Multispectral image classification

From Wikipedia, the free encyclopedia
Jump to: navigation, search

Most satellites that image the Earth provide Spectral imaging. These images may be called multispectral, hyperspectral or ultraspectral because they are acquired for the same area using different wavelengths of the electromagnetic spectrum. For example Landsat satellite, Multispectral scanner - band 1 image, band 2 image etc. Since these remote sensing images are typically multispectral responses of various features it is hard to identify directly the feature type by visual inspection. Hence the remote sensing data has to be classified first, followed by processing by various data enhancement techniques so as to help the user to understand the features that are present in the image.

Such classification is a complex task which involves rigorous validation of the training samples depending on the classification algorithm used. The techniques can be grouped mainly into two types.

  • Supervised classification techniques
  • Unsupervised classification techniques

Supervised classification techniques[edit]

Supervised classification makes use of training samples. Training samples are areas on the ground for which there is Ground truth, that is, what is there is known. The spectral signatures of the training areas are used to search for similar signatures in the remaining pixels of the image, and we will classify accordingly. This type of classification which uses the training samples for classification is called supervised classification. Expert knowledge is very important in this method since the selection of the training samples and adopting a bias can badly affect the accuracy of classification. One popular technique is the Maximum Liklihood principle. In this we will calculate the probability of a pixel belonging to a class (i.e. feature) and will allot the pixel to its most probable class.

Unsupervised classification[edit]

In case of unsupervised classification no prior knowledge is required for classifying the features of the image. In this, the natural clustering or grouping of the pixel values i.e., gray levels of the pixels are observed. Then a threshold level is defined for adopting the no of classes in the image. The finer the threshold value more will be the no of classes. But beyond a certain limit same class is represented in different classes in the sense variation in the class is represented. After forming the clusters, ground truth validation is done to identify the class the image pixel belongs to. Thus in this unsupervised classification apriori information about the classes is not required. One of the popular methods in unsupervised classification is K means classifier algorithm.