CLASSIFICATION OF REMOTELY SENSED IMAGE DATA USING MULTITYPE INFORMATION

Classification of multispectral image data based on spectral information has been a common practice in the analysis of remote sensing data. However, the results produced by current classification algorithms necessarily contain residual inaccuracies and class ambiguity. By the use of other available sources of information, such as spatial, temporal, and ancillary information, it is possible to reduce this class ambiguity and in the process improve the accuracy. Therefore, the purpose of this research is to improve the accuracy of the classification by utilizing such multitype information. To accomplish this objective, three approaches are proposed. The first approach is a stochastic model in the time domain which utilizes spectral and temporal characteristics. The second approach involves the probabilistic and supervised relaxation methods which utilize multitype information. The third approach is a stochastic model in the spatial domain which attempts to extract interpixel class-conditional correlation and use this information with spectral characteristics to classify an object. As a result of adapting the above approaches to the problem, the following five new classifiers are developed. (1) Markov pixel classifier; (2) Non-iterative probabilistic relaxation; (3) Modified minimum distance object classifier; (4) Modified maximum likelihood object classifier; (5) Linear minimum distance object classifier. For all the above algorithms, software systems are developed or the existing software programs at the Laboratory for Applications of Remote Sensing (LARS), Purdue University are modified. All these methods are experimentally evaluated