Geodeziya, kartografiya, geografiya


ISODATA unsupervised classification



Yüklə 29,69 Mb.
səhifə55/121
tarix09.10.2023
ölçüsü29,69 Mb.
#153376
1   ...   51   52   53   54   55   56   57   58   ...   121
Geodeziya, kartografiya, geografiya

ISODATA unsupervised classification
The ISODATA algorithm is the most common unsupervised classification tool. It is used to create a user-defined number of clusters or classes in an image that must later be labeled to create a land cover map. Before you run an ISODATA program you need to define several parameters that control how many clusters will be generated, how many iterations will be run, and other parameters that control how clusters are combined and split as the program executes.
When the program is finished you will have an image with several classes which will not be labeled. For example, if you specified that the ISODATA program should create 25 classes the output map will have 25 classes with each comprised of pixels with similar pixel values. The next step is to label these classes with the correct land cover type. If you were trying to create a map with two classes, forest and non-forest, you would look at each of the 25 classes in the ISODATA output image and label them as either forest or non-forest. In some cases it will be difficult to decide how to label a particular class because it will contain more than one land cover type. When this happens you can choose to run the ISODATA program again to output more classes or you can apply the ISODATA program to run only on those classes that contain multiple land cover types, a technique known as "cluster-busting".
Supervised statistical classification
There are several types of statistics-based supervised classification algorithms. Some of the more popular ones are (in increasing complexity); parallelepiped, minimum distance, maximum likelihood, and mahalanobis distance. With supervised statistical classification algorithms the analyst must first locate and define samples in the image of each class that are required for the final map. For example if you were interested in creating a map with forest and non-forest classes you would select sample areas in the image that represent the different types of forest and non-forest. These samples are called training areas. Remote sensing software that supports supervised classification provides tools to allow users to draw lines around training areas and label them. Once a sufficient number of training areas are selected you can run the supervised classification. The algorithm then compares each pixel in the image with the different training areas to determine which training area is most "similar" to the pixel in the image. Once the most "similar" training area is found the image pixel is labeled with the according land cover class.
The difference between the different types of supervised statistical classification algorithms is how they determine similarity between pixels. In general, the more complex algorithms take longer to process but they also tend to do a better job at assigning the right land cover label to image pixels.

Yüklə 29,69 Mb.

Dostları ilə paylaş:
1   ...   51   52   53   54   55   56   57   58   ...   121




Verilənlər bazası müəlliflik hüququ ilə müdafiə olunur ©azkurs.org 2024
rəhbərliyinə müraciət

gir | qeydiyyatdan keç
    Ana səhifə


yükləyin