Principle of least distance classifier

The minimum distance classification is one of the methods for supervised classification. Its specific classification process is as follows:

(1) Calculate the mean vector and standard deviation (mean variance) vector of each class using training sample data;

(2) Take the mean vector as the center position of the class in the feature space, and calculate the distance from each pixel in the input image to various centers. There are two simple distance functions that are most widely used in remote sensing image classification:

Euclidean distance and absolute distance
.

(3) According to the calculated distance, the image element is included in the class with the smallest distance.

Due to the direct use of Euclidean distance and absolute distance for classification, there are obvious defects:

First, the variation range of gray values of different categories is that the variance size is different, and the distance from the pixel to the class center cannot be used to divide the attribution of the pixel. Second, the distribution of point groups of natural terrain types is not necessarily circular or spherical, that is, the radius is different in different directions, so the distance measurement should be different in different directions. Considering the above factors, we can improve the distance classification as follows to improve the classification accuracy:

Euclidean distance

Absolute Distance

WhereIThe standard deviation of the band. You can also use
To replace
Or use other weighting methods.

The minimum distance method is used to classify images. Its accuracy depends on the understanding of known object categories and the accuracy of training statistics. In general, this classification method is relatively effective and easy to calculate, and supports sequential pixel scanning and classification.