EVALUATION OF OBSERVATION PARAMETERS BY REDUCTION METHOD

Abstract

A set of parameters of observed sub-objects forms a set of points in the multidimensional space of the indicated parameters. The task of recognition in the given conditions is reduced to partitioning this space into solution regions corresponding to characteristics of individual sub-objects. A possible method for resolving this problem is preliminary clustering of the set of points at the learning stage without a teacher and then fixing the boundaries of the regions. The system for recognizing complex objects is implemented in the form of a learning neural network which alphabet corresponds to the aforesaid solution regions, and complex objects are the words of this alphabet. The formation of these words is preceded by geo-referencing of sub-objects, which makes it possible to identify them as constituents of the object and the formation is performed by triangulation method. It is proposed to evaluate the parameters of sub-objects using the reduction method based on the measurement model, which is based on the basic relationship of the passive location. Adjustment of the model parameters interpreted as parameters of the observed signal is performed with the use of statistical method or by straightforward enumeration. Thus, the reduction method is a model-oriented method for evaluation of parameters of the individual emitter obtained from the observation of a signal received from it. Triangulation is used for geographic referencing of the object, and a scale-invariant converter is implemented for transforming a spectrographic signal into a spectrometric signal.

Keywords

• recognition
• parametric space
• splitting
• solution regions
• clustering
• neural network
• learning without a teacher
• learning with a teacher
• reduction method
• statistical search
• straightforward enumeration
• spectrograph
• spectrometer
• scale invariance