# US 2024/0004032 A1 · Deep neural network (dnn)-based multi-target constant false alarm rate (cfar) detection methods

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2024/0004032 A1
- Document ID: US20240004032A1
- Status: Published application
- Kind: A1
- Application number: 18/451,818
- Filed: 2023-08-17
- Published: 2024-01-04
- Earliest claimed: 2021-07-14
- Assignees: Zhejiang University; Donghai Laboratory
- Applicants: Zhejiang University; Donghai Laboratory
- Inventors: Chunyi SONG; Zhihui CAO; Zhiwei XU; Yuying SONG; Fuyuan AI; Jingxuan WU
- Web page: https://patentgrep.com/patent/US20240004032A1

## CPC

- G01S 7/417: {involving the use of neural networks}
- G01S 7/354: {Extracting wanted echo-signals (Doppler systems G01S13/50)}
- G01S 13/536: using transmission of continuous unmodulated waves, amplitude-, frequency-, or phase-modulated waves
- G01S 13/726: {Multiple target tracking}
- G01S 13/89: for mapping or imaging
- G06N 3/04: Architecture, e.g. interconnection topology
- G06N 3/045: Combinations of networks
- G06N 3/084: Backpropagation, e.g. using gradient descent
- G01S 13/5246: {post processors for coherent MTI discriminators, e.g. residue cancellers, CFAR after Doppler filters}
- G01S 2013/9314: {Parking operations}
- Y02A 90/10: Information and communication technologies [ICT] supporting adaptation to climate change, e.g. for weather forecasting or climate simulation
- G01S: RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- Y02A: TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE

## Abstract

The embodiment of the present disclosure provides a deep neural network (DNN)-based multi-target constant false alarm rate (CFAR) detection method. The method includes: obtaining target values to be measured based on radar IF (IF) signals to be detected, the target values to be measured including a measured frequency value and a measured intensity value of the radar IF signals; obtaining peak sequences based on the target values to be measured; generating a target detection result by processing the peak sequences based on a DNN detector, the DNN detector being a machine learning model; generating approximated maximum likelihood estimation (AMLE) of a scale parameter based on an approximated maximum likelihood estimator; generating a false alarm adjustment threshold based on a preset false alarm rate and the AMLE; and generating a constant false alarm detection result by processing the target detection result based on the false alarm adjustment threshold.

## Claims

### Claim 1

A deep neural network (DNN)-based multi-target constant false alarm rate (CFAR) detection method, implemented based on a processor, comprising:
obtaining target values to be measured based on radar intermediate frequency (IF) signals to be detected, the target values to be measured including a measured frequency value and a measured intensity value of the radar IF signals to be detected;
obtaining peak sequences based on the target values to be measured;
generating a target detection result by processing the peak sequences based on a DNN detector, the DNN detector being a machine learning model;
generating approximated maximum likelihood estimation (AMLE) of a scale parameter based on an approximated maximum likelihood estimator;
generating a false alarm adjustment threshold based on a preset false alarm rate and the AMLE; and
generating a constant false alarm detection result by processing the target detection result based on the false alarm adjustment threshold.

### Claim 2 (depends on 1)

The method of claim 1, wherein the DNN detector is obtained by training an initial DNN detector based on a simulation data set satisfying a first preset condition.

### Claim 3 (depends on 2)

The method of claim 2, wherein the first preset condition includes that: radar IF signal samples in the simulation data set have dynamic signal-to-clutter ratios (SCRs) and a dynamic count of targets.

### Claim 4 (depends on 3)

The method of claim 3, wherein the first preset condition is implemented based on a data augmentation manner, and the data augmentation manner includes:
augmenting the simulation data set based on the dynamic SCRs by setting the dynamic SCRs corresponding to the radar IF signal samples; and
augmenting the simulation data set by setting the dynamic count of targets for the radar IF signal samples and generating a distance sequence based on an additive random sampling manner.

### Claim 5 (depends on 4)

The method of claim 4, wherein the dynamic SCRs and the dynamic count of targets are determined based on an estimated scene feature distribution of a target place;
the estimated scene feature distribution includes an estimated target distribution and an estimated environmental distribution; and
the estimated target distribution includes a target count distribution and a target density distribution.

### Claim 6 (depends on 5)

The method of claim 5, wherein a variation factor of the dynamic SCRs obeys a first Gaussian distribution; and a value of a first standard deviation of the first Gaussian distribution is related to the estimated environmental distribution.

### Claim 7 (depends on 6)

The method of claim 6, wherein the first standard deviation is also related to the target density distribution.

### Claim 8 (depends on 5)

The method of claim 5, wherein the dynamic count of targets is randomly generated based on a first preset distribution; and the first preset distribution is determined based on the estimated target distribution.

### Claim 9 (depends on 8)

The method of claim 8, wherein a mean value of the first preset distribution is related to an upper limit of the target count distribution.

### Claim 10 (depends on 5)

The method of claim 5, wherein a scaling factor and a distance variation factor of the additive random sampling manner are determined based on the estimated target distribution.

### Claim 11 (depends on 10)

The method of claim 10, wherein the scaling factor and the distance variation factor are randomly generated based on a second preset distribution; and the second preset distribution is determined based on the estimated target distribution.

### Claim 12 (depends on 11)

The method of claim 11, wherein a mean value of the second preset distribution is related to an upper limit of the target density distribution.

### Claim 13 (depends on 1)

The method of claim 1, wherein the DNN detector adopts a full connect neural network (FCNN).

### Claim 14 (depends on 1)

The method of claim 1, wherein the approximated maximum likelihood estimator is related to curtailing samples; and
the curtailing samples are obtained by:
in response to the target values to be measured satisfying a second preset condition, obtaining the curtailing samples based on the target values to be measured and the target detection result.

### Claim 15 (depends on 1)

The method of claim 1, wherein the false alarm adjustment threshold is determined based on the preset false alarm rate and the scale parameter.

### Claim 16

A deep neural network (DNN)-based multi-target constant false alarm rate (CFAR) detection system, comprising:
a detection device configured to obtain radar intermediate frequency (IF) signals to be detected; and
a processor configured to:
obtain target values to be measured based on the radar IF signals to be detected, the target values to be measured including a measured frequency value and a measured intensity value of the radar IF signals to be detected;
obtain peak sequences based on the target values to be measured;
generate a target detection result by processing the peak sequences based on a DNN detector, the DNN detector being a machine learning model;
generate approximated maximum likelihood estimation (AMLE) of a scale parameter based on an approximated maximum likelihood estimator;
generate a false alarm adjustment threshold based on a preset false alarm rate and the AMLE; and
generate a constant false alarm detection result by processing the target detection result based on the false alarm adjustment threshold.

### Claim 17 (depends on 1)

A non-transitory computer-readable storage medium storing computer instructions, wherein after reading the computer instructions in the storage medium, a computer performs the DNN-based multi-target CFAR detection method of claim 1.
