System and method for identifying an object

• US 20040054499A1
• Filed: 02/19/2003
• Published: 03/18/2004
• Est. Priority Date: 07/21/2000
• Status: Active Grant

• Current Assignee: Ohio University
• Original Assignee: Ohio University
• Inventors: Starzyk, Janusz A., Nelson, Dale E.
• Granted Patent: US 7,035,754 B2
• Time in Patent Office: Days
• US Class Current: 702/182
• CPC Class Codes: G06F 18/21 Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation

Abstract

A system and method for determining the classification of a signal, or the identification of an object is provided. Based on rough set theory, or data mining, a training data set (105) is partitioned (105) and labeled (125) with a multi-class entropy method. Reducts (145) are calculated from a sub-set of the best-performing columns (130) of the partitioned and labeled training set data. These reducts are applied to test signals and combined for each signal classification. The present system and method produces a more accurate, robust and efficient classification result.

Claims

1. A system for identifying an object wherein said object is represented by a signal, comprising:

a computer system;

a training information data set, said training set in the form of a table having a plurality of rows and a plurality of columns, wherein each row represents a signal and each column represents attributes associated with each given signal;

a labeler;

a reduct calculator;

a testing information data set; and

a reduct classification fuser.

2. The system of claim 1 further comprising normalization logic for normalizing said signals in said training information data set.

3. The system of claim 2 further comprising a partitioner.

4. The system of claim 3 further comprising a wavelet transformer.

5. The system of claim 1 further comprising a column selector.

6. The system of claim 5 further comprising duplicate and ambiguous signal removal logic.

7. The system of claim 6 further comprising a core calculator.

8. The system of claim 1 further comprising a core calculator.

9. The system of claim 1 wherein the labeler binary labels the training information data set and wherein the labeler further labels the test information data set.

10. The system of claim 6, wherein the computer system tests the performance of the full training set and wherein the computer system automatically identifies the signals in the test set.

11. The system of claim 9 wherein the labeler includes fuzz factor logic.

12. A system for classifying a signal, comprising:

a computer system;

a training data set;

a testing data set; and

a reduct classification fuser, wherein each of the data sets are in the form of tables of data each having a plurality of rows and columns, each of the rows representing a signal.

13. The system of claim 12 further comprising a core calculator and reduct calculator.

14. The system of claim 13 further comprising a labeler.

15. The system of claim 12 further comprising normalization logic and a partitioner.

16. The system of claim 15 further comprising a wavelet transformer.

17. The system of claim 12 further comprising a user interface and a network.

18. The system of claim 14 wherein the labeler includes fuzz factor logic.

19. The system of claim 14 further comprising a column selector and duplicate and ambiguous signal removal logic.

20. A data-mining method of determining what classification a signal belongs to, the method comprising the steps of:

providing a training information set in the form of a table of data, the table having a plurality of columns and rows, wherein each column of the table represents an attribute of the classification and wherein each row of the table is a signal and represents all the attributes associated with a specific classification;

binary labeling the signals;

selecting a subset of the plurality of columns, each column in the subset having a higher information index than any of the remaining columns in the plurality of columns that were not selected;

calculating each of the reducts;

providing a set of test signals in the form of a table with a plurality of columns of attributes and rows of test signals;

determining a separate reduct classification of each of a plurality of test signals using each of the reducts; and

determining a final classification of each test signal of the plurality of test signals by combining each of the separate reduct classifications for each of the test signals.

21. The method of claim 20 further comprising the step of binary labeling each of the test signals.

22. The method of claim 21 wherein the step of binary labeling each of the test signals comprises the substeps of using a training threshold value determined for each column in the training set and establishing a threshold value tolerance bond about each threshold value such that a column value falling within a respective threshold value tolerance band will not be considered in classifying that respective signal.

23. The method of claim 20 further comprising the step of normalizing the training signals.

24. The method of claim 20 further comprising the step of partitioning the training information set into a plurality of partitions of columns.

25. The method of claim 24 wherein the partitioning step is done via a block partitioning method whereby pluralities of adjacent columns are grouped into the partitions.

26. The method of claim 24 wherein the partitioning step is done via an interleave partitioning method whereby pluralities of non-adjacent columns are grouped into the partitions.

27. The method of claim 24 further comprising the step of:

wavelet transforming each of the signals in each of the partitions.

28. The method of claim 20 further comprising the step of removing duplicate and ambiguous signals from the training information set.

29. The method of claim 20 further comprising the step of calculating the core.

30. The method of claim 20 further comprising the step of test performing the classification of each signal in the training information set using each of the reducts.

31. The method of claim 21 further comprising the step of normalizing each of the test signals prior to accomplishing the step of binary labeling each of the test signals.

32. The method of claim 20 further comprising the step of wavelet transforming each of the test signals after they are normalized and before they are binary labeled.