Radar target discrimination systems using artificial neural network topology

• US 5,235,339 A
• Filed: 11/13/1992
• Issued: 08/10/1993
• Est. Priority Date: 11/13/1992
• Status: Expired due to Fees

• Current Assignee: The United States of America As Represented By The Secretary of Agriculture
• Original Assignee: the united states of america as represented by the secretary of the navy
• Inventors: Roth, Duane, Morrison, Larry M.
• Primary Examiner(s): Sotomayor, John B.
• Application Number:
  US07/977,301
• Time in Patent Office: 270 Days
• Field of Search: 342/159, 342/160, 342/162, 342/98, 342/99, 342/101, 342/115, 342/192, 342/194, 342/193, 342/195, 342/196
• US Class Current: 342/159
• CPC Class Codes: G01S 13/52 Discriminating between fixed and moving objects or between objects moving at different speeds
• CPC Class Codes: G01S 7/417 involving the use of neural networks
• CPC Class Codes: G06F 2218/12 Classification; Matching

Abstract

A system for distinguishing between a target and clutter analyzes frequency components of returned wave energy by one or more networks each having inputs receiving successive samples of the returned energy and having outputs individually connected to the inputs through multiplier elements providing selectable factors. The multipliers corresponding to each output are connected to the output through a summing element and a selectable and generally sigmoidal activation function. The factors may be bandpass filter coefficients or discrete Fourier transform coefficients so as to generate frequency components of the energy. Predetermined frequency characteristics of the returned energy may be detected by providing the outputs of a network to a network in which the factors are selected as correlation or convolution coefficients, are selected to integrate fed back outputs, or are selected to sum several outputs within a predetermined range. The activation functions may be selected for thresholding, linearity, limiting, or generation of logarithms.

Claims

1. A method for discriminating between target wave energy and clutter wave energy present together in combined wave energy and having different, predetermined frequency characteristics, the method comprising:

sampling said combined energy at a predetermined rate to generate sample signals representing successive amplitudes of said combined energy;

providing a network receptive to a plurality of input signals and generating a plurality of output signals, the network having;

a plurality of multiplier elements, each of said multiplier elements corresponding to one of said input signals and to one of said output signals and generating a product signal representing the product of the amplitude of the input

signal and a selectable factor individual to the multiplier element,

a plurality of summing elements, each of said summing elements corresponding to one of said output signals and generating a sum signal representing the sum of the product signals corresponding to the output signal, and

a plurality of activation elements, each of said activation elements corresponding to one of said output signals, receiving the sum signal corresponding to the output signal, and generating the output signal in accordance with a selectable activation function of the sum signal;

selecting said factor of each multiplier element so that the factors corresponding to each of said output signals are coefficients effective at said predetermined rate to select a predetermined frequency component of said combined energy;

providing a predetermined number of said sample signals as said input signals to said network so that the product signals corresponding to each of said output signals are summed by the corresponding said summing elements so as to generate said sum signals thereof representing frequency components of said combined energy; and

selecting a predetermined said activation function for each of said activation elements so that said activation elements generate said output signals to represent said frequency components in a predetermined form for analysis to detect said predetermined frequency characteristics.

2. The method of claim 1 wherein said factors of said multiplying elements corresponding to each of said output signals are coefficients of a bandpass filter for a predetermined one of said frequency components so that the sum signal corresponding to the output signal represents a signal having the amplitude of the frequency component in said combined energy at each of said sample signals.

3. The method of claim 1 wherein said factors of said multiplying elements corresponding to each of said output signals are coefficients of a discrete Fourier transform for a predetermined one of said frequency components so that the sum signal corresponding to the output signal represents the amplitude of energy present in said combined energy at the frequency component during said predetermined number of said sample signals.

4. The method of claim 1 wherein said network is a first such network, wherein said output signals thereof are in a first such predetermined form for analysis to detect said predetermined frequency characteristics and wherein the method further comprises:

providing a second such network receiving as the input signals thereof said output signals of said first network representing said frequency components in said first predetermined form;

selecting said factor of each multiplier element of said second network so that the factors thereof corresponding to each of said output signals of said second network select such a predetermined frequency characteristic in at least one of the input signals of said second network; and

selecting a predetermined such activation function for each of said activation elements of said second network so that said activation elements of said second network generate said output signals thereof to represent such a predetermined frequency characteristic in at least one of the input signals of said second network in a second predetermined form for analysis to detect said predetermined frequency characteristics.

5. The method of claim 4 wherein said predetermined frequency characteristic is a correlation of a plurality of said frequency components represented in said first predetermined form and wherein the method further comprises selecting said factors of each multiplier element corresponding to one of said output signals of said second network as a correlation factor for said frequency characteristic of a plurality of frequency components corresponding to predetermined input signals of said second network representing said frequency components related to one of said frequency components so that said one output signal represents said correlation for said one frequency component.

6. The method of claim 4 wherein said predetermined frequency characteristic is a variation of said frequency components during a predetermined plurality of said sample signals and wherein the method further comprises:

receiving said output signals of said first network in said first predetermined form as a set of initial input signals of said second network;

receiving said output signals of said second network as a set of feedback input signals of said second network so that a summing element of said second network and an activation element thereof generate one of said feedback signals;

selecting the factors of said multiplier elements corresponding to each initial signal of said set of initial input signals as a predetermined initial factor to provide the initial signal to a predetermined one of said summing elements of said second network; and

selecting the factor of each multiplier element corresponding to one of said feedback signals and to the one of said summing elements generating the feedback signal as a predetermined feedback factor to provide the feedback signal to the summing element for summing with the initial signal provided thereto by the corresponding initial factor, each feedback factor being selected in relation to the corresponding initial factor so as to integrate successive such initial input signals over said predetermined plurality of said sample signals.

7. The method of claim 4 wherein said predetermined frequency characteristic involves the sum of a plurality of said output signals of said first network and wherein the method further comprises selecting said factors of each multiplier element of said second network corresponding to said plurality of said output signals of said first network and to one of said output signals of said second network so that said plurality of output signals are provided to the summing element corresponding to the one output signal of said second network for summing by said summing element, said factors being selected so that the sum signal thereof is within a predetermined range corresponding to the range of one signal of said plurality of output signals of said first network.

8. The method of claim 2 wherein said network is a first such network and the method further comprises:

providing a second such network;

providing said sample signals in succession as said input signals of said first network;

selecting said factors of said multiplying elements of said first network so that such factors corresponding to a first half of said output signals are said coefficients of said bandpass filter, and so that such factors corresponding to a second half of said output signals are said coefficients of said filter having the same magnitudes as said factors corresponding to said first half of said output signals and being opposite in sign;

selecting said activation function of each activation element of said first network to generate an output signal corresponding to zero when said sum signal to the element is below a zero value and corresponding directly to said sum signal when said sum signal is above a zero value, so that said output signals of said first network represent only positive sum signals thereof, and so that each pair of said output signals of said first network corresponding to a pair of said coefficients having the same magnitude and opposite in sign represent the absolute value of said amplitude of such a frequency component;

providing said first half of said output signals of said first network as a first set of input signals to said second network;

providing said second half of said output signals of said first network as a second set of input signals to said second network; and

selecting said factors of said multiplying elements of said second network to provide each first set signal and each second set signal corresponding to the same said frequency component to the summing element corresponding to a predetermined one of said output signals of said second network, said factors of said second network being selected so that

such factors corresponding to the same frequency component in said first set and in said second set and also corresponding to one of the output signals of said second network have the same value,

such factors of the other multiplying elements corresponding to the one output signal have an effectively zero value, and

the sum signal corresponding to the one output signal corresponds to said absolute value,

whereby the output signals of the second network correspond to rectified such frequency components.

9. The method of claim 3 wherein said network is a first such network and the method further comprises approximating a discrete Fourier transform by:

providing a second such network;

providing said predetermined sample signals as a first set of said input signals of said first network and as a second set of said input signals thereof;

selecting said factors of said multiplying elements of said first network so that

such factors corresponding to a first half of said output signals are in-phase coefficients of a discrete Fourier transform, such in-phase coefficients corresponding to said second set of input signals having the same magnitudes as such in-phase coefficients corresponding to said first set and being opposite in sign, and

such factors corresponding to a second half of said output signals are quadrature coefficients of said discrete Fourier transform for the same frequency components thereof as said first half of said output signals, said quadrature coefficients corresponding to said second set of input signals having the same magnitudes as said quadrature coefficients corresponding to said first set and being opposite in sign;

selecting said activation function of each activation element of said first network to generate an output signal corresponding to zero when said sum signal to the element is below a zero value and corresponding directly to said sum signal when said sum signal is above a zero value, so that

said output signals of said first network represent only positive sum signals thereof,

each pair of said output signals of said first network corresponding to each pair of in-phase coefficients having the same magnitude represent the absolute value of an in-phase frequency component of the discrete Fourier transform, and

each pair of said output signals of said first network corresponding to each pair of quadrature coefficients having the same magnitude represent the absolute value of a quadrature frequency component of a discrete Fourier transform;

providing said first half of said output signals of said first network as an in-phase set of input signals to said second network;

providing said second half of said output signals of said first network as a quadrature set of input signals to said second network; and

selecting said factors of said multiplying elements of said second network to provide each in-phase set signal and each quadrature set signal corresponding to the same said frequency component to the summing element corresponding to a predetermined one of said output signals of said second network, said factors of said second network being selected so that

such factors corresponding to the same frequency component in said in-phase set and in said quadrature set and also corresponding to one of the output signals of said second network have the same value,

such factors of the other multiplying elements corresponding to the one output signal have an effectively zero value, and

the sum signal corresponding to the one output signal represents the sum of said absolute values of the in-phase component and the quadrature component,

whereby the output signals of the second network correspond to the sum of the absolute values of the in-phase and quadrature components of the frequency components and to said discrete Fourier transform.

10. The method of claim 1 wherein said activation function is generally sigmoidal.

11. The method of claim 10 wherein said predetermined form has a first value when said sum signal is below a predetermined threshold and a second value when said sum signal is above said threshold and said activation function is selected as a signum having an asymptotic portion at said first value, a step at said threshold, and another asymptotic portion at said second value.

12. The method of claim 10 wherein said predetermined form corresponds directly to a predetermined range of said sum signal and said activation function is a sigmoid selected to have a substantially linear central portion disposed at said range.

13. The method of claim 10 wherein said predetermined form has one value when said sum signal is below a predetermined level and corresponds directly to said sum signal when said sum signal is above said level and said activation function is selected as a sigmoid having an asymptotic portion at said one level and a substantially linear central portion extending from said asymptotic portion.

14. The method of claim 10 wherein said predetermined form corresponds to a logarithm of a positive said sum signal and said activation function is selected to have a generally logarithmical curved asymptotic portion.