Claims:

1) A method for classifying a target from micro-Doppler signatures of the target, the method comprising:
determining a micro-Doppler signature of said target from echo signals of said target received from said target;
accumulating the determined micro-Doppler signatures over a pre-determined time period;
converting the accumulated micro-Doppler signatures into an image of said target;
comparing said image of said target with one or more pre-stored images;
determining a class of said image of said target based on the comparison; and
in the event of a successful determination, assigning the class to said target, or
in the event of an unsuccessful determination, creating a new class of said image of said target and assigning the new class to said target.

2) The method as claimed in claim 1, wherein the method, prior to the step of determining the micro-doppler signature, comprises:
receiving echo signals reflected from said target in response to electromagnetic waves transmitted by a radar towards said target; and
detecting said target based on a constant false alarm rate (CFAR) value derived from the echo signals.

3) The method as claimed in claim 1, wherein the micro-doppler signature is selected from short-time Fourier transform (STFT), bispectrum and bicoherence.

4) The method as claimed in claim 1, wherein the method comprises creating a catalogue of the one or more pre-stored images corresponding to their respective micro-Doppler signatures.

5) The method as claimed in claim 1, wherein the new class is created using a neural network model.

6) The method as claimed in claim 5, wherein the method comprises:
training the neural network model;
testing the trained neural network model; and
creating the new class in the event of failure of the neural network model during said testing and training the neural network model with the new class.

7) A system for classifying a target from micro-Doppler signatures of said target, said system comprising at least a processing device configured to:
determine a micro-Doppler signature of said target from echo signals of said target received from said target;
accumulate the determined micro-Doppler signatures over a pre-determined time period;
convert the accumulated micro-Doppler signatures into an image of said target;
compare said image of said target with one or more pre-stored images;
determine a class of said image of said target based on the comparison; and
in the event of a successful determination, assign the class to said target, or
in the event of an unsuccessful determination, create a new class of said image of said target and assign the new class to said target.

8) The system as claimed in claim 7, wherein said processing device is configured to:
receive echo signals reflected from said target in response to electromagnetic waves transmitted by a radar towards said target; and
detect said target based on a constant false alarm rate (CFAR) value derived from the echo signals.

9) The system as claimed in claim 7, wherein said processing device is further configured to create the new class through a neural network.

10) The system as claimed in claim 9, wherein said processing device is further configured to:
train the neural network model;
test the trained neural network model; and
create the new class in the event of failure of said neural network model during said testing and train the neural network model with the new class.
, Description:FORM 2

THE PATENTS ACT, 1970
(39 of 1970)
&
THE PATENTS RULES, 2003

COMPLETE SPECIFICATION
(see Section 10, Rule 13)

METHOD FOR CLASSIFYING A TARGET FROM MICRO-DOPPLER SIGNATURES OF THE TARGET AND SYSTEM THEREFOR

BHARAT ELECTRONICS LIMITED
Address: Outer Ring Road, Nagavara, Bangalore 560045,
Karnataka, India.

The following specification particularly describes the invention
and the manner in which it is to be performed. 

TECHNICAL FIELD
The present invention relates to signal processing. The invention, more particularly, relates to radar signal processing and classification of the radar targets in real-time.

BACKGROUND
Electromagnetic waves transmitted by a radar towards a target are reflected from the target as echo signals which are used to extract the target characteristics. In case of a moving target, during the reception of the echo signals, the radar carrier frequency may be shifted due to Doppler effect induced by the moving target. Additionally, if the target or any part of the target undergoes micro-motions such as mechanical vibrations or rotations, further frequency modulation or Doppler modulations are induced on the received echo signals, referred to micro-Doppler effect. These Doppler modulations become a distinctive signature of the target which typically provides the identity of the target and is used to classify the target.
Conventionally various micro-Doppler measurement tools such as short-time Fourier transform (STFT), bispectrum and bicoherence have been used for radar target classification. However, the success of the classification process depends on the quality of available catalogue of the micro-Doppler signatures corresponding to various targets and waveforms. The operator of a radar must be well acquainted with the catalogue of the targets, radar waveforms and micro-Doppler signatures. Based on the knowledge of the catalogue, the operator would analyse the observed micro-Doppler signatures in the radar display and conclude the types and classes of the targets. This kind of manual or visual analysis may not be practical in many situations where the classification must be done as quickly as possible. The quality of the classification depends on the visual and analytical capability of the individual operator and may vary from person to person.
Therefore, there is a need for an invention which can perform classification of targets independently of any operator and with acceptable accuracy in real-time.

SUMMARY
This summary is provided to introduce concepts of the present invention. This summary is neither intended to identify essential features of the present invention nor is it intended for use in determining or limiting the scope of the present invention.
In accordance with an embodiment of the present invention, there is provided a method for classifying a target from micro-Doppler signatures of the target. The method comprises: determining a micro-Doppler signature of the target from echo signals of the target received from the target; accumulating the determined micro-Doppler signatures over a pre-determined time period; converting the accumulated micro-Doppler signatures into an image of the target; comparing the image of the target with one or more pre-stored images; determining a class of the image of the target based on the comparison; and in the event of a successful determination, assigning the class to the target, or in the event of an unsuccessful determination, creating a new class of the image of the target and assigning the new class to the target.
In an aspect, the method, prior to the step of determining the micro-doppler signature, comprises: receiving echo signals reflected from the target in response to electromagnetic waves transmitted by a radar towards the target; and detecting the target based on a constant false alarm rate (CFAR) value derived from the echo signals.
Typically, the micro-doppler signature is selected from short-time Fourier transform (STFT), bispectrum and bicoherence.
In an aspect, the method comprises creating a catalogue of the one or more pre-stored images corresponding to their respective micro-Doppler signatures.
Preferably, the new class is created using a neural network model.
In an aspect, the method comprises: training the neural network model; testing the trained neural network model; and creating the new class in the event of failure of the neural network model during the testing and training the neural network model with the new class.
In accordance with another embodiment of the present invention, there is provided, a system for classifying a target from micro-Doppler signatures of the target. The system comprises at least a processing device configured to: determine a micro-Doppler signature of the target from echo signals of the target received from the target; accumulate the determined micro-Doppler signatures over a pre-determined time period; convert the accumulated micro-Doppler signatures into an image of the target; compare the image of the target with one or more pre-stored images; determine a class of the image of the target based on the comparison; and in the event of a successful determination, assign the class to the target, or in the event of an unsuccessful determination, create a new class of the image of the target and assign the new class to the target.
In an aspect, the processing device is configured to: receive echo signals reflected from the target in response to electromagnetic waves transmitted by a radar towards the target; and detect the target based on a constant false alarm rate (CFAR) value derived from the echo signals.
Preferably, the processing device is further configured to create the new class through a neural network.
In an aspect, the processing device is further configured to: train the neural network model; test the trained neural network model; and create the new class in the event of failure of the neural network model during the testing and train the neural network model with the new class.

BRIEF DESCRIPTION OF ACCOMPANYING DRAWINGS
The detailed description is described with reference to the accompanying figures. In the figures, the left-most digit(s) of a reference number identifies the figure in which the reference number first appears. The same numbers are used throughout the drawings to reference like features and modules.
Figure 1 illustrates a flowchart depicting the steps of a method for classifying a target from micro-doppler signatures of the target, according to an embodiment of the present invention.
Figure 2(a-h) illustrates the micro-Doppler signatures of different types of images of different targets for classifying the targets, according to an embodiment of the present invention.
Figure 3 illustrates a flowchart depicting the steps involved in training a neural network to be utilized in the method for classifying a target, according to an embodiment of the present invention.
Figure 4 illustrates a flowchart depicting the steps involved in testing a neural network to be utilized in the method for classifying a target, according to an embodiment of the present invention.
Figure 5 illustrates a flowchart depicting the steps involved in creating a new target class by a neural network to be utilized in the method for classifying a target, according to an embodiment of the present invention.
It should be appreciated by those skilled in the art that any block diagrams herein represent conceptual views of illustrative methods embodying the principles of the present invention. Similarly, it will be appreciated that any flow charts, flow diagrams, and the like represent various processes which may be substantially represented in computer readable medium and so executed by a computer or processor, whether or not such computer or processor is explicitly shown.

DETAILED DESCRIPTION
The various embodiments of the present invention describe about a method for classifying a target from micro-doppler signatures of the target and a system for the same. The micro-doppler signatures of the target are analyzed without human intervention and classified in real time thereby contributing to the smartness of a radar.
In the following description, for purpose of explanation, specific details are set forth in order to provide an understanding of the present invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced without these details. One skilled in the art will recognize that embodiments of the present invention, some of which are described below, may be incorporated into a number of systems.
However, the methods and systems are not limited to the specific embodiments described herein. Further, structures and devices shown in the figures are illustrative of exemplary embodiments of the present invention and are meant to avoid obscuring of the present invention.
It should be noted that the description merely illustrates the principles of the present invention. It will thus be appreciated that those skilled in the art will be able to devise various arrangements that, although not explicitly described herein, embody the principles of the present invention. Furthermore, all examples recited herein are principally intended expressly to be only for explanatory purposes to help the reader in understanding the principles of the invention and the concepts contributed by the inventor to furthering the art and are to be construed as being without limitation to such specifically recited examples and conditions. Moreover, all statements herein reciting principles, aspects, and embodiments of the invention, as well as specific examples thereof, are intended to encompass equivalents thereof.
Micro-Doppler content appears in radar returns or echo signals due to the Doppler modulations produced by micro-motions. The time-varying response of micro-motions forms a micro-Doppler signature in the form of multi-component frequency modulated signals. Micro-Doppler effect begins with the movement of a non-rigid object/part of the target, when the global change consists of many local micro-motions.
A primary technique for micro-Doppler study is time-frequency representation. A traditional spectrogram named as the squared modulus of the short-time Fourier transform (STFT) is a broadly developed tool utilized for the time-frequency study of micro-Doppler information. The spectrogram is computed using a sliding window for truncating a received echo signal by the sequence of short-time divisions (messages) and Fourier transform related to that short-time segmented message. A more substantial window length gives higher frequency resolution in the STFT, but, as a result, the more inadequate temporal resolution appears at the same point in time.
In the present invention Deep Convolutional Neural Network (DCNNs) has been used for classification of micro-Doppler signatures’ images estimated by the radar. Notwithstanding that machine learning techniques having been applied widely in radar signature image classification in the last decades, their performance could not be fully implemented in the radar area due to the deficiency of the training examples. It is a critical task because of the very high labour work and financial costs associated with taking radar images. However, with recent advances in radar technologies it is possible to collect high training data. This allows training of authentic classification model from authentic data instead of the data generated using transfer learning and Generative Adversarial Networks (GANs) used conventionally. Hence, in the present invention, DCNNs are used to classify the micro-Doppler signatures and the different activities obtained from the Doppler radar. In the present invention short-time Fourier transform (STFT) signatures’ images of multiple different classes are used for training purpose. As per the results, the application of DCNNs modifies the accuracy of classification.
In accordance with the present invention, the method for classifying a target from micro-doppler signatures of the target is developed based on some basic assumptions which are as follows: each of the techniques used for obtaining micro-Doppler signatures of different radar targets produces signatures which are periodic in nature with respect to time and the micro-motion in the targets; the micro-Doppler signatures are created and made available in real-time while the operation of the radar is not hampered in terms of performance; and the signatures corresponding to motions other than micro-motion or those corresponding to stationary targets are similar in nature and therefore are avoidable.
In accordance with the present invention, the method for classifying a target from micro-Doppler signatures of the target involves reception of the micro-Doppler signatures in real-time, conversion of the micro-Doppler signatures into images with valid pixel values, creation of a catalogue of such images with respect to various targets and radar waveforms, analysis of a test image with the help of the catalogue, and determination of a class of the test target based on the analysis. This method does not affect the performance of the radar as the reception and recording (if necessary) of the micro-Doppler signatures are done in hardware independent of the signal processing chain. Further, the conversion of the received and recorded micro-Doppler signatures into images with pixel values may be processed consequently. The catalogue created for the images with respect to different types of targets and radar waveforms will be essential for the target classification as a reference to be used for declaring the type or class of the radar target. The analysis of the test image with the help of the catalogue involves finding the image in the catalogue having the maximum similarity with the test image by comparing the test image with the images in the catalogue. The class or type of the radar target based on the analysis is determined based on the similarity found between the test image and the catalogue images which helps estimating the class of the radar target.
The main contributions of the method are: use of incoming short-time Fourier transform (STFT) images of the radar targets for training of the DCNN model, using only the received STFT signatures of the target to carry out identification of the target during the test time, and use of multiple different classes of objects to train the classification model; use of multiple convolutional layers and a plurality of edge detection convolutional filters in each layer to extract the features of the target; and use of authentically generated data for training the DCNN architecture.
The method for classifying a target according to the present invention uses tensor flow framework based DCNN, and incorporates multiple different classes of radar targets. In an exemplary non-limiting embodiment, eight different classes of radar targets are incorporated as follows: (1) Breathing Stationary (2) Fan AZ 90 1 (3) Fan AZ 90 2 (4) Fan EI 90 1 (5) Fan EI 90 2 (6) Fan LOS (7) Walking (8) Walking Breathing. Figures 2(a-h) show the STFT signatures of images of different classes. As can be seen from the different class signatures, each class has some kind of similarity in a signature. These received target signatures are used to train the DCNN model. In testing phase, signatures of images are tested and the class is predicted.
In accordance with the aforesaid exemplary embodiment, DCNN training model with 3 convolutional layers having filter size of 3 X 3 is implemented, such that in the first, second and third layer, thirty-two, thirty-two, and sixty-four filters, respectively are used, with a learning rate of 0.0001 and same padding, to calculate total number of epochs.
Total epochs =
1 + int(num_iterations/int(data.train.num_example/batch_size))
The above formula is very effective to find the total number of epochs. In accordance with the aforesaid exemplary embodiment, number of iteration (num_iterations) = 1000, batch_size = 20, img_size = 128 x 128 x 3 (where 3 is number channels), validation size = 0.2. There are total eight numbers of classes. Breathing stationary class consist of 2115 images. Fan AZ 90 1 consist of 2178 images. Fan AZ 90 2 consist of 2246 images. Fan EI 90 1 consist of 2266 images. Fan EI 90 2 consist of 2211 images. Fan LOS consist of 2483 images. Walking breathing consist of 2244 images. Walking consist of 2158 images.
Referring now to Figure 1, a flowchart depicting the steps of the method 100 for classifying a target from micro-doppler signatures of the target, in accordance with an embodiment of the present invention is illustrated. Electromagnetic waves transmitted by a radar towards a target are reflected from the target as echo signals which are received at step 101. At step 102, the detection of the radar targets is performed using Constant False Alarm Rate (CFAR) value P_fa which is derived from the echo signals by known processes. At step 103, the calculation of a relevant micro-Doppler signature measurement tool is carried out. The tool includes short-time Fourier transform (STFT) and/or bicepstrum (bispectrum and bicoherence). The tool generates signatures only when the output of the measurements are accumulated over time. Hence, this accumulation is performed at step 104. The time period over which the accumulation is done is T. At step 105, the accumulated micro-Doppler signatures are transformed into an N×N pixel image of the target which may be used later for classifying the signature. The classification of the target is achieved by the comparison of the image with pre-stored images corresponding to their respective micro-Doppler signatures in a catalogue. This comparison is performed at step 106 where the neural network techniques elaborated in the Figure 2 are employed. Once a new image is generated in the system, updating the existing catalogue becomes necessary as the new image may come from a class or type of target which may not have been encountered by the radar before. The comparison of the image with the pre-stored images in the catalogue performed in the previous step 106 is also used in updating the catalogue. At step 107, the presence of a class of the image of the target corresponding the micro-Doppler signature in the catalogue is determined. If the answer is yes, then the target is assigned the determined class as shown at step 109. However, if the answer is no, then the method passes to step 108 which deals with the task of the creation of new class of the image corresponding to the micro-Doppler signature in the catalogue and the declaration of the creation as shown in Figure 3.
Referring to Figure 3, a flowchart depicting the steps involved in training a neural network to be utilized in the method for classifying the target, according to an embodiment of the present invention, is illustrated. In training phase shown in Figure 3, data of images corresponding to micro-Doppler signatures of the targets is collected and stored as per their class at step 301. One or more data pre-processing tasks are performed at step 302. At step 303, a classification model of a neural network, for example DCNN model, is prepared and the training data is fed to the model for training the model. At step 304, the trained model is saved for testing purpose.
Referring to Figure 4, a flowchart depicting the steps involved in testing the neural network to be utilized in the method for classifying the target, according to an embodiment of the present invention, is illustrated. Figure 4 represents the framework of testing phase of the neural network model of Figure 3. At step 401, test data of test image corresponding to micro-Doppler signatures of a test target is pre-processed as per need. Generally in this step, test data is prepared in the same manner as the training data is prepared. This is done so that the neural network model should able to read the test data and process the test data. In model testing step 402, the test data is fed to the saved neural network model. Afterwards, the neural network model performs the classification task at step 403. Finally, a suitable class label is assigned to the test data/image in module 404. This label indicates the potential class for which this test data/image belongs.
Referring to Figure 5, a flowchart depicting the steps involved in creating a new target class by the neural network to be utilized in the method for classifying the target, according to an embodiment of the present invention, is illustrated. In Figure 5, a case is shown where a test target does not belong to existing list of targets as represented in step 501. At step 502, the neural network model will fail to recognize that unknown test target. At step 503, the test image for which the neural network model is not able to identify its class is saved. Further, additional data from that test target which is unknown is collected. At step 504, the neural network model is again trained with newly added class datasets (i.e. data of the image corresponding to micro-Doppler signatures of that test target) which makes the neural network model in much better position to recognize that unknown test target.
With the neural network model of Deep Convolutional neural network (DCNN) about 85% of test accuracy has been achieved in classifying targets. The accuracy can be further increased by using enhanced dataset for classification task.
Thus, the method for classifying a target from micro-Doppler signatures of the target comprises the following steps: receiving echo signals reflected from the target in response to electromagnetic waves transmitted by a radar towards the target; detecting the target based on a constant false alarm rate (CFAR) value derived from the echo signals; determining a micro-Doppler signature of the target from echo signals of the target, wherein the micro-Doppler signature is selected from short-time Fourier transform (STFT), bispectrum and bicoherence; accumulating the determined micro-Doppler signatures over a pre-determined time period; converting the accumulated micro-Doppler signatures into an image of the target; comparing the image of the target with one or more pre-stored images corresponding to their respective micro-Doppler signatures in a catalogue; determining a class of the image of the target based on the comparison; and in the event of a successful determination, assigning the class to the target, or in the event of an unsuccessful determination, creating a new class of the image of the target using a neural network model such as DCNN and assigning the new class to the target; wherein the method includes training the neural network model, testing the trained neural network model, and creating the new class in the event of failure of the neural network model during the testing and training the neural network model with the new class.
The aforesaid method is typically performed by a system for classifying a target from micro-Doppler signatures of the target. In exemplary embodiments, the system may be implemented through at least a processing device such as microcontroller, microprocessor, digital signal processor (DSP), field programmable gate array (FPGA), application specific integrated circuit (ASIC), or other like electronic processing devices. In other exemplary embodiments, the processing device may be a computing device such as a computer, a tablet, etc.
In an exemplary embodiment, the processing device is configured to: receive echo signals reflected from the target in response to electromagnetic waves transmitted by a radar towards the target; detect the target based on a constant false alarm rate (CFAR) value derived from the echo signals; determine a micro-Doppler signature of the target from echo signals of the target, wherein the micro-Doppler signature is selected from short-time Fourier transform (STFT), bispectrum and bicoherence; accumulate the determined micro-Doppler signatures over a pre-determined time period; convert the accumulated micro-Doppler signatures into an image of the target; compare the image of the target with one or more pre-stored images corresponding to their respective micro-Doppler signatures in a catalogue; determine a class of the image of the target based on the comparison; and in the event of a successful determination, assign the class to the target, or in the event of an unsuccessful determination, create a new class of the image of the target using a neural network model such as DCNN and assign the new class to the target; wherein the system is further configured to train the neural network model, test the trained neural network model, and create the new class in the event of failure of the neural network model during the testing and train the neural network model with the new class.
The method and system in accordance with the present invention enables classification of radar targets by analysing the micro-Doppler signatures generated by the targets. The shape and periodicity of the micro-Doppler signatures corresponding to a radar target depends on the target’s RCS (Radar Cross Section), movement, the radar waveform, etc. Visual inspection of the micro-Doppler signature may be enough to know the nature or class of the target based on the catalogue of micro-Doppler signatures of various targets previously recorded. But in real-time applications visual inspection may not always be possible. Therefore, the present invention is necessary which will receive the micro-Doppler signature in real-time, analyse the signature in real-time, and point out the class of the radar target as the result of its analysis in real time.
The foregoing description of the invention has been set merely to illustrate the invention and is not intended to be limiting. Since modifications of the disclosed embodiments incorporating the substance of the invention may occur to person skilled in the art, the invention should be construed to include everything within the scope of the invention.