Designing a rule-based classifier using syntactical approach

Abstract

In this research, a rule-based system has been developed for pattern recognition, making use of both statistical and syntactical approaches in classifying remote sensing data. This system attempts to automate the process of recognizing the output patterns from the unsupervised classifier based on a developed syntactic model. In the first phase of this work, a library of statistics of spectral characteristics of different land-use features in six bands of Thematic Mapper (TM) data (bands: 1, 2, 3, 4, 5 and 7) has been constructed. The statistical library is based on a prior knowledge (ground data) of the concerned pattern features, namely: maize, cotton, trees, soil, urban, water and salty water. The second phase is designing a grammar rule tree for recognizing each pattern, based on the outcome of this library, using waveform peak method. The developed system has a builtin intelligence so as to accept the unlabelled output spectral classes from the unsupervised classifier, parsing the pattern for each, and outputting descriptive classes.