Urban land use mapping with multi-spectral and SAR satellite data using neural networks
Statistical, textural and Gabor features were extracted from integrated multi-temporal, multi-spectral thematic mapper (TM) data and ERS-1 synthetic aperature radar (SAR) imagery for urban land use mapping. These features extract structure and context information from the imagery and therefore may be suitable to recognize complex land use classes. The computed features are first normalized using the Self Organising Map algorithm and then a decision tree algorithm is applied to select the most appropriate features for subsequent processing. The classification procedure was carried out with a multi-layer perceptron, trained with the resilient backpropagation algorithm, with weight decay regularization. Results demonstrate the potential of the proposed methodology.
Bibliographic Reference: Paper presented: IGARSS '97 (SG), August 5-8, 1997
Availability: Available from (1) as Paper EN 40661 ORA
Record Number: 199710867 / Last updated on: 1997-07-08
Original language: en
Available languages: en