what's the difference between spectral signature and features in remote sensing?

Question

I would like to understand if is there any difference or relation between spectral signature concept and features in remote sensing?

Answer 1

The term spectral signature refers to the relationship between the wavelength (or frequency) of electromagnetic radiation and the reflectance of the surface. The signature is affected by several things including the material composition and structure. Some parts of the EMR spectrum, such as the microwave region, are more sensitive to surface structure than other regions. We use the spectral signature (or more often sampled parts of it--bands of satellite imagery) to infer things about the surface such as composition (e.g. vegetation, bare soil, etc.).

A feature on the other hand is simply an object in landscape. For example, a feature may be a field of uniform crop, a road, or building, or any other part of the landscape. We often try to identify features by using their spectral signatures, assuming uniformity, which is not always the case. Sometimes, rather than classifying pixels based on their spectral signatures alone, we also try to account for spatial relations such as the proximity of similar pixels. This is common for example with object-based image segmentation, which attempts to identify features using a combination of spectral and spatial characteristics.

Answer 2

A spectral signature is some measurable quantity (e.g., reflectivity, emissivity), which varies as a function of wavelength and can be used to identify a material. To obtain a signature, the quantity must be measured at a sufficient number of wavelengths (and at fine enough spectral resolution) such that the material can be discriminated from other materials. For example, an RGB image (converted to reflectance), provides reflectivity information at three wavelengths (red, green, & blue); however, that typically wouldn't be considered a spectral signature because it doesn't provide adequate information to discriminate various materials (e.g., pixels containing grass, artificial turf, or green tennis court can look nearly identical in RGB imagery). Spectral signatures are typically obtained either from hyperspectral images or with handheld spectrometers.

The term "feature" can have multiple meanings. While it can refer to a spatial characteristic (or object), in the spectral domain it usually means something quite different. A spectral feature could be the original spectral measurement data (e.g., reflectivity) but it is often something derived from the spectral measurements, typically by creating linear or nonlinear transformations of the original data values. Spectral features are often created to reduce the dimensionality of the spectral data prior to further processing. Examples of linear features are those obtained from Principal Components Analysis (PCA) or Linear Discriminant Analysis (LDA). An example of a nonlinear feature is the Normalized Difference Vegetation Index (NDVI), which is a scaled difference between the red and near-infrared bands in an image pixel.

There are numerous types of spectral features that can be extracted from spectral data and the best one depends on the details of what you are trying to accomplish. This is an area of active research. To learn more about some the techniques and applications, I recommend starting with a web search for "spectral feature extraction".