A classification method using spatial information extracted by neural network

Author

Inoue, Akira ; Fukue, Kiyonari ; Shimoda, Haruhisa ; Sakata, Toshibumi

Author_Institution

Res. & Inf. Center, Tokai Univ., Tokyo, Japan

fYear

1993

fDate

18-21 Aug 1993

Firstpage

893

Abstract

A land cover classification method using a neural network is applied for the purpose of utilizing spatial information. The adopted model of the neural network has a three layered architecture, and the training method of the network is the back-propagation algorithm. Co-occurrence matrices, which are extracted from original image data, are used for the input pattern to the neural network. To evaluate the method, classification was conducted with this method for images from the Landsat TM and SPOT HRV. Obtained classification accuracies were 7-12% higher than that of the conventional pixel-wise maximum likelihood method based on spectral information

Keywords

backpropagation; environmental science computing; geophysics computing; image recognition; matrix algebra; neural nets; remote sensing; Landsat TM data; SPOT HRV data; back propagation algorithm; classification method; cooccurrence matrices; image data; input pattern; land cover; neural network; spatial information; three layered architecture; training method; Data mining; Entropy; Frequency; Heart rate variability; Neural networks; Neurons; Pixel; Remote sensing; Satellites; Statistics;

fLanguage

English

Publisher

ieee

Conference_Titel

Geoscience and Remote Sensing Symposium, 1993. IGARSS '93. Better Understanding of Earth Environment., International

Conference_Location

Tokyo

Print_ISBN

0-7803-1240-6

Type

conf

DOI

10.1109/IGARSS.1993.322178

Filename

322178