Concurrent Self-Organizing Maps for Classification of Multispectral Satellite Imagery

摘要

We present a new neural classification model called Concurrent Self-Organizing Maps (CSOM), representing a winner-takes-all collection of small Kohonen networks. Each SOM of the system is trained individually to provide best results for one class only. We have applied the proposed CSOM model for multispectral satellite image classification. We have experimented the CSOM for classification of multispectral pixels belonging to a LANDSAT TM image with 7 bands. The image is composed by a set of 7-dimensional pixels, out of which a subset contains labeled pixels that have been classified by an expert into seven thematic categories. For the test lot, the experimental results lead to the recognition rate of 94.31 % using a circular SOM with 784 neurons, while we have obtained a recognition rate of 95.29 % using a CSOM containing 7 circular SOMs with 112 neurons each. Simultaneously, CSOM leads to a significant reduction of training time by comparison to SOM.