Multi-resolution hidden markov model using class specific features
AssigneeThe United States of America as represented by the Secretary of the Navy
InventorPaul M. Baggenstoss
Life of this patent
Estimating the term…
Abstract
A method for classifying data includes selecting an elemental size and features for the data that are representative of possible subclasses. Resolution widths are selected in conjunction with these features. Models associated with symbols are developed from these resolution widths and features. Data is compared with these models to give a likelihood that the model applies. The best model is determined and a signal is provided related to the symbol associated with the best model.
Claims20
A method for classifying data in a signal processing system by determining a best model for classifying data comprising:
- receiving acoustic data from a one dimensional data source;
- selecting an elemental segment size as a finest subdivision of the data;
- selecting features that are representative of possible subclasses of the data in a signal processing system;
- selecting resolution widths for analysis windows based on selected features as multiples of the elemental segment size;
- developing models based on all selected subclasses and resolution widths, each model being associated with a particular symbol;
- comparing the data with each model to give a likelihood that the data fall within the developed model;
- determining the best model from the developed models for the data from a largest likelihood from the step of comparing;
- and
- providing a signal related to the symbol associated with the determined best model for the data.
A method for classifying data by determining a best model for classifying acoustic data comprising:
- receiving data from an acoustic data source;
- selecting an elemental segment size as a finest subdivision of the acoustic data;
- selecting features that are representative of possible subclasses of the acoustic data;
- selecting resolution widths for analysis windows based on selected features as multiples of the elemental segment size;
- developing models in a signal processing system based on all selected subclasses and resolution widths, each model being associated with a particular symbol, said developed models being for classes of acoustic signals;
- comparing the acoustic data with each model to give a likelihood that the acoustic data fall within the developed model;
- determining the best model from the developed models for the acoustic data from a largest likelihood from the step of comparing;
- and
- providing a signal related to the symbol associated with the determined best model for the acoustic data.
Dependencies are inferred from the claim text.