This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: Spectral regions for training categories may intersect or overlap (in such a case classes are assigned in sequence of classification). A second disadvantage is that other parts of the image may remain unclassified because they do not ‘fall’ into a box. The maximum likelihood classifier is the most advanced classifier but it requires a considerable amount of computation time. As computers have become very fast and powerful, the latter is no longer a problem and the maximum likelihood classifier is widely used nowadays. The maximum likelihood approach does not only take into account the average DN values of the training areas, it also accounts for the variance of pixel values of the training areas. The variances are used to estimate the probability of membership for a certain land cover class. Figure 5.21 shows a threedimensional plot with the probability density functions of several land cover classes in spectral band 3 (visible red) and 4 (near infrared). The variance of the pixel values of the class ‘sand’ is very small (a distinct peak), the variance for urban is large. (Compare this with the Veluwe TM image exercise). Notice that the variance for the class ‘water’ is larger in band 3 than in band 4. Figure 5.18 Illustration of the Euclidean distance measure (Campbell, 1987). 82 Figure 5.19 Parallelepiped classifier (Campbell, 1987). The equiprobability contours shown in figure 5.20 serve as a decision rule to assign pixels to certain land cover classes. The equiprobability contours are drawn using information from the training areas. The probability differs from one spectral band to another. The maximum like lihood classifier is a very powerful classifier but it is sensitive for the quality of the training data: the likelihoods (or probabilities) are computed based on the assumption that the training data have a multivariate, normal (Gaussian) frequency distribution. Remote sensing data do not always fulfil this condition, you should check the distribution of your training data before using this classifier. The maximum likelihood classifier is based on the Bayesian decision rule. This technique allows to use socalled ‘ a priori information’ during the classification. If a certain crop grows e.g. only on a specific type of soil and a digital soil map is available, the classification can be directed by using this information by: ‘it is very unlikely that this crop will occur on other soil type than...’. Examples and procedures are described by Strahler (1980), De Jong and Riezebos (1991) and Jansen (1994). 83 Figure 5.20 & 5.21 Equiprobability contours defined by the maximum likelihood classifier. Probability density functions defined by the maximum likelihood classifier (Lillesand & Kiefer, 1994). 84 Only a few types of classifiers are described in these lecture notes. Many other approaches to the classification of digital images have been developed. It is not possible to state that a given classifier is ‘best’ for all circumstances, because the characteristics of each image, the conditions...
View
Full Document
 Winter '12
 JOHN
 Remote Sensing, Electromagnetic spectrum, µm, Infrared

Click to edit the document details