What are the features that are present in a multi-spectral image which can be used for classification?

Also, how do we get these values from the multi-spectral image?

I am using Python environment.

closed as too broad by PolyGeo Jun 22 '18 at 7:36

Please edit the question to limit it to a specific problem with enough detail to identify an adequate answer. Avoid asking multiple distinct questions at once. See the How to Ask page for help clarifying this question. If this question can be reworded to fit the rules in the help center, please edit the question.

  • 1
    welcome to GIS SE. There are at least two different questions in this question. Please edit your question to be more focused (and ask several questions if needed.) Please give more detail about what you mean by "extract" – radouxju Jun 22 '18 at 5:35
  • please move the "how to get" in another question, and specify what tools you are using. This question can be trivial if you use an image processing software, and complex if you are in a specific environment with a software that is not designed to manipulate images. – radouxju Jun 22 '18 at 7:02
  • Welcome to GIS SE! As a new user be sure to take the Tour where you will see that there should be only one question asked per question under our focussed Q&A format. – PolyGeo Jun 22 '18 at 7:36

Each pixel of a multispectral image is a vector of values related to the reflectance of the surface observed by the sensor (the exact contribution of the surface to the measurement is defined by the point spread function of the sensor). Multispectral sensor are typically sensitive to several portions of the spectrum ranging from blue to near infrared ( 300 to 1000 nm of wavelength. Some are also capable to measure in the SWIR (short wave infrared) or the TIR (thermal infrared). So you have between 3 and around 15 components per pixel, all of which can be used for specific purposes.

All those values in the vector are distributed as integer values in the raw files. Those values can be calibrated into reflectance values (poportion of the light that is reflected by a surface) and correction for the atmosphere effects as well as the topography can be applied to further standardise the values. Those steps are however not always necessary if you want to classify a single image, but they are recommended (especially if there is haze on a portion of your image, or in case of topographic effect).

In addition to those primary components, the context around each individual pixel can be used to improve the classification, so some algorithms are basedon the neighborhood of the pixel and not the pixel alone.

  • Is there a universal formula for calculating the reflectance value or it differs between sensors? If it differs, then where do we get the formula? – Rhea Jun 22 '18 at 7:06
  • it depends on the sensor and on the preprocessing level. For instance, you can download Sentinel2 image L1C product, which are already calibrated, and then you just need to divide by 10000 in order to get the reflectance. In Europe, there is also L2A product, which include the atmospheric corrections. With raw data, you must first apply a linear conversion of DN to luminance, and then there is a "universal" formula – radouxju Jun 22 '18 at 7:36
  • if your question is "how do I convert from raw data to reflectance", please specify which sensor you are using. And use "radiometric calibration" as your serach keywords. But as I said, calibration is only necessary if you work on several images. – radouxju Jun 22 '18 at 7:39

Not the answer you're looking for? Browse other questions tagged or ask your own question.