Land-use classification of remotely sensed data using Kohonen Self-Organizing Feature Map neural networks

The use of Kohonen Self-Organizing Feature Map (KSOFM, or feature map) neur al networks for land-use/land-cover classification from remotely sensed dat a is presented. Different from the traditional multi-layer neural networks, the KSOFM is a two-layer network that creates class representation by self -organizing the connection weights from the input patterns to the output la yer. A test of the algorithm is conducted by classifying a Landsat Thematic Mapper (TM) scene for seven land-use/land-cover types, benchmarked with th e maximum-likelihood method and the Back Propagation (BP) network. The netw ork outperformes the maximum-likelihood method for per-pixel classification when four spectral bands are used. A further increase in classification ac curacy is achieved when neighborhood pixels are incorporated. A similar acc uracy is obtained using the sp networks for classifications both with and w ithout neighborhood information. The feature map network has the advantage of faster learning but has the drawback of being a slow classification proc ess. Learning by the feature map is affected by a number of factors such as the network size, the codebooks partitioning, the available training sampl es, and the selection of the learning rate. The feature map size controls t he accuracy at which class borders are formed, and a large mop may be used to obtain accurate class representation. If is concluded that the feature m ap method is a viable alternative for land-use classification of remotely s ensed data.