Unsupervised & Supervised Classification

One of the most common uses of remotely sensed imagery is to determine the extent and distribution of various land uses and land cover classes across a particular area. There are two broad categories of digital image classification -- unsupervised and supervised. Unsupervised classification relies on clustering algorithms to determine which land cover type each pixel represents, while supervised classification uses carefully selected training sites to guide the computer's classification. 

In this weeks lab, we experimented with both unsupervised and supervised classification techniques. Below is a land cover map of Germantown, MD, that was created using supervised classification within ERDAS Imagine. Training sites were generated using the provided lat/longs for various land cover types including agriculture, fallow fields, water, urban areas, deciduous and mixed forests, and grass. The smaller inset map shows the Output Distance file. Brighter pixels in this image indicate areas that are likely to be misclassified and require additional training sites.