Classification of land cover using optimized neural nets on SPOT data
- Jutland Telephone, Operations Research Group, Aarhus (Denmark)
An optimized neural net was developed for land-cover classification in a multispectral SPOT satellite image covering 10 km x 10 km region which contains a mixture of densely built-up areas, suburbs, rural land, and waterbodies. In the technique, segments in the image are described by textural features calculated from gray-level difference statistics. The size of the input layer (i.e., the input variables to be used), as well as the size of the hidden layer in the neural net are determined using the optimization algorithm proposed by Mozer and Smolensky (1989). The textural features are calculated in segments generated by region growing in an image which has been processed iteratively with an edge enhancing adaptive filter. 15 refs.
- OSTI ID:
- 6209226
- Journal Information:
- Photogrammetric Engineering and Remote Sensing; (United States), Vol. 59:5; ISSN 0099-1112
- Country of Publication:
- United States
- Language:
- English
Similar Records
Selection of optimal textural features for maximum likelihood image classification
Change detection and classification of land cover in multispectral satellite imagery using clustering of sparse approximations (CoSA) over learned feature dictionaries