image classification

Remotely-sensed imagery can be difficult to interpret for a number of reasons. The ultimate goal in remote sensing is to derive information from the raw image data - that is, we want to identify, or classify, what we see in the image.

Because identifying each pixel by hand (or even large groups of pixels) is tedious and time-consuming, we want to try to avoid this as much as possible, at least on a large scale. For that, we turn to (semi) automated classification, so that we can get the computer to do as much of the work for us as possible.

In general, we can characterize classifications in a few different ways; for these tutorials, we focus on three:

• unsupervised methods, where we have nearly no input into how the computer classifies pixels;

• (supervised) pixel-based methods, where individual pixels are classified based on input data;

• (supervised) object-based, where pixels are grouped together before being classified based on input data.

• unsupervised classification
  • selecting sample points
  • training a clusterer
  • applying the classification
  • viewing the clusters
  • stratified sampling
  • creating a scatter plot
  • plotting spectral signatures
  • next steps
  • references and notes
• pixel-based classification
  • adding training points
  • adding classes
  • using require to load another script
  • getting training data
  • splitting the training data
  • training a classifier
  • classifying the image
  • accuracy assessment
  • spectral signatures
  • next steps
  • references and notes
• object-based classification
  • image segmentation
  • object texture and contrast
  • next steps
  • notes and references