Learning type waveform recognizer

• US 5,497,448 A
• Filed: 03/23/1995
• Issued: 03/05/1996
• Est. Priority Date: 04/30/1992
• Status: Expired due to Fees

• Current Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Original Assignee: Matsushita Electric Industrial Company Limited (Panasonic Holdings Corporation)
• Inventors: Sakaue, Shigeo, Maruno, Susumu, Shimeki, Yasuharu
• Primary Examiner(s): MacDonald, Allen R.
• Assistant Examiner(s): Dorvil, Richemond
• Application Number:
  US08/409,947
• Time in Patent Office: 348 Days
• Field of Search: 395/22, 395/23, 395/24, 395/2-41, 382/165, 382/170, 382/181
• US Class Current: 706/20
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 2218/08 Feature extraction

Abstract

The invention is a learning type waveform recognizer that has a sequential signal extractor that extracts changes in the values of an input waveform signal at each point on the transition axis, a learning and recognition section that learns the relations between the changes extracted from the input signal and classified waveform signals and recognizes the degree of matching with the classified waveform at each point on the transition axis based on the learned relations, and an integrated recognition section that integrates the recognized results at points in an interval and performs an overall judgment by identifying the waveform of the input waveform signal.

Claims

1. A learning type waveform recognizer for recognizing a waveform of a signal comprising:

a sequential signal extractor for receiving input terminals signal and for extracting sequential changes of said input signal at each point on a transition axis;

a learning and recognition section, including an input connected to said sequential signal extractor for receiving extracted sequential changes of said input signal, and a plurality of output terminals, wherein each of said plurality of output terminals corresponds to each one of a plurality of labelled known waveforms, said learning and recognition section performing a learning operation to obtain a relationship between an output terminal and a known waveform by inputting each one of a plurality of labelled known waveforms to said learning and recognition section via said sequential signal extractor, obtaining a signal at an output terminal indicative of the maximum degree of coincidence between an inputted known waveform and an output terminal, and labelling each output terminal to correspond to each inputted known waveform, said learning and recognition section performing a recognition operation by inputting an unknown signal via said sequential signal extractor, and outputting signals from said output terminals indicative of a degree of coincidence between said unknown signal and the labelled output terminals based on said relationship between an output terminal and a known waveform; and

an integrated recognition section for integrating each of said signals output from said plurality of output terminals along a transition axis and recognizing a waveform of said unknown signal based on values obtained by integration.

2. The learning type waveform recognizer according to claim 1, wherein said learning and recognition section comprises:

a network having a plurality of layers including at least first and second layers comprising a plurality of first and second recognition units interlinked to form said network, respectively,

each of said plurality of first recognition units comprising;

a signal inputting unit for receiving a sequential signal extracted by said extractor,

a quantizer, coupled to said signal inputting unit, for quantizing said sequential signal received by said signal inputting unit, and

a path selector unit, coupled to said quantizer and having at least one path input terminal and at least one path output terminal, wherein paths are selected according to an output of said quantizer, and

each of said plurality of second recognition units comprising;

at least one path input terminal,

at least one path output terminal linked to said at least one path input terminal, wherein said linkage has a varying strength, and

a path learning unit, for receiving a teaching signal and, according to said teaching signal, modifying the linkage strength between said at least one path input terminal and said at least one path output terminal.

3. The learning type waveform recognizer according to claim 1, wherein said sequential signal extractor comprises:

a current signal extractor for extracting a value of a signal at each point on a transition axis, and

a change amount extractor for calculating each difference between a current value extracted by said current signal extractor and a value at a respective point on the transition axis.

4. The learning type waveform recognizer according to claim 3, wherein said change amount extractor comprises:

a signal memory for successively storing values at points within a predetermined width on the transition axis extracted by said current signal extractor, and

a difference calculator for calculating differences between the current value and previous values stored in said signal memory.

5. The learning type waveform recognizer according to claim 1, wherein said integrated recognition section comprises:

a local recognition result memory for successively storing a predetermined number of results output from said learning and recognition section with respect to a predetermined point on the transition axis,

an integration number setting section for setting a number of local recognition results to be integrated, and

an arithmetic integration section for integrating local recognition results stored in said local recognition result memory and outputting integrated results.

6. The learning type waveform recognizer according to claim 5, wherein said arithmetic integration section comprises:

a plurality of integration weights, each corresponding to one of the local recognition results stored in said local recognition result memory, and

a sum-of-products calculator for calculating a sum of products of said local recognition results and said corresponding weights.

7. The learning type waveform recognizer according to claim 5, wherein said integration number setting section comprises:

an integrating weight setter for setting a plurality of integrating weights, each corresponding to one of the local recognition results stored in said local recognition results memory.

8. The learning type waveform recognizer according to claim 1, wherein said integrated recognition section comprises a plurality of integration units, wherein

the number of said integration units are equal to the number of output terminals of said learning and recognition section,

each integration unit being exclusively connected to one of the output terminals of said learning and recognition section,

each integration unit comprising an input terminal for reading an output signal of said learning and recognition section, a feedback terminal, two weights respectively corresponding to said input terminal and said feedback terminal, a sum-of-products calculator for calculating a sum of a product of a value input from said input terminal and a corresponding weight and a product of a value input from said feedback terminal and a corresponding weight, a threshold processor for performing threshold operation to an output of said sum-of-product calculator, and an attenuator for attenuating the output of said sum-of-product calculator with a predetermined time constant,

the output of said sum-of-product calculator being applied to said feedback terminal through said attenuator, and

the output of said sum-of-product calculator also being output from said integrated recognition section through said threshold processor.