Common raster data concepts

As we mentioned earlier, remotely sensed raster data is a matrix of numbers. Remote sensing contains thousands of operations that can be performed on data. This field changes on almost a daily basis as new satellites are put into space and computer power increases.

Despite its decade-long history, we haven’t even scratched the surface of the knowledge that this field can provide to the human race. Once again, similar to the common GIS processes, this minimal list of operations allows you to evaluate any technique that’s used in remote sensing.

Band math

Band math is multidimensional array mathematics. In array math, arrays are treated as single units, which are added, subtracted, multiplied, and divided. However, in an array, the corresponding numbers in each row and column across multiple arrays are computed simultaneously. These arrays are termed matrices, and computations involving matrices are the focus of linear algebra.

Change detection

Change detection is the process of taking two images of the same location at different times and highlighting those changes. A change could be due to the addition of something on the ground, such as a new building, or the loss of a feature, such as coastal erosion. Many algorithms detect changes among images and also determine qualitative factors such as how long ago the change took place.

The following figure from a research project by the US Oak Ridge National Laboratory (ORNL) shows rainforest deforestation between 1984 and 2000 in the state of Rondonia, Brazil:

Figure 1.22 – This US Department of Energy satellite image analysis illustrates change detection by comparing the deforestation of a rainforest over time

Colors are used to show how recently the forest was cut. Green represents virgin rainforests, white represents a forest that was cut within 2 years of the end of the date range, red represents within 22 years, and the other colors fall in between, as described in the legend.

Histogram

A histogram is the statistical distribution of values in a dataset. The horizontal axis represents a unique value in a dataset, while the vertical axis represents the frequency of this unique value in the raster. The following example, which was generated from a NASA Landsat image, shows a histogram showing pixel value distributions for the first three bands:

Figure 1.23 – Histogram distribution of red, green, and blue pixels in a satellite image that can be redistributed to enhance an image for visualizing or analyzing certain features

A histogram is a key operation in most raster processing. It can be used for everything from enhancing contrast in an image to serving as a basis for object classification and image comparison.

Feature extraction

Feature extraction is the process of manually or automatically digitizing features in an image to points, lines, or polygons. This process serves as the basis for the vectorization of images, in which a raster is converted into a vector dataset. An example of feature extraction is extracting a coastline from a satellite image and saving it as a vector dataset.

If this extraction is performed over several years, you could monitor the erosion or other changes along this coastline.

Supervised and unsupervised classification

Objects on the Earth reflect different wavelengths of light, depending on the materials that they are made of. In remote sensing, analysts collect wavelength signatures for specific types of land cover (for example, concrete) and build a library for a specific area. A computer can then use this library to automatically locate classes in the library in a new image of the same area.

In unsupervised classification, a computer groups pixels with similar reflectance values in an image without any other reference information other than the histogram of the image.