Pattern classification system

• US 5,337,371 A
• Filed: 08/27/1992
• Issued: 08/09/1994
• Est. Priority Date: 08/09/1991
• Status: Expired due to Term

• Current Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Shimizu, Makoto, Sato, Masaaki, Yoshida, Kunio, Shida, Takehiko, Fujiwara, Katsuhiko, Saitoh, Mie, Yokouchi, Akira, Naka, Motohiko, Akamine, Ikuo
• Primary Examiner(s): Moore, David K.
• Assistant Examiner(s): ANDERSON, DAVID
• Application Number:
  US07/935,099
• Time in Patent Office: 712 Days
• Field of Search: 382/14, 382/15, 382/36, 382/37, 382/38, 382/39
• US Class Current: 382/100
• CPC Class Codes: F24F 11/00 Control or safety arrangements
• CPC Class Codes: G06F 18/24143 Distances to neighbourhood prototypes, e.g. restricted Coulomb energy networks [RCEN]
• CPC Class Codes: G06F 18/24317 Piecewise classification, i.e. whereby each classification requires several discriminant rules
• CPC Class Codes: G06N 3/04 Architecture, e.g. interconnection topology
• CPC Class Codes: Y10S 706/904 Manufacturing or machine, e.g. agricultural machinery, machine tool

Abstract

A pattern classification system includes a plurality of classification sections. Each of the classification sections includes a device for storing information of N coefficients W representing a reference pattern, a device for calculating an evaluation value V on the basis of N input signals S and the N coefficients W, the N input signals representing an input pattern, the evaluation value V representing a relation between the input pattern and the reference pattern, a device for storing information of a fixed threshold value R, and a device for comparing the evaluation value V and the threshold value R and for outputting an estimation signal depending on a result of the comparing, the estimation signal including a category signal P which represents a category. A selection section is operative for selecting one of categories in response to the category signals outputted from the classification sections and for outputting a signal Px representing the selected one of the categories. An adjustment section is operative for adjusting parameters in the classification sections in response to the output signal Px of the selection section and a teacher signal T, the parameters including the coefficients W.

Claims

1. A pattern classification system comprising:

a plurality of classification sections each including;

means for storing information of N coefficients W representing a reference pattern,

means for calculating an evaluation value V on the basis of N input signals S and the N coefficients W, the N input signals representing an input pattern, the evaluation value V representing a relation between the input pattern and the reference pattern,

means for storing information of a fixed threshold value R,

means for comparing the evaluation value V and the threshold value R and for outputting a first part of an estimation signal depending on a result of the comparing, the estimation signal first part including a category signal P which represents a category;

means for storing a counter signal C, representing a frequency in use of the classification section, and

means for outputting the stored counter signal as a second part of the estimation signal in response to the result of the comparing the evaluation value V and threshold value R;

a selection section for selecting one of a plurality of categories represented by the category signals P outputted from the classification sections in response to the counter signals outputted from the classification sections and for outputting a signal Px representing the selected one of the categories; and

an adjustment section for adjusting parameters in the classification sections in response to the output signal Px of the selection section and a teacher signal T representing a target state of the output signal Px of the selection section, the parameters including the coefficients W.

2. The pattern classification system of claim 1, wherein the calculating means in each of the classification sections comprises means for calculating a distance D between the input signals S and the coefficients W as the evaluation value V.

3. The pattern classification system of claim 1, wherein the calculating means in each of the classification sections comprises means for calculating a distance D between the input signals S and the coefficients W as the evaluation value V, and wherein each of the classification sections comprises means for storing the category signal P, means for outputting the stored category signal P as a part of an evaluation signal in response to the result of the comparing the evaluation value V and the threshold value R, said means for storing a counter signal C, and said means for outputting the stored counter signal as a part of the estimation signal in response to the result of the comparing the evaluation value V and the threshold value R.

4. The pattern classification system of claim 1, wherein the adjustment section comprises means for adjusting the counter signal stored in each of the classification sections.

5. The pattern classification system of one of claims 1 wherein the selection section comprises means for detecting one of the categories which contains a largest number of the category signals outputted from the classification sections, and means for using the detected one of the categories as the selected one of the categories.

6. The pattern classification system of claim 1,

wherein the selection section comprises means for selecting a maximum counter signal from among the counter signals outputted from the classification sections, means for selecting one of the category signals outputted from the classification sections which corresponds to the selected maximum counter signal, and means for generating the output signal Px in accordance with the selected one of the category signals.

7. The pattern classification system of claim 1, wherein the selection section comprises means for outputting a label signal representative of one of the classification sections which outputs the selected one of the category signals.

8. The pattern classification system of claim 1, wherein the adjustment section comprises means for deciding the output signal Px of the selection section to be true and false when the output signal Px of the section signal is equal to and different from the teacher signal T respectively, and for generating a true/false signal TF representative thereof, and true/false processing means for adjusting the parameters in the classification sections in response to the true/false signal TF.

9. The pattern classification system of claim 1, wherein the adjustment section comprises means for deciding whether the output signal Px of the selection section is true or false, and for generating a true/false signal TF representative thereof, and true/false processing means for adjusting the parameters in the classification sections in response to the true/false signal TF, and wherein the true/false processing means comprises a first memory for storing information of a counter adjustment amount DC1 suited to a true case, a second memory for storing information of a counter adjustment amount DC2 suited to a false case, means for reading out one of the counter adjustment amounts DC1 and DC2 from the first and second memories in response to the true/false signal TF, and for outputting the readout one of the counter adjustment amounts DC1 and DC2, and means for correcting a counter signal Cx into a corrected counter signal Cx'"'"' in accordance with the outputted one of the counter adjustment amounts DC1 and DC2, and for outputting the corrected counter signal Cx'"'"'.

10. The pattern classification system of claim 9, wherein the true/false processing means comprises a first memory for storing information of a coefficient adjustment amount DW1 suited to a true case, a second memory for storing information of a coefficient adjustment amount DW2 suited to a false case, means for reading out one of the coefficient adjustment amounts DW1 and DW2 from the first and second memories in response to the true/false signal TF, and for outputting the readout one of the coefficient adjustment amounts DW1 and DW2, and means for correcting the coefficients Wx into corrected coefficients Wx'"'"' in accordance with the outputted one of the coefficient adjustment amounts DW1 and DW2, and for outputting the corrected coefficients Wx'"'"'.

11. The pattern classification system of claim 1, wherein the adjustment section comprises means for deciding whether the output signal Px of the selection section is true or false, and for generating a true/false signal TF representative thereof and true/false processing means for adjusting the parameters in the classification sections in response to the true/false signal TF, and further comprising generation processing means which includes a first memory for storing information of a counter lower limit Cleast, means for outputting a label signal "x" in response to the true/false signal TF and the lower limit Cleast, the label signal "x" corresponding to a classification section having a counter signal equal to or smaller than the lower limit Cleast, a second memory for storing a counter initial value Cstart, means for reading out the initial value Cstart from the second memory in response to the true/false signal TF, the teacher signal T, and the input signals S, means for outputting the coefficients W in response to the input signals S, and means for outputting the category signal P in response to the teacher signal T.

13. The pattern classification system of claim 11, wherein each of the classification sections comprises means for rejecting reception of the input signals S when the counter signal C is equal to or smaller than the lower limit Cleast.

14. The pattern classification system of claim 11, further comprising means for initializing the counter signal C in each of the classification sections to the lower limit Cleast.

15. The pattern classification system of claim 3, wherein each of the classification sections comprises means for outputting a signal representative of the calculated distance D as a part of the evaluation signal.

16. The pattern classification system of claim 15, wherein the selection section comprises means for selecting one of distance-representing signals outputted from the classification sections, the selected distance-representing signal corresponding to a minimum distance, and means for outputting the signal Px corresponding to the selected distance-representing signal.

12. The pattern classification system of claim 12, wherein the adjustment section comprises means for selectively outputting the true/false signal to at least one of the true/false processing means and the generation processing means in accordance with contents of the true/false signal TF.