Lec 10 - Information Extraction Flashcards
Purposes of Image Extraction (4)
(1) Categorizing data
(2) Data simplification
(3) Data interpretation
(4) Mapping
Overall objective of classification
Automatically categorize all pixels in an image into land cover classes or themes
Selection of Classification (3)
(1) The nature of the data being analyzed
(2) The computational resources available
(3) The intended application of the classified data
Pattern Recognitions (3)
(1) Spatial pattern recognition
(2) Spectral pattern recognition
(3) Temporal pattern recognition
Use spatial context to distinguish between different classes
Spatial Pattern Recognition
Most widely-used pattern recognition that distinguish between different land cover classes from differences in the spectral reflectance
Spectral Pattern Recognition
The ability to distinguish patterns based on spectral or spatial considerations that may vary over the year
Temporal Pattern Recognition
Land Cover Mapping and Applications of Remote Sensing
(1) Understanding of the type and amount of land cover in an area is an important characteristic
(2) Remote sensing has become a powerful tool for land cover identification and classification
(3) The investment in the development of this technology has contributed to Precision Agriculture
Steps in Thematic Information Extraction from Satellite Images
(1) Definition of the mapping approach
(2) Geographical stratification
(3) Image segmentation (for object-oriented classification)
(4) Feature Identification and Selection
(5) Classification
The unit to which the classification algorithms will be applied.
Spatial unit of analysis
The study area is divided into smaller areas (strata) so that each strata can be processed independently
Geographical stratification
The division of an image into spatially continuous, disjoint and homogenous regions
Image segmentation
The manipulation and selection of features are used to reduce the number of features without sacrificing accuracy
Feature Identification and Selection
The process of partitioning an image data set into a discrete number of classes in accordance with specific criteria that are based, in part, on the individual image point data values (recognize patterns).
Classification
This depends on the type of information you want to extract from the original data. May simply represent areas that look different to the computer or may be associated with known features on the ground.
Classes
Types of Classification (based on type of learning
(1) Supervised
(2) Unsupervised
(3) Hybrid
Type of classification that requires “training pixels”, pixels where both the spectral values and the class is known.
Supervised
Type of classification where no extraneous data is used: classes are determined purely on difference in spectral values.
Unsupervised
Use unsupervised and supervised classification together
Hybrid
Supervised Classification Steps
(1) Training stage
(2) Classification stage
(3) Output stage
Pixel observations from training sites are plotted on scatter diagrams and are used as a basis of setting up the statistical parameters used to classify pixels outside these sites.
Classification stage
This type of classifier determines the mean value (DN) of each class in each band, and then assigns unknown pixels to classes whose means are most similar to the value of the unknown pixel.
Minimum Distance to Means
Benefit of Minimum Distance to Means
This method is quite computationally efficient since it is mathematically simple, which made it a good choice before the advent of modern computers.
Drawback of Minimum Distance to Means
Insensitive to different degrees of variance in spectral response data
The most powerful classifier in common use. A statistical decision rule that examines the probability function of a pixel for each of the classes, and assigns the pixel to the class with the highest probability.
Maximum Likelihood
Benefit of Maximum Likelihood
Takes variation in spectral response into consideration
Drawback of Maximum Likelihood
Computationally intensive; multimodal or non-normally distributed classes require extra care when training the classifier, if high accuracy is to be achieved
This works by delineating the boundaries of a training class using straight lines.
Parallelepiped
Benefit of Parallelepiped
This method is computationally fast, simple to train and use
Drawbacks of Parallelepiped
Using straight lines to delineate the classes limits the method’s effectiveness. Also, having pixels classified as unknown may be undesirable for some applications
It is based on correlations (statistics) between variables by which different patterns can be identified and analyzed.
Mahalanobis Distance
Any individual pixel is compared to each discrete cluster to see which one it is closest to.
Unsupervised Classification
Advantage of Unsupervised Classification
No extensive prior knowledge required opportunities of human error minimized unique classes recognized as distinct units
Disadvantage of Unsupervised Classification
Limited control over classes and identities
A more sophisticated version of the K-Means classifier which allows classes to be created and destroyed.
Iterative self-organizing data analysis (ISODATA)
Each point has a degree of belonging to clusters (fuzzy logic) rather than belonging completely to just one cluster.
Fuzzy C Means