Image classification is interpreting and labeling pixels within an image based on precise rules. The pixel delineation law can be devised using one or more spectral or textural characteristics.
A spectral response pattern represents a graph that exhibits the degree/intensity of reflected energy in different spectrum regions. It’s a familiar perception of the human color concept. Thus, spectral response pattern is a “signature” of individual earth surface features.
Hyperspectral images have rich spectral information. The classification technique delineated based on spectral information, the pixel is used to extract spectral pattern or to obtain certain specific features from spectral response through feature extraction to classify. Every single pixel is interpreted and identified spectrally.
A pure pixel is a pixel in the image having one material/feature associated. Images comprise mixed pixels due to low spatial resolution. Mixels are pixels containing more than one distinct material called endmembers. Often a pure pixel is overshadowed by the near pixel as shown in the figure. The return of reflected energy/radiation is from the top of the atmosphere, thus a tangled or an overshadowed pixel cannot be considered as a pure pixel. Pixel Purity Index (PPI) helps to identify the most spectrally pure pixels in multispectral and hyperspectral images.
The effectiveness of the hyperspectral image lies in their ability to compare a pixel spectrum with the spectra of known pure. This makes the classification accurate and precise.