-Oriented Classification of Satellite Imagery
Supervisor: Dr. Mehdi Momeni
Abstract: The building detection algorithms using satellite imagery are divided into two groups:
Pixel-based and object-based and sometimes a combination of both. In the first methods,
only pixels and only their spectral attributes are used in the image analysis.
According to the previous studies, these methods are not proper for high-resolution image classification. Against, in the object-based methods, the processing unit is (homogeneous) group of pixels. Also, these methods can use the non-spectral attributes of the objects (such as proximity and geometry attributes) moreover their spectral attributes. These methods could produce proper and better results than pixel-based methods. However, the segmentation errors and the selection of the segmentation parameters have caused the limitations in the object-based methods.
In this master dissertation, we showed these limitations, too and then improved the results of (boundary) building extraction by defining the other elements (besides segments) as the objects. The proposed method was implemented on fifteen diverse Pan-sharped images of QuickBird satellite. Also, five previous algorithms (include of object-based and pixel-based methods) were implemented on these 15 images. Then, the proposed method was compared with other methods in terms of the performance and the reliability. For a better comparison, the images were divided into two groups: high contrast and low contrast datasets.
The results of group 1 showed the performance of the proposed method is higher than all other methods. In addition, the reliability of its results was not bad in comparison with other methods. Against, the performance of the proposed method was lower than the object-based methods at group 2. It shows that the proposed method may be confronted with the difficulties in the building extraction when there is the low contrast between the building regions and the adjacent regions. However, this method had the high performance than some methods at group 2.
The discussion on the Achilles heel and the advantages of these five methods and specially, the discussion on the limitations of the object-based methods (segmentation) was another basic goal of this dissertation.
Finally, this dissertation discussed on the quality control of a building extraction algorithm. All of the previous studies have used the extracted elements of the error matrix for the validation of their algorithm. However, some factors that are effective in the extracting of the building in an urban area have not been considered in the selection of the images of these studies. In addition, the current indices do not show the evaluating of an algorithm in relation to these factors. Therefore, in this dissertation we developed the evaluation indices for a building extraction algorithm using a variety of images.
Keywords: Map updating, Building detection, Satellite imagery, High resolution, Object-based method.
Presenting a Developed Method Based on Enhanced Multiple Classifier Systems for Crop- land Classification from Full-Polarimetric Synthetic Aperture Radar and Optical Images
Supervisor: Dr. Abdolreza Safari, and Dr. Saeid Homayouni
Abstract: Optical and polarimetric synthetic aperture radar (PolSAR) earth observations offer valuable sources of information for agricultural
applications and crop mapping.
Various spectral features, vegetation indices and textural indicators can be extracted from optical data.
These features contain information about the reflectance and spatial arrangement of crop types.
By contrast, PolSAR data provide quad-polarization
backscattering data and target decompositions, which give information about the structural properties and scattering mechanisms of different
Combining these two sources of information can present a complementary data set with a significant number of spectral, textural, and polarimetric features for crop mapping and monitoring. Moreover, a temporal combination of both observations may lead to obtaining more reliable results compared to the use of single-time observations. However, there are several challenges in cropland classification using this large amount of information.
The first challenge is the possibility correlation among some optical features or radar features which leads to redundant features. Moreover, some optical or radar features may have a low relevancy with some or all crop types. These two challenges cause to increase complexity and computational load of classification. In addition, when the ratio of number of samples to the number of features is very low, the curse of dimensionality may be occur. Another challenge of classification is the imbalanced distribution among various crop types, the so called imbalanced data.
For solving the first challenge, feature selection algorithms have been used in some studies. However, these algorithms may choose some irrelevant features. Thus, it is necessary to select optimum radar-optic features for cropland classification so that they have highest relevancy with respect to crop types and low correlation among themselves. One of the main aim of this thesis is to propose a strategic algorithm which has a simple and comprehensible structure relying on separability and dependency concepts. One of the unique advantage of this method is the directly selection of optimum features for each crop type.
In addition to feature selection methods, various classifier have been presented for cropland classification from optical and radar data. Among these classifiers, the multiple classifier systems (MCS) especially the random forest (RF) and stacked generalization have shown higher efficiency and flexibility. Thus, as another aim of this thesis, relying on these two concepts, a comprehensive crop mapping framework was presented which used the capability of various optical and radar features. Regarding the second challenge, none of classical algorithms even RF can solve it. Therefore, another aim of this thesis is an alternative to RF which is able to solve these two challenges, the curse of dimensionality and the imbalanced data, simultaneously. The proposed MCSs have other modifications in feature selection and fusion steps of RF.
Keywords: Multiple classifier systems, Cropland classification, Radar imagery, Optical imagery, Random forest.
Agricultural Crop Acreage Estimation Using Remote Sesning Technology
Supervisor: Dr. Seyed Kazem Alavipanah
Abstract: Combining optical and polarimetric synthetic aperture radar (PolSAR) earth observations offers a
complementary data set with a significant number of spectral, textural, and polarimetric features
for crop mapping and monitoring.
Moreover, a temporal combination of both sources of information may lead to obtaining more reliable results compared to the use of single-time observations.
In this project, an operational framework based on the stacked generalization of random forest (RF), which efficiently employed bi-temporal observations of optical and radar data, was proposed for crop mapping.
In the first step, various spectral, vegetation index, textural, and polarimetric features were extracted from both data sources and placed into several groups. Each group was classified separately using a single RF classifier. Then, several additional classification tasks were accomplished by another RF classifier. The earth observations used in this paper were collected by RapidEye satellites and the Unmanned Aerial Vehicle Synthetic Aperture Radar (UAVSAR) system over an agricultural region near Winnipeg, Manitoba, Canada.
The results confirmed that the proposed methodology was able to provide a higher overall accuracy and kappa coefficient than traditional stacking method, and also than all the individual RFs using each group. These accuracy metrics were also better than those of the RFs using the stacked features. Moreover, only the proposed methodology could achieve standard accuracy (F-score >= 85%) for all crop types in the study area. The visual comparison also demonstrated that the crop maps produced by the proposed methodology had more homogeneous, uniform appearances. Moreover, the mixed pixels of crop types, which abundantly existed in the traditional stacking and individual RFs' maps, were significantly eliminated.
Keywords: Multiple classifier systems, Cropland classification, Radar imagery, Optical imagery, Random forest.
Photography Camera Calibration: An Overview of Concept, Methods and Equations
Supervisor: Dr. Khosravi
Abstract: Camera calibration is favored as an important issue in photogrammetry and computer vision literatures.
The importance of this issue can be due to two reasons: firstly, every recently camera should be calibrated before being used to correct its
lens distortion and interior orientation elements.
In addition, it is a main preprocessing step at many vision applications. This researche provided an overview of concept, objective, methods and mathematical equations for camera calibration.
Section 1 dedicates on the elementary subjects in relation to definition and history of photogrammetry and also, the definition and the types of aerial photos. For pre-processing and photo processing, we need to define a coordinate system. Therefore, the types of object (ground) and photo coordinate systems are described as detail in section 2. Moreover, the most important subject of photogrammetry, i.e. collinearity condition principle, which is the relation of object and photo space and also, the base of all the calibration equations is discussed in this section. The errors of aerial photo, which the most important are the geometric errors of lens distortion are explained in section 3. The errors of lens distortion are the radial and tangential (decentering) distortions. In addition, the radial lens distortions are divided into pincushion and barrel distortion. Formularization and modeling of radial and decentering lens distortions are described as details in section 4. Some mathematical models for lens distortions are pointed out in this section which the most famous and most applicable model is model which is presented by Brown. In section 5, the algebraic, physical and hybrid methods for the modeling of camera systematic errors based on the basic and converted collinearity condition equations are described as details. Section 6 and 7 are the most important sections, in which the most important subject of this project, i.e. camera calibration is explained. In section 6, the concept and the goal of camera calibration and also, the methods in computer vision and photogrammetry for calibration are described. Finally, section 7 presents the mathematical equations of single-image calibration and multi-image self-calibration.
Keywords: Camera calibration, Self-calibration, Lens distortion, Interior orientation elements, Exterior orientation elements.