Segmentation and classification in SAR imagery

The need for segmentation as an image analysis tool for agriculture and forestry is demonstrated and related to classification strategies. Methods to compare segmentation algorithms are described and shown to suggest that segmentation based on global maximum a posteriori (MAP) recruitment provides the best approach currently. Segmentation of multi-channel imagery as a single entity is shown to confer advantages over combined single channel methods. Attempts to classify and segment tropical forest regions for forest discrimination are seen to be closely related to methods for change detection. The MAP classifier for multi-channel data is displayed; currently this does not yield good results, and more ad hoc methods appear superior, perhaps due to inadequate descriptions of the class identifier.