Adaptive filtering neural network classifier

• US 5,761,383 A
• Filed: 06/03/1997
• Issued: 06/02/1998
• Est. Priority Date: 04/27/1995
• Status: Expired due to Term

• Current Assignee: Northrop Grumman Systems Corporation (Northrop Grumman Corporation)
• Original Assignee: Northrop Grumman Corporation
• Inventors: Engel, Stephen J., Buckland, Dennis
• Primary Examiner(s): Downs, Robert W.
• Assistant Examiner(s): Shah, Sanjiv
• Application Number:
  US08/867,719
• Time in Patent Office: 364 Days
• Field of Search: 395/21, 395/22, 395/23, 382/190
• US Class Current: 706/14
• CPC Class Codes: G06F 18/00 Pattern recognition
• CPC Class Codes: G06F 2218/08 Feature extraction

Abstract

An adaptive filtering neural network classifier for classifying input signals, includes a neural network and one or more adaptive filters for receiving input analog signals to be classified and generates inputs for the classifier. Each adaptive filter is characterized as having a predetermined number of operating parameters. An analog to digital converter converts each input signal into a digital signal before input to the neural network. The neural network processes each digital signal to generate therefrom a plurality of weighted output signals in accordance with the type of network implemented. One of the weighted output signals represents a class for the input signal, and an error signal representing a difference between the weighted output signals and a predetermined desired output is also generated by the network. A control device responsive to the error signal generates a further set of operating filter parameters for input to each of the adaptive filters to change the operating response thereof and minimize the error signal.

Claims

1. An adaptive neural network classifier for classifying an input analog signal comprising:

one or more programmable filter means for receiving said input analog signal to be classified and generating filter output signals, each said programmable filter means being characterized as having a predetermined number of operating parameters;

an analog to digital converter means for converting each of said filter output signals from said one or more programmable filter means into a digital signal;

adaptive neural network means for receiving and processing each said digital signal to generate therefrom a plurality of weighted output signals in accordance with a predetermined processing structure of said adaptive neural network means, one of said weighted output signals representing a class for said input analog signal, said neural network means further generating an error signal representing a difference between said one of said weighted output signals and a predetermined desired class; and

control means responsive to said error signal for generating a further set of operating filter parameters for input to each said programmable filter means to change the operating response thereof to minimize said error signal.

2. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 1, wherein said adaptive filters are configured as bandpass filters having operating parameters including center frequency f₀ and quality factor Q, said operating parameters initially predetermined based on said input signal to be classified.

3. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 1, wherein said operating parameters are predetermined in accordance with a training technique performed on said neural network classifier.

4. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 3, wherein said training technique involves inputting predetermined known input signals and desired output signals into said neural network classifier and for generating a plurality of weighted output signals.

5. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 4, wherein said operating filter parameters are generated in accordance with said plurality of weighted signals generated by said training technique.

6. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 1, wherein said one of said weighted output signals represents an unknown class for said input signal, said neural network means further clustering like weighted output signals to form a class thereof.

7. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 1, wherein said neural network means has a weighted distribution and adjusts said weighted distribution to minimize said error signal.

8. An adaptive neural network classifier for classifying an input analog signal comprising:

one or more programmable filter means for receiving said input analog signal to be classified and generating a filter output signals, each said programmable filter means being characterized as having a predetermined number of operating parameters;

adaptive neural network means for receiving and processing each said digital output signal to generate therefrom a plurality of weighted output signals in accordance with a predetermined processing structure of said adaptive neural network means, one of said weighted output signals representing a class for said input analog signal, said neural network means further generating an error signal representing a difference between said one of said weighted output signals and a predetermined desired class; and

control means responsive to said error signal for generating a further set of operating filter parameters for input to each said programmable filter means to change the operating response thereof to minimize said error signal.

9. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 8, further including an analog to digital converter means for converting each said filter output signals from said one or more adaptive filter means into a digital signal, wherein said neural network means receives each digital signal for processing.

10. The adaptive neural network classifier for classifying an input analog signal as claimed in claim 8, wherein said neural network means has a weighted distribution and adjusts said weighted distribution to minimize said error signal.

11. A method for classifying signals with a neural network classifier, said method comprising the steps of:

inputting an analog signal to be classified into one or more programmable filter means for generating filter output signals for input to an adaptive neural network means, each said programmable filter means being characterized as having a predetermined number of operating parameters;

processing each of said filter output signals in said adaptive neural network means to generate therefrom a plurality of weighted output signals in accordance with an internal processing structure of said adaptive neural network means, one of said weighted output signals representing a class for said input analog signal, and further generating an error signal representing a difference between said one of said weighted output signals and a predetermined desired class; and

inputting said error signal to a control means and generating a further set of operating filter parameters for input to each said programmable filter means to change the operating response thereof to minimize said error signal.

12. The method for classifying signals in accordance with claim 11, wherein said inputting of said analog signal additionally includes inputting of desired output signals into said neural network classifier for generating said error signal.

13. The method for classifying signals in accordance with claim 12, further including the step of performing analog to digital conversion of each said filter output signals prior to input to said neural network.

14. The method for classifying signals in accordance with claim 11, further comprising adjusting a weighted distribution of said neural network to minimize said error signal.