Neural networks are learning algorithms that are used for pattern classification, estimation and modeling complex relationships between inputs and outputs. In my work, I have used a variety of neural networks including backpropagation, ART (Adaptive Resonance Theory) models, and SOM (Self-organizing Maps) for landcover classification, change detection, estimation and other problems.

The map below shows ARTMAP classification of vegetation in the Sierras using Landsat data

The second map shows the classification differences between ARTMAP and Decision Tree classification of landcover in N. America using MODIS. Darkest areas show the highest differentiation between two classifiers while green areas represent agreement.

Collaborators: Gail Carpenter (CNS Tech Lab), Curtis Woodcock and Alan Strahler (Boston University).

See some of our papers:
1. Carpenter et al. Mixture Modeling
2. Carpenter, Gopal, Woodcock. Vegetation Mapping

3. Abuelgasim, A., Ross, W. D., Gopal, S., and Woodcock, C. E. (1999). Change detection using adaptive neural networks: Environmental damage assessment after the Gulf War, Remote Sensing of the Environment, 70 (2), 208-223.

4. Carpenter, G., Gjaja, M., Gopal, S., and Woodcock, C.  (1997). ART networks in Remote Sensing, IEEE Transactions on Geoscience and Remote Sensing, 35(2), 308-325.

5. Liu, W., Gopal, S., and Woodcock, C.  (2001). Spatial data mining for classification, visualization and interpretation with ARTMAP neural networks, in Robert L. Grossman (Editor), et al Data mining for scientific and engineering applications, Kluwer Academic Publishers, Netherlands.