Using support vector machine SVM to classify hyperspectral remote sensing data

Classification:

Support Vector Machines

material:

ENVI5.3x,Indian_pines

Indian Pines aerial hyperspectral remote sensing image (referred to as "Indian Pines data"), the data was acquired by NASA AVIRIS sensor in the agricultural area of ​​Northwest Indiana, USA on June 12, 1992.

u The spatial resolution of the data is 20m, the image size is 145×145 pixels (21025 pixels), the spectral range is 0.4~2.5 um , and it is composed of 224 bands. This area contains 16 types of land cover in agricultural areas. The total number of labeled pixels in this area is 148,152, and the number of labeled pixels on different features is uneven. The above picture shows the false-color composite image of Indian Pines data (left) and the real surface coverage type (right).

Process:

Sample making

SVM is a supervised classification method. First, samples are collected and random points are generated according to the type. Each type generates 50 random points. A total of 900 points in 16 categories are used as samples. In theory, any proportion of sample points can be generated. Some land types are labeled If the total number of samples is less than 50, theoretically, all should be selected as samples, and we will not deal with them here.

ENVI opens the hyperspectral data and sample hsapefile, and generates roi (REGION OF INTRESTING), Vector To ROI tool from the shapefile,

0-15 corresponds to 1-16

The results are as follows:

use

Use the same sample to verify the classification effect of different classification methods