Training strategies for neural network soft classification of remotely sensed imagery
Evidence from the recent literature demonstrates little progress in discrete (hard) classification of land cover from remotely sensed imagery. An empirical study is presented to test training procedures with neural networks for soft (mixture) classification. The results show that land cover mixtures are best recognized following training with two component mixed pixels, and that linearly rescaled or binned target vector representations are equally satisfactory.
Bibliographic Reference: Article: International Journal of Remote Sensing (1996)
Record Number: 199610878 / Last updated on: 1996-09-16
Original language: en
Available languages: en