Neural Network Processing of Satellite Imagery
In this study a neural computing approach is taken for the classification and land feature extraction of satellite imagery. The satellite data consists of multispectral Thematic Mapper recordings. The applied neural network is a three layer feedforward network with one hidden layer using the backpropagation learning paradigm. Several network configuration concerning data scaling and activation functions are examined. For a comparative study and validation of the neural network model, the maximum likelihood classifier and the k nearest neighbor classifier are used as statistical conventional classification techniques. The neural network is found to be superior in terms of overall performance to the conventional techniques. An accuracy assessment analyzes the classification results and gives a detailed picture about the advantages and weaknesses of each method.
Thesis96.E53.pdf
11.94 MB
Adobe PDF
fca35678b3f64b9fd1c4cb1da35aec50