H. Murai et al., PRINCIPAL COMPONENT ANALYSIS FOR REMOTELY-SENSED DATA CLASSIFIED BY KOHONENS FEATURE MAPPING PREPROCESSOR AND MULTILAYERED NEURAL-NETWORK CLASSIFIER, IEICE transactions on communications, E78B(12), 1995, pp. 1604-1610
There have been many developments on neural network research, and abil
ity of a multi-layered network for classification of multi-spectral im
age data has been studied. We can classify non-Gaussian distributed da
ta using the neural network trained by a back-propagation method (BPM)
because it is independent of noise conditions. The BPM is a supervise
d classifier, so that we can get a high classification accuracy by usi
ng the method, so long as we can choose the good training data set. Ho
wever, the multi-spectral data have many kinds of category information
in a pixel because of its pixel resolution of the sensor. The data sh
ould be separated in many clusters even if they belong to a same class
. Therefore, it is difficult to choose the good training data set whic
h extract the characteristics of the class. Up to now, the researchers
have chosen the training data set by random sampling from the input d
ata. To overcome the problem, a hybrid pattern classification system u
sing BPM and Kohonens feature mapping (KFM) has been proposed recently
. The system performed choosing the training data set from the result
of rough classification using KFM. However, how the remotely sensed da
ta had been influenced by the KFM has not been demonstrated quantitati
vely. In this paper, we propose a new approach using the competitive w
eight vectors as the training data set, because we consider that a com
petitive unit represents a small cluster of the input patterns. The ap
proach makes the training data set choice work easier than the usual o
ne, because the KFM can automatically self-organize a topological rela
tion among the target image patterns on a competitive plane. We demons
trate that the representative of the competitive units by principal co
mponent analysis (PCA). We also illustrate that the approach improves
the classification accuracy by applying it on the classification of th
e real remotely sensed data.