Supervised classification can be performed using either image-derived or field-measured spectral signatures. These spectral signatures are selected to represent the various cover classes present in an image. However chosen, these spectral signatures are used to assign the associated cover classes on a pixel-by-pixel basis. The two most common classification algorithms are maximum likelihood and minimum-distance classification.
In many cases, the spectral signatures used for supervised classification of a remotely sensed image are derived from a spectral library. To maximize the success of this approach, the spectral signatures contained in the spectral library must be representative of materials on the ground at the time of overflight. This is particularly true of vegetation since the spectral signature of vegetation changes continually during the growing season. The most common approach is to have a FieldSpec 4 spectroradiometer in the field at the time of overflight. This allows for both the collection of field spectral signatures that will best match the imagery and measurements necessary for conversion of the image to reflectance.
When image derived spectral signatures are used as their primary source of information for the classification, field collected reflectance spectral libraries are often used to define the relationship between the various images classes and the spectral signatures of known materials on the ground. Image derived spectral signatures are a mixture of the spectral signatures of the materials present in a given image pixel. Spectral un-mixing techniques are often used in conjunction with a spectral library of "pure" end-member spectra to understand these mixed image signatures. As in the previous example, the spectral signatures contained in the spectral library must be representative of materials on the ground at the time of overflight. Using a FieldSpec 4 spectroradiometer in the field at the time of overflight allows for the measurement of materials at the sub-pixel scale that contribute to the overall mixed spectral signature observed in the collected imagery.
The spatial scale of reflectance spectra collected in the field depends of the type of analysis performed. When using field collected reflectance spectra directly for classification, it is important that the spectra are collected at a spatial scale similar to image pixel. This is most easily achieved by performing a collection of a large number of individual measurements from each cover class. The average of these measurements is then used to represent the spectral signature of a given cover class. The measurements for each cover class are often collected on a grid pattern laid out in advance. When the field collected reflectance spectra are to be used to un-mix observed image spectral signatures, field reflectance spectra should be collected of each distinct surface type that contributes to the average spectral signatures observed in the image.
The person operating the FieldSpec 4 spectroradiometer should always be positioned such that their shadow is furthest from the collection point since this minimizes errors from light scattered off the instrument operator. Care should be taken to avoid disturbing the surface prior the the collection of spectra using the FieldSpec 4 spectroradiometer. This is best achieved by collecting spectra on a pre-determined grid where the collection of the portion of the grid furthest from the sun is collected first. That way the operator will always be walking over areas that have already been measured.
Contact us to learn more about image classification.