OPTIMIZATION OF TRAINING DATA REQUIRED FOR NEURO-CLASSIFICATION
Xin Zhuang, Graduate Research Assistant
Agricultural Engineering Department
and
Laboratory for Applications of Remote Sensing
Purdue University
West Lafayette, IN 47907-1146
(317)494-1187
zhuang@ecn.purdue.edu
D. Fabiän Lozano-Garcia, Remote Sensing Application Manager
Laboratory for Applications of Remote Sensing
Purdue University
West Lafayette, IN 47907
Bernard A. Engel, Associate Professor
Agricultural Engineering Department
and
Laboratory for Applications of Remote Sensing
Purdue University
West Lafayette, IN 47907-1146
R. Norberto Fernändez, Manager
Global Resources Information Database, United Nations Environment Programme
Nairobi, Kenya
Chris J. Johannsen, Director
Laboratory for Applications of Remote Sensing
Purdue University
West Lafayette, IN 47907
Commission III: Mathematical Analysis of Data
ABSTRACT:
Classification of remotely sensed data with artificial neural networks is called neuro-classification. Artificial neural networks
have shown great potential in classification of remotely sensed data. The amount of data used for training a neural network
affects accuracy and efficiency of the neural network classifier. A neural network was trained separately with 5%, 10%, 15%,
and 20% image data from a LANDSAT Thematic Mapper scene, which was acquired 29 July 1987. At a risk level of 5%, the
results showed that (a) classifiers NN-5% (neuro-classification with 5% of the image data used for training), NN-10%, and
NN-15% did not differ from one another, (b) classifiers NN-15% and NN-20% did not differ from each other, but (c)
classifiers NN-5% and NN-10% differed from classifier NN-20%. The training rates were reduced by more than 10
seconds/cycle as we increased the percentage of the image data for training a neural network. Ten percent image data are
needed to adequately train a neural network classifier, the classifier provides satisfactory performance.
KEY WORDS: Neuro-Classification, Artificial Neural Networks, Image Processing.
529