Search for a command to run...
We present a software package for the supervised classification of images useful for cover-type mapping of freshwater habitat (e.g., water surface, gravel bars, vegetation). The software allows the user to select a representative subset of pixels within a specific area of interest in the image that the user has identified as a cover-type habitat of interest. We developed a graphical user interface (GUI) that allows the user to select single pixels using a dot, line, or group of pixels within a defined polygon that appears to the user to have a spectral similarity. Histogram plots for each band of the selected ground-truth subset aid the user in determining whether to accept or reject it as input data for the classification processes. A statistical model, or classifier, is then built using this pixel subset to assign every pixel in the image to a best-fit group based on reflectance or spectral similarity. Ideally, a classifier incorporates both spectral and spatial information. In our software, we implement quadratic discriminant analysis (QDA) for spectral classification and choose three spatial methods—mode filtering, probability label relaxation, and Markov random fields—to incorporate spatial context after computation of the spectral type. This multi-step interactive process makes the software quantitatively robust, broadly applicable, and easily usable for cover-type mapping of rivers, their floodplains, wetlands often components of these functionally linked freshwater systems. Indeed, this supervised classification approach is helpful for a wide range of cover-type mapping applications in freshwater systems but also estuarine and coastal systems as well. However, it can also aid many other applications, specifically for automatic and quantitative extraction of pixels that represent the water surface area of rivers and floodplains.