CN112099014B - Road millimeter wave noise model detection estimation method based on deep learning - Google Patents
Road millimeter wave noise model detection estimation method based on deep learning Download PDFInfo
- Publication number
- CN112099014B CN112099014B CN202010859256.3A CN202010859256A CN112099014B CN 112099014 B CN112099014 B CN 112099014B CN 202010859256 A CN202010859256 A CN 202010859256A CN 112099014 B CN112099014 B CN 112099014B
- Authority
- CN
- China
- Prior art keywords
- noise
- road
- layer
- feature
- data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 48
- 238000013135 deep learning Methods 0.000 title claims abstract description 43
- 238000001514 detection method Methods 0.000 title claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 16
- 230000008569 process Effects 0.000 claims abstract description 12
- 238000006243 chemical reaction Methods 0.000 claims abstract description 11
- 238000007781 pre-processing Methods 0.000 claims abstract description 8
- 238000013507 mapping Methods 0.000 claims description 31
- 239000011159 matrix material Substances 0.000 claims description 30
- 238000012549 training Methods 0.000 claims description 30
- 238000000605 extraction Methods 0.000 claims description 24
- 238000005070 sampling Methods 0.000 claims description 23
- 238000011084 recovery Methods 0.000 claims description 17
- 238000010606 normalization Methods 0.000 claims description 12
- 238000010586 diagram Methods 0.000 claims description 10
- 238000003491 array Methods 0.000 claims description 3
- 238000001914 filtration Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 3
- 230000010355 oscillation Effects 0.000 claims description 3
- 230000035945 sensitivity Effects 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 11
- 230000004913 activation Effects 0.000 description 6
- 206010039203 Road traffic accident Diseases 0.000 description 4
- 230000007613 environmental effect Effects 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 102100034112 Alkyldihydroxyacetonephosphate synthase, peroxisomal Human genes 0.000 description 2
- 101000799143 Homo sapiens Alkyldihydroxyacetonephosphate synthase, peroxisomal Proteins 0.000 description 2
- 238000000848 angular dependent Auger electron spectroscopy Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012935 Averaging Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- 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
- G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
- G01S13/88—Radar or analogous systems specially adapted for specific applications
- G01S13/93—Radar or analogous systems specially adapted for specific applications for anti-collision purposes
- G01S13/931—Radar or analogous systems specially adapted for specific applications for anti-collision purposes of land vehicles
-
- G—PHYSICS
- G01—MEASURING; TESTING
- 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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/415—Identification of targets based on measurements of movement associated with the target
-
- G—PHYSICS
- G01—MEASURING; TESTING
- 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
- G01S7/00—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00
- G01S7/02—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00
- G01S7/41—Details of systems according to groups G01S13/00, G01S15/00, G01S17/00 of systems according to group G01S13/00 using analysis of echo signal for target characterisation; Target signature; Target cross-section
- G01S7/418—Theoretical aspects
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T90/00—Enabling technologies or technologies with a potential or indirect contribution to GHG emissions mitigation
Abstract
The invention relates to a road millimeter wave noise model detection estimation method based on deep learning, which comprises the following steps: s1: preprocessing millimeter wave radar echo signals of R road sections; s2: performing time-frequency conversion on the digitized intermediate frequency signal, and extracting road noise clutter characteristics; s3: processing the noise clutter characteristics to generate a normalized noise clutter characteristic data set; s4: adding road coordinate information and uploading the road coordinate information to a road end server; s5: the roadside server processes the data. And S6, the road end server deep learning network trains the parameters of the millimeter wave noise model of the road. The invention can improve the sensitivity and accuracy of the vehicle radar detection target and enhance the safety in the driving process.
Description
Technical Field
The invention relates to the technical field of millimeter wave radar application, in particular to a road millimeter wave noise model detection and estimation method based on deep learning.
Background
With the advent of the 5G age, the information age has seen a great advance. In the traffic field, advances in information technology can greatly enhance the safety performance of vehicles. ADAS, advanced driving assistance system, is a research hotspot in the current automotive unmanned field. Signal processing of automotive radars is one of the key technologies of ADAS, and common automotive radars include ultrasonic radars, laser radars and millimeter wave radars. The millimeter wave radar has the effective spectrum bandwidth of 30GHz to 300GHz, short wavelength, small volume, light weight and high precision, can work in all weather and is little influenced by the climate environment.
In the research field of radar, the most important is the accurate detection of target signals, including target object information such as target distance, speed, angle and the like, as is the vehicle millimeter wave radar. At present, a great deal of research work is done on millimeter wave radar signal processing methods such as noise removal, interference suppression and the like. In the aspect of sensing the whole road environment by adopting millimeter wave radar signals, corresponding literature and technical data are relatively less. The conventional research thought is that a millimeter wave radar receives radar echo signals, and then analyzes and processes the signals to extract useful target information.
When the traditional signal processing thought encounters a road noise and noise wave is relatively large, the accuracy of detecting a target object by the millimeter wave radar is greatly reduced, and the false alarm probability of radar detection is also improved along with the increase of interference noise. In the road with higher environmental noise, the automatic driving system is affected by noise clutter in the road, if the radar of the vehicle finds the target vehicle ahead later, the braking distance is short, and the riding comfort of a driver and personnel in the vehicle can be affected by severe braking in a short time, so that serious traffic accidents can be even caused. In order to avoid casualties and property loss caused by traffic accidents, more advanced methods must be adopted to deal with the problem of road environmental noise.
Disclosure of Invention
The invention provides a road millimeter wave noise model detection estimation method based on deep learning, which aims to overcome the defect that the sensitivity and accuracy of a vehicle-mounted radar detection target are not high enough due to environmental noise in the prior art.
The method comprises the following steps:
s1: preprocessing millimeter wave radar echo signals of R road sections; the preprocessing process comprises the steps of mixing a received signal with a radar local oscillation signal to obtain an intermediate frequency signal, filtering the intermediate frequency signal, and digitizing the filtered intermediate frequency signal through AD conversion.
S2: and performing time-frequency conversion on the digitized intermediate frequency signal to extract the noise clutter characteristic of the road.
S3: the noise clutter characteristics are processed to generate a normalized noise clutter characteristic dataset.
S4: road information data is acquired, and the road information data and noise characteristic data are integrated.
S5: the road end server builds a road millimeter wave noise deep learning network model, and performs data processing on road information data and noise characteristic data;
s6: and deeply learning and training a road millimeter wave noise deep learning network model to obtain optimal network model parameters, wherein the noise model is represented by the trained model parameters, and detection and estimation of the noise model are realized according to the model parameters.
Preferably, S2 comprises the steps of:
s2.1: taking a radar echo intermediate frequency signal of an R-th road section from the preprocessed millimeter wave radar echo signals of the R road sections;
s2.2, taking an nth Chirp signal, wherein the value range of N is 1 to N, and N represents the number of all Chirp signals in the radar echo intermediate frequency signal of the nth road section;
s2.3: performing time-frequency conversion on the nth Chirp radar echo intermediate frequency signal; obtaining a frequency domain signal;
s2.4: extracting road noise clutter characteristics from the obtained frequency domain signals; and store noise clutter characteristic data in an arrayWherein, the array subscript R represents the R-th road section, and the value of R is 1 to R; the value of the upper index i of the array is 1 to 9, and the upper index i represents 9 noise clutter characteristic data;
the road noise clutter characteristic data are respectively as follows: noise clutter characteristics I; noise clutter characteristics ii; noise clutter characteristics III;
s2.5: judging whether N is greater than or equal to N, if not, making n=n+1, and returning to S2.2; if yes, go to step S2.6.
S2.6: output N Chirp frequency domain information noise clutter characteristic arrays
Preferably, S2.4 comprises the steps of:
s2.4.1: noise clutter characteristic i data are calculated from frequency domain signal data:
calculating a frequency domain signal mean value:store in array->
Calculating the frequency domain signal range: r=x max -X min Stored in array
Calculating the median of the frequency domain signal: z=media (X i ) The method comprises the steps of carrying out a first treatment on the surface of the Stored array
Step S2.4.2: calculating noise clutter characteristic II data from the frequency domain signal and the noise clutter characteristic I data:
calculating frequency domain signal variance:store in array->
Calculating standard deviation of the frequency domain signals:store in array->
Calculating the average deviation of the frequency domain signals:store in array->
Step S2.4.3: calculating noise clutter characteristic III data from the frequency domain signal and the noise clutter characteristic I data and the noise clutter characteristic II data:
calculating the relative average deviation of the frequency domain signals:store in array->
Calculating the relative standard deviation of the frequency domain signals:store in array->
Calculating the absolute deviation of the median of the frequency domain signal: mad=media i (|X i -median j (X j ) I), stored in an array
Preferably, S3 comprises the steps of:
s3.1: the array from S2In the recipe, get->Normalization processing is carried out to obtain normalized noise clutter characteristic data +.>
S3.2: judging whether i is greater than or equal to 9, if not, making i=i+1 and returning to S3.1; if yes, executing S3.3;
s3.3: from normalized noise clutter characteristic dataMatrix m forming 3×3 n The method comprises the steps of carrying out a first treatment on the surface of the Wherein n represents an nth Chirp signal;
s3.4: judging whether N is greater than or equal to N, if not, making n=n+1, and returning to S3.3; if yes, executing step S3.5; after this step is completed, N matrixes m are obtained n ；
S3.5: from matrix m n Noise clutter characteristic matrix M with composition length of sxs r The method comprises the steps of carrying out a first treatment on the surface of the Wherein the value of s isM r The subscript r of (2) represents the r-th road segment;
s3.6: judging whether R is greater than or equal to R, if not, making r=r+1, and returning to S3.1; if so, outputting a noise characteristic matrix M r I.e. noise clutter characteristic data sets.
Preferably, the method comprises the steps of,the normalization process includes the steps of:
s3.1.1: extracting maximum value in noise clutter characteristic data
S3.1.2: let n=1, starting with the 1 st Chirp signal.
S3.1.3: taking noise clutter characteristic data
S3.1.4: according to the normalization formula:calculating a t value;
s3.1.5: the t value is rounded down and stored in an arrayIn (a) and (b);
s3.1.6: judging whether N is greater than or equal to N, if not, making n=n+1, and returning S3.1.3; if yes, outputting all
Preferably, S4 comprises the steps of:
s4.1: acquiring road information data;
s4.2: extracting road name r n { G, S, T }, wherein G represents national road, S represents provincial road, and T represents town road;
extracting vehicle specific position information c l { ak+/-b }, wherein ak represents the central position of the road-end server responsible for the road section length, and b represents the current position of the vehicle from the road-end server;
extracting the vehicle travel direction c d { E, S, W, N }, wherein E, S, W, N represents northeast, northwest;
s4.3: integrating road information data Tag r {r n ,c l ,c d }。
S4.4: integrating road information and noise characteristic data M { Tag ] r ,M r }。
S4.5: and uploading M to a road end server.
Preferably, S5 comprises the steps of:
s5.1: fetch data M { Tag r ,M r }。
S5.2: constructing a road millimeter wave noise model deep learning network model; taking the noise characteristic matrix M r And matrix the noise characteristics M r Inputting the model training data to a road millimeter wave noise model deep learning network model to perform model training;
s5.3: taking road information data Tag r {r n ,c l ,c d And Tag road information data r {r n ,c l ,c d And stored in a server.
Preferably, the road millimeter wave noise model deep learning network model in S5.2 is composed of L layers of subnets, and each layer of subnets comprises a depth feature extraction subnet and a depth feature mapping subnet.
Preferably, the depth feature extraction sub-network comprises a four-layer network structure:
the depth feature extraction sub-network first sub-layer is used for shallow feature coding; the depth feature extraction sub-network second sub-layer is used for shallow feature sampling; a third sub-layer of the depth feature extraction sub-network is used for deep feature coding; a fourth sub-layer of the depth feature extraction sub-network is used for deep feature sampling;
the depth feature mapping subnetwork comprises a five-layer network structure:
the first sub-layer of the depth feature mapping sub-network is used for reconstructing primary deep features; the second sub-layer of the depth feature mapping sub-network is used for primary depth feature recovery; a third sub-layer of the depth feature mapping sub-network is used for secondary depth feature reconstruction; a fourth sub-layer of the depth feature mapping sub-network is used for secondary deep feature recovery; the depth feature map subnet is a fifth sublayer for mapping the depth features to output data.
Preferably, S6 comprises the steps of:
s6.1: initializing depth network parameters: setting the number of layers L of the sub-network of the road millimeter wave noise characteristic deep learning network model and the training iteration number J of the noise characteristic data; initializing weight parameter w 0 And bias parameter b 0 ；
S6.2: matrix the noise characteristics M r Inputting the road millimeter wave noise characteristic deep learning network model to perform one-time iteration unsupervised training;
outputting deep noise characteristics after training of intermediate jth subnetworkWherein j is E [1, L]Calculating the weight w of each layer of sub-network j And bias b j As the input data of the j+1th subnet, continuing training the j+1th subnet of the next layer;
s6.3: noise characteristic matrix M r Inputting the depth features into a first subnet;
the depth feature extraction first sub-network comprises a first feature coding layer, a first sampling layer, a second feature coding layer and a second sampling layer;
first feature encoding layer noise characteristic matrix M r Extracting the primary characteristics to obtain a shallow noise characteristic diagramFirst sampling layer vs. shallow noise characteristic diagram +.>Performing feature sampling to enhance noise features; the second feature coding layer performs feature extraction again on the shallow noise feature map to obtain a deep noise feature map +.>The second sampling layer carries out feature sampling again on the deep noise feature map and outputs a deep noise feature map of enhanced noise features +.>
S6.4: deep noise feature mapInputting the depth features of the first subnet to map the first subnet;
the first subnet depth feature mapping first subnet comprises a feature first reconstruction layer, a first upsampling layer, a feature second reconstruction layer, a second upsampling layer and a feature mapping layer;
feature first reconstruction layer and feature second reconstruction layer versus deep noise feature mapPerforming feature reconstruction twice, performing feature recovery twice on the depth noise feature map by the first upsampling layer and the second upsampling layer, and alternately repeating the feature reconstruction and the feature recoveryAnd (3) row.
S6.5: the feature mapping layer will reconstruct the recovered deep noise featuresAccording to the size of the set characteristic diagram, mapping the characteristic of deep noise outputted by the jth sub-network +.>Calculating the weight w of the first subnet 1 And bias b 1 Outputting to the next layer of sub-network;
s6.6: judging whether j is greater than or equal to L, if so, executing the step S6.8, and if not, executing the step S67;
s6.7: inputting data into the subnet layer j+1, and returning to S6.3;
s6.8: taking noise characteristic data stored in a road side server as a label to perform supervised training, and adjusting network parameters of each layer of sub-network; error e recovery based on noise characteristics of each layer of subnetwork j The difference adjusts the weight w of each layer of sub-network j And bias b j ；
S6.9: determining a reconstruction error e j Whether or not to be less than or equal to the minimum reconstruction error e jmin If yes, executing S6.10, and if not, executing S6.3;
s6.10: determining the optimal network layer number structure and the optimal parameters of each layer; training completely and outputting noise characteristics
Compared with the prior art, the technical scheme of the invention has the beneficial effects that:
according to the invention, the complex noise model is obtained by obtaining the noise characteristic parameters under different road conditions and uploading the noise characteristic parameters to the road end server and applying deep learning in the road end server.
The complex road millimeter wave noise model acquired by the road side server can be provided for vehicles passing through the road, the vehicle acquires the road millimeter wave noise model parameters in advance, the vehicle-mounted radar detection algorithm is updated, the sensitivity and the accuracy of the vehicle-mounted radar detection target are improved, and the safety in the driving process is further enhanced.
By applying the method disclosed by the invention, the vehicle-mounted radar can more accurately detect the position, the speed and the angle of the vehicle on the road in front, and meanwhile, the vehicle-mounted system can remind a driver to slow down in advance, so that the riding comfort of passengers in the vehicle is improved.
Drawings
Fig. 1 is a flowchart of a road millimeter wave noise model detection estimation method based on deep learning in embodiment 1.
Fig. 2 is a flowchart of a method of extracting road clutter characteristics.
FIG. 3 is a flow chart for generating a normalized clutter characteristic dataset.
Fig. 4 is a flowchart of a noise clutter characteristic normalization algorithm.
Fig. 5 is a flowchart of adding road information marks.
Fig. 6 is a flow chart of a data processing by the roadside server.
Fig. 7 is a diagram of a road millimeter wave noise feature deep learning network model structure.
Fig. 8 is a flowchart of training a road millimeter wave noise model.
Detailed Description
The drawings are for illustrative purposes only and are not to be construed as limiting the present patent;
for the purpose of better illustrating the embodiments, certain elements of the drawings may be omitted, enlarged or reduced and do not represent the actual product dimensions;
it will be appreciated by those skilled in the art that certain well-known structures in the drawings and descriptions thereof may be omitted.
The technical scheme of the invention is further described below with reference to the accompanying drawings and examples.
Example 1:
example 1
The embodiment provides a road millimeter wave noise model detection estimation method based on deep learning, as shown in fig. 1, the method comprises the following steps:
s1: preprocessing millimeter wave radar echo signals of R road sections; the preprocessing process comprises the steps of mixing a received signal with a radar local oscillation signal to obtain an intermediate frequency signal, filtering the intermediate frequency signal, and digitizing the filtered intermediate frequency signal through AD conversion;
s2: performing time-frequency conversion on the digitized intermediate frequency signal, and extracting road noise clutter characteristics;
s3: processing the noise clutter characteristics to generate a normalized noise clutter characteristic data set;
s4: acquiring road information data, and integrating the road information data and noise characteristic data;
s5: constructing a road millimeter wave noise deep learning network model, and carrying out data processing on road information data and noise characteristic data;
s6: and deeply learning and training a road millimeter wave noise deep learning network model to obtain optimal network model parameters, and detecting and estimating the noise model according to the model parameters.
According to the method, the millimeter wave noise model of the road is extracted, so that the vehicle-mounted radar can obtain the environmental noise parameters of the current road in advance, and corresponding algorithm processing is performed before the vehicle enters the specific road condition, so that the target detection sensitivity of the vehicle is improved, the running safety performance of the vehicle is improved, the occurrence of traffic accidents is reduced, and the comfort of passengers in the driving process is improved.
Example 2:
the road millimeter wave noise model detection estimation method based on deep learning provided by the embodiment is consistent with embodiment 1, and only each step is further limited.
Step S1: and preprocessing millimeter wave radar echo signals of R road sections.
Step S2: and performing time-frequency conversion on the digitized intermediate frequency signal to extract the noise clutter characteristic of the road.
Step S3: the noise clutter characteristics are processed to generate a normalized noise clutter characteristic dataset.
Step S4: and adding the road coordinate information and uploading the road coordinate information to a road end server.
Step S5: the roadside server processes the data.
Step S6: the road end server deep learning network trains parameters of the road millimeter wave noise model.
As shown in fig. 2, the step 2 of extracting the road noise clutter characteristics specifically includes the following steps:
step S21: and taking N Chirp signals in total of radar echo intermediate frequency signals of the r-th road section.
Step S22, taking an nth Chirp signal, wherein the value range of N is 1 to N, which represents a total of N Chirp signals.
Step S23: and performing time-frequency conversion on the nth Chirp radar echo intermediate frequency signal.
Step S24: extracting road noise clutter characteristics from the obtained frequency domain signals; the road noise clutter characteristic data are respectively: noise clutter characteristics i: average, very poor, median; noise clutter characteristics ii: variance, standard deviation, mean deviation; noise clutter characteristics iii: relative mean deviation, relative standard deviation, median absolute deviation.
Step S25: storing noise clutter characteristic data and arrayWherein, the array subscript R represents the R-th road section, the value of R is 1 to R, and the R-th road section is totally represented; the values of the array superscript i are 1 to 9, which represent 9 noise clutter characteristic data.
Step S26: judging whether N is greater than or equal to N, if not, executing step S27; if yes, go to step S28.
Step S27: index value n=n+1.
Step S28: output N Chirp frequency domain information noise clutter characteristic arrays
The step S24 extracts the noise clutter characteristic data of the frequency domain signal and the step S25 stores the noise clutter characteristic in an arrayThe method comprises the following steps:
step S241: calculating noise clutter characteristic I data from frequency domain signal data, and averaging:extremely bad: r=x max -X min The method comprises the steps of carrying out a first treatment on the surface of the Median: z=media (X i ) The method comprises the steps of carrying out a first treatment on the surface of the Respectively store in the array->
Step S242: calculating noise clutter characteristic II data from the frequency domain signal and the noise clutter characteristic I data, and variance:standard deviation: />Mean deviation:>respectively store in the array->
Step S243: calculating noise clutter characteristic III data from the frequency domain signal and the noise clutter characteristic I data and the noise clutter characteristic II data, and relatively average deviation:relative standard deviation: />Median absolute deviation: mad=media i (|X i -median j (X j ) |) is provided; respectively store in the array->
As shown in fig. 3, the step S3 further processes the noise clutter characteristics to generate a normalized noise clutter characteristic data set. The method comprises the following steps:
step S31: taking the array obtained in the previous step
Step S32: taking outAnd (5) carrying out normalization processing.
Step S33: judging whether i is greater than or equal to 9, if not, executing step S311; if yes, go to step S34. Step S34: from normalized noise clutter characteristic dataMatrix m forming 3×3 n . Where n represents the nth Chirp signal.
Step S35: judging whether N is greater than or equal to N, if not, executing step S310; if yes, go to step S36. After this step is completed, N matrixes m are obtained n 。
Step S36: from matrix m n Noise clutter characteristic matrix M with composition length of sxs r . Wherein the value of s isM r The subscript r of (2) denotes the r-th road segment.
Step S37: judging whether R is greater than or equal to R, if not, executing step S39; if yes, go to step S38.
Step S38: output noise characteristic matrix M r 。
Step S39: index value r=r+1.
Step S310: index value n=n+1.
Step S311: index value i=i+1.
As shown in the figure4, the step S32 is takenAnd (5) carrying out normalization processing. The method comprises the following steps:
step S321: extracting maximum value in noise clutter characteristic data
Step S322: the index value is assigned n=1, and normalization starts from the 1 st Chirp signal.
Step S323: taking noise clutter characteristic data
Step S324: according to the normalization formula:and calculating a t value.
Step S325: the t value is rounded down and stored in an arrayIs a kind of medium.
Step S326: judging whether N is greater than or equal to N, if not, executing the index value n=n+1 in step S328; if yes, go to step S327.
Step S327: output all
Step S328: index value n=n+1.
As shown in fig. 5, step S4 adds road coordinate information and uploads the information to the road side server, and includes the following steps:
step S41, obtaining road information data.
Step S42: extracting road name r n { G, S, T }, wherein G represents national road, S represents provincial road, and T represents town road. Extracting vehicle specific position information c l { ak.+ -. B }, wherein ak represents the roadside serverAnd b represents the current position of the vehicle from the road end server. Extracting the vehicle travel direction c d { E, S, W, N }, wherein E, S, W, N represents northeast, northwest.
Step S43 of integrating road information data Tag r {r n ,c l ,c d }。
Step S44, integrating road information and noise characteristic data M { Tag ] r ,M r }。
And step S45, uploading the M to a road end server.
As shown in fig. 6, the step S5 of the path end server processing data includes the following steps:
step S51, data M { Tag { uploaded by the vehicle is obtained r ,M r }。
Step S52, taking the noise characteristic matrix M r 。
Step S53, taking the road information data Tag r {r n ,c l ,c d }。
Step S54 noise characteristic matrix M r And inputting the model training data to a deep learning network for model training.
Step S55, road information data Tag r {r n ,c l ,c d And stored in a server.
As shown in FIG. 7, the step S54 is a noise characteristic matrix M r And inputting the model training data to a deep learning network for model training. The road millimeter wave noise model feature deep learning network has the following structure: the road millimeter wave noise feature deep learning network model is composed of L layers of subnets, and each layer of subnets comprises a depth feature extraction subnet and a depth feature mapping subnet.
The depth feature extraction subnetwork comprises a four-layer network structure:
the first sub-layer of the sub-network is extracted by the depth features, and the main function is shallow feature coding; the second sub-layer of the depth feature extraction sub-network has the main function of shallow feature sampling; the third sub-layer of the depth feature extraction sub-network has the main function of depth feature coding. The fourth sub-layer of the depth feature extraction sub-network has the main function of depth feature sampling. For example, the depth feature extraction subnet may be set to: the system comprises a first convolution active layer, a first sampling layer, a second convolution active layer and a second sampling layer.
The depth feature mapping subnetwork comprises a five-layer network structure:
the first sub-layer of the depth feature mapping sub-network has the main function of primary depth feature reconstruction; the second sub-layer of the depth feature mapping sub-network has the main function of primary depth feature recovery; the third sub-layer of the depth feature mapping sub-network has the main function of secondary depth feature reconstruction; the depth feature mapping sub-network is a fourth sub-layer, and the main function is secondary deep feature recovery; the depth feature map subnet is a fifth sublayer and has the main function of mapping the depth features into output data. For example, the depth feature mapping subnet may be set to: a first deconvolution activation layer, a first upsampling layer, a second deconvolution activation layer, a second upsampling layer, and a third deconvolution layer.
As shown in fig. 8, the step S6 is to train the parameters of the road millimeter wave noise model by using the road millimeter wave noise feature deep learning network model. The method comprises the following steps:
step S61, initializing the depth network parameters. Setting the number of layers L of the sub-network of the road millimeter wave noise characteristic deep learning network model and the training iteration number J of the noise characteristic data; initializing weight parameter w 0 And bias parameter b 0 。
Step S62 noise characteristic matrix M r And inputting the data to a road millimeter wave noise characteristic deep learning network model to perform one-time iterative unsupervised training. Outputting deep noise characteristic diagram after training of intermediate j-th subnetworkWherein j is E [1, L]Calculating the weight w of each layer of sub-network j And bias b j As input data of the j+1th subnet, training is continued for the next layer of the j+1th subnet.
Step S63 noise characteristic matrix M r (28×28×1) input to the depth feature extraction first subnet. Convolutional active layer 1 pair noise feature matrix M r (28×28×1) performing primary feature extraction to obtain shallow noise feature map(28X 16). First sampling layer vs. shallow noise characteristic diagram +.>(28×28×16) feature sampling to enhance shallow noise feature +.>(14X 16). The convolution activation layer 2 performs feature extraction again on the shallow noise feature map to obtain a deep noise feature map(14X 8). The second sampling layer carries out feature sampling again on the deep noise feature map and outputs a deep noise feature map of enhanced noise features +.>(7X 8). Wherein the convolutional layer output may be represented by the following formula:w is a convolution kernel matrix which is successfully initialized, b is a bias parameter matrix matched with the convolution kernel, BN (·) is a batch normalization function, sigma (·) is an activation function, and h (M) r ) For the convolution kernel w and the input M r And (5) a characteristic diagram output after convolution.
Step S64 deep noise feature map(7 x 8) input to the first subnet depth feature map the first subnet. Deconvolution activation layer 1 vs deep noise profile +.>(7×7×8) performing primary reconstruction to obtain deep noise characteristic reconstruction pattern(7×7×8), first upsampling layerReconstruction of deep noise features>(7×7× 8) performing feature recovery to obtain deep noise feature recovery map +.>(14X 8). Deconvolution activation layer 2 restoration map for deep noise characteristics +.>(14×14×8) performing reconstruction to obtain deep noise feature reconstruction +.>(14×14×16) the second upsampling layer reconstruct the deep noise features +.>(14×14×16) performing feature recovery to obtain deep noise feature recovery map +.>(28×28×16)。
Step S65, the feature mapping layer restores deep noise feature to the image(28×28×16) deep noise feature of mapped single channel +.>(28X 1). Wherein the deconvolution layer can be represented by the following formula: /> An image is reconstructed for the feature map. Calculating the weight w of the first subnet 1 And bias b 1 And outputting to the next layer of sub-network.
Step S66, judging whether j is greater than or equal to L, if so, executing step S68, and if not, executing step S67.
And step S67, inputting data into the subnet layer j+1.
Step S68: taking noise characteristic data stored in a road side server as a label to perform supervised training, and fine-tuning network parameters of each layer of sub-network. Error e recovery based on noise characteristics of each layer of subnetwork j Differential fine tuning weight w of each layer of subnetwork j And bias b j 。
Step S69, judging the reconstruction error e j Whether or not to be greater than or equal to the minimum reconstruction error e jmjn If yes, step S610 is executed, and if no, step S63 is executed.
Step S610: and determining the optimal network layer structure and the optimal parameters of each layer. Training completely and outputting noise characteristics
The embodiment can detect noise model parameters of various road environments. The vehicles passing through the road section can download the road millimeter wave noise model parameters from the road end server, and the road environment and the noise level of the current running road are perceived through algorithm processing. The vehicle-mounted system updates a vehicle-mounted radar detection algorithm according to the road millimeter wave noise model, so that the overall anti-interference performance of the vehicle-mounted radar system is improved, the omission ratio is reduced, the identification accuracy of radar detection targets is improved, and traffic accidents are reduced. According to the embodiment, on the premise of improving the running safety of the vehicle, a driver can make sufficient pre-judgment on the condition of the road ahead, the speed of the vehicle is eased and controlled, and the driving comfort of passengers in the vehicle is improved.
The terms describing the positional relationship in the drawings are merely illustrative, and are not to be construed as limiting the present patent;
it is to be understood that the above examples of the present invention are provided by way of illustration only and not by way of limitation of the embodiments of the present invention. Other variations or modifications of the above teachings will be apparent to those of ordinary skill in the art. It is not necessary here nor is it exhaustive of all embodiments. Any modification, equivalent replacement, improvement, etc. which come within the spirit and principles of the invention are desired to be protected by the following claims.
Claims (8)
1. The road millimeter wave noise model detection estimation method based on deep learning is characterized by comprising the following steps of:
s1: preprocessing millimeter wave radar echo signals of R road sections; the preprocessing process comprises the steps of mixing a received signal with a radar local oscillation signal to obtain an intermediate frequency signal, filtering the intermediate frequency signal, and digitizing the filtered intermediate frequency signal through AD conversion;
s2: performing time-frequency conversion on the digitized intermediate frequency signal, and extracting road noise clutter characteristics;
the method comprises the following steps:
s2.1: taking a radar echo intermediate frequency signal of an R-th road section from the preprocessed millimeter wave radar echo signals of the R road sections;
s2.2, taking an nth Chirp signal, wherein the value range of N is 1 to N, and N represents the number of all Chirp signals in the radar echo intermediate frequency signal of the nth road section;
s2.3: performing time-frequency conversion on the nth Chirp radar echo intermediate frequency signal; obtaining a frequency domain signal;
s2.4: extracting road noise clutter characteristics from the obtained frequency domain signals; and store noise clutter characteristic data in an arrayWherein, the array subscript R represents the R-th road section, and the value of R is 1 to R; the value of the upper index i of the array is 1 to 9, and the upper index i represents 9 noise clutter characteristic data;
the road noise clutter characteristic data are respectively as follows: noise clutter characteristics I; noise clutter characteristics ii; noise clutter characteristics III;
s2.5: judging whether N is greater than or equal to N, if not, making n=n+1, and returning to S2.2; if yes, executing the step S2.6;
s2.6: output N Chirp frequency domain information noise clutter characteristic arrays
The step S2.4 comprises the following steps:
s2.4.1: noise clutter characteristic i data are calculated from frequency domain signal data:
calculating a frequency domain signal mean value:store in array->
Calculating the frequency domain signal range: r=x max -x min Stored in array
Calculating the median of the frequency domain signal: z=media (X i ) The method comprises the steps of carrying out a first treatment on the surface of the Stored array
Step S2.4.2: calculating noise clutter characteristic II data from the frequency domain signal and the noise clutter characteristic I data:
calculating frequency domain signal variance:store in array->
Calculating standard deviation of the frequency domain signals:store in array->
Calculating the average deviation of the frequency domain signals:store in array->
Step S2.4.3: calculating noise clutter characteristic III data from the frequency domain signal and the noise clutter characteristic I data and the noise clutter characteristic II data:
calculating the relative average deviation of the frequency domain signals:store in array->
Calculating the relative standard deviation of the frequency domain signals:store in array->
Calculating the absolute deviation of the median of the frequency domain signal: mad=media (|x) i -median(X i ) I), stored in an array
S3: processing the noise clutter characteristics to generate a normalized noise clutter characteristic data set;
s4: acquiring road information data, and integrating the road information data and noise characteristic data;
s5: the road end server builds a road millimeter wave noise deep learning network model, and performs data processing on road information data and noise characteristic data;
s6: and deeply learning and training a road millimeter wave noise deep learning network model to obtain optimal network model parameters, and detecting and estimating the noise model according to the model parameters.
2. The road millimeter wave noise model detection estimation method based on deep learning as claimed in claim 1, wherein S3 comprises the steps of:
s3.1: the array from S2In the recipe, get->Normalization processing is carried out to obtain normalized noise clutter characteristic data +.>
S3.2: judging whether i is greater than or equal to 9, if not, making i=i+1 and returning to S3.1; if yes, executing S3.3;
s3.3: from normalized noise clutter characteristic dataMatrix m forming 3×3 n The method comprises the steps of carrying out a first treatment on the surface of the Wherein n represents an nth Chirp signal;
s3.4: judging whether N is greater than or equal to N, if not, making n=n+1, and returning to S3.3; if yes, executing step S3.5; after this step is completed, N matrixes m are obtained n ；
S3.5: from matrix m n Noise clutter characteristic matrix M with composition length of sxs r The method comprises the steps of carrying out a first treatment on the surface of the Wherein the value of s isM r The subscript r of (2) represents the r-th road segment;
s3.6: judging whether R is greater than or equal to R, if not, making r=r+1, and returning to S3.1; if so, outputting a noise characteristic matrix M r I.e. noise clutter characteristic data sets.
3. The road millimeter wave noise model detection estimation method based on deep learning according to claim 2, wherein,the normalization process includes the steps of:
s3.1.1: extracting maximum value in noise clutter characteristic data
S3.1.2: let n=1, normalize starting from the 1 st Chirp signal;
s3.1.3: acquiring noise clutter characteristic data
S3.1.4: according to the normalization formula:calculating a t value;
s3.1.5: the t value is rounded down and stored in an arrayIn (a) and (b);
s3.1.6: judging whether N is greater than or equal to N, if not, making n=n+1, and returning S3.1.3; if yes, outputting all
4. The road millimeter wave noise model detection estimation method based on deep learning as claimed in claim 3, wherein S4 comprises the steps of:
s4.1: acquiring road information data;
s4.2: extracting road name r n { G, S, T }, wherein G represents national road, S represents provincial road, and T represents town road;
extracting vehicle specific position information c l { ak+/-b }, wherein ak represents the central position of the road-end server responsible for the road section length, and b represents the current position of the vehicle from the road-end server;
extracting the vehicle travel direction c d { E, S, W, N }, wherein E, S, W, N represents northeast, northwest;
s4.3: integrating road information data Tag r {r n ,c l ,c d }；
S4.4: integrating road information and noise characteristic data M { Tag ] r ,M r }；
S4.5: and uploading M to a road end server.
5. The road millimeter wave noise model detection estimation method based on deep learning as claimed in claim 4, wherein S5 comprises the steps of:
s5.1: acquiring road information and noise characteristic data M { Tag ] r ,M r }；
S5.2: constructing a road millimeter wave noise model deep learning network model; acquiring a noise characteristic matrix M r And matrix the noise characteristics M r Inputting the model training data to a road millimeter wave noise model deep learning network model to perform model training;
s5.3: acquiring road information data Tag r {r n ,c l ,c d And Tag road information data r {r n ,c l ,c d And stored in a server.
6. The road millimeter wave noise model detection estimation method based on deep learning according to claim 5, wherein the road millimeter wave noise model deep learning network model of S5.2 is composed of L layers of subnets, and each layer of subnets comprises a depth feature extraction subnets and a depth feature mapping subnets.
7. The road millimeter wave noise model detection estimation method based on deep learning of claim 6, wherein the deep feature extraction subnet comprises a four-layer network structure:
the depth feature extraction sub-network first sub-layer is used for shallow feature coding; the depth feature extraction sub-network second sub-layer is used for shallow feature sampling; a third sub-layer of the depth feature extraction sub-network is used for deep feature coding; a fourth sub-layer of the depth feature extraction sub-network is used for deep feature sampling;
the depth feature mapping subnetwork comprises a five-layer network structure:
the first sub-layer of the depth feature mapping sub-network is used for reconstructing primary deep features; the second sub-layer of the depth feature mapping sub-network is used for primary depth feature recovery; a third sub-layer of the depth feature mapping sub-network is used for secondary depth feature reconstruction; a fourth sub-layer of the depth feature mapping sub-network is used for secondary deep feature recovery; the depth feature map subnet is a fifth sublayer for mapping the depth features to output data.
8. The road millimeter wave noise model detection estimation method based on deep learning as claimed in claim 7, wherein S6 comprises the steps of:
s6.1: initializing depth network parameters: setting the number of layers L of the sub-network of the road millimeter wave noise characteristic deep learning network model and the training iteration number J of the noise characteristic data; initializing weight parameter w 0 And bias parameter b 0 ；
S6.2: matrix the noise characteristics M r Inputting the road millimeter wave noise characteristic deep learning network model to perform one-time iteration unsupervised training;
outputting deep noise characteristics after training of intermediate jth subnetworkWherein j is E [1, L]Calculating the sub-network of each layerWeight w j And bias b j As the input data of the j+1th subnet, continuing training the j+1th subnet of the next layer;
s6.3: noise characteristic matrix M r Inputting the depth features into a first subnet;
the depth feature extraction first sub-network comprises a first feature coding layer, a first sampling layer, a second feature coding layer and a second sampling layer;
first feature encoding layer noise characteristic matrix M r Extracting the primary characteristics to obtain a shallow noise characteristic diagramFirst sampling layer vs. shallow noise characteristic diagram +.>Feature sampling is carried out, noise features are enhanced, and a shallow enhanced noise feature map is obtainedSecond feature coding layer +.>Extracting features again to obtain deep noise feature map->Second sample layer versus deep noise profile +.>Performing feature sampling again to output deep enhanced noise feature map of enhanced noise feature>
S6.4: deep enhanced noise feature mapInputting the depth features of the first subnet to map the first subnet;
the first subnet depth feature mapping first subnet comprises a feature first reconstruction layer, a first upsampling layer, a feature second reconstruction layer, a second upsampling layer and a feature mapping layer;
feature first reconstruction layer and feature second reconstruction layer versus deep enhanced noise feature mapPerforming feature reconstruction twice, wherein the first upsampling layer and the second upsampling layer are respectively used for deep enhanced noise feature map +.>Performing feature recovery twice, and alternately repeating feature reconstruction and feature recovery to obtain deep enhanced noise feature map (I) with reconstructed recovery>
S6.5: the feature mapping layer reconstructs the restored deep enhanced noise feature mapAccording to the size of the set characteristic diagram, mapping the characteristic of deep noise outputted by the jth sub-network +.>Calculating the weight w of the first subnet 1 And bias b 1 Outputting to the next layer of sub-network;
s6.6: judging whether j is greater than or equal to L, if so, executing the step S6.8, and if not, executing the step S67;
s6.7: inputting data into the subnet layer j+1, and returning to S6.3;
s6.8: acquiring noise characteristic data stored in a road-end server as a label to perform supervised training and adjusting each layer of sub-Network parameters; error e recovery based on noise characteristics of each layer of subnetwork j The difference adjusts the weight w of each layer of sub-network j And bias b j ；
S6.9: determining a reconstruction error e j Whether or not to be less than or equal to the minimum reconstruction error e jmin If yes, executing S6.10, and if not, executing S6.3;
s6.10: determining the optimal network layer number structure and the optimal parameters of each layer; training completely and outputting noise characteristics
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010859256.3A CN112099014B (en) | 2020-08-24 | 2020-08-24 | Road millimeter wave noise model detection estimation method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010859256.3A CN112099014B (en) | 2020-08-24 | 2020-08-24 | Road millimeter wave noise model detection estimation method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112099014A CN112099014A (en) | 2020-12-18 |
| CN112099014B true CN112099014B (en) | 2023-08-22 |
Family
ID=73754366
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010859256.3A Active CN112099014B (en) | 2020-08-24 | 2020-08-24 | Road millimeter wave noise model detection estimation method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112099014B (en) |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108169745A (en) * | 2017-12-18 | 2018-06-15 | 电子科技大学 | A kind of borehole radar target identification method based on convolutional neural networks |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108226892A (en) * | 2018-03-27 | 2018-06-29 | 天津大学 | A kind of radar signal restoration methods under complicated noise based on deep learning |
| CN108599765A (en) * | 2018-04-14 | 2018-09-28 | 上海交通大学 | The device and method of the noise suppressed distortion correction of analog-digital converter based on deep learning |
| CN109389058A (en) * | 2018-09-25 | 2019-02-26 | 中国人民解放军海军航空大学 | Sea clutter and noise signal classification method and system |
| DE102018219255A1 (en) * | 2018-11-12 | 2020-05-14 | Zf Friedrichshafen Ag | Training system, data set, training method, evaluation device and deployment system for a road vehicle for recording and classifying traffic noise |
| WO2020139355A1 (en) * | 2018-12-27 | 2020-07-02 | Didi Research America, Llc | System for automated lane marking |
| CN111507233A (en) * | 2020-04-13 | 2020-08-07 | 吉林大学 | Multi-mode information fusion intelligent vehicle pavement type identification method |
-
2020
- 2020-08-24 CN CN202010859256.3A patent/CN112099014B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108169745A (en) * | 2017-12-18 | 2018-06-15 | 电子科技大学 | A kind of borehole radar target identification method based on convolutional neural networks |
| CN108229404A (en) * | 2018-01-09 | 2018-06-29 | 东南大学 | A kind of radar echo signal target identification method based on deep learning |
| CN108226892A (en) * | 2018-03-27 | 2018-06-29 | 天津大学 | A kind of radar signal restoration methods under complicated noise based on deep learning |
| CN108599765A (en) * | 2018-04-14 | 2018-09-28 | 上海交通大学 | The device and method of the noise suppressed distortion correction of analog-digital converter based on deep learning |
| CN109389058A (en) * | 2018-09-25 | 2019-02-26 | 中国人民解放军海军航空大学 | Sea clutter and noise signal classification method and system |
| DE102018219255A1 (en) * | 2018-11-12 | 2020-05-14 | Zf Friedrichshafen Ag | Training system, data set, training method, evaluation device and deployment system for a road vehicle for recording and classifying traffic noise |
| WO2020139355A1 (en) * | 2018-12-27 | 2020-07-02 | Didi Research America, Llc | System for automated lane marking |
| CN111507233A (en) * | 2020-04-13 | 2020-08-07 | 吉林大学 | Multi-mode information fusion intelligent vehicle pavement type identification method |
Non-Patent Citations (1)
| Title |
|---|
| Millimeter Wave Radar Target Tracking Based on Adaptive Kalman Filter road millimeter wave radar noise-model deep learning;Guangyao Zhai等;《2018 IEEE Intelligent Vehicles Symposium (IV)》;第453-458页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112099014A (en) | 2020-12-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN106529493B (en) | Robust multi-lane line detection method based on perspective view | |
| DE102017126877B4 (en) | motor vehicle | |
| DE102007023888B4 (en) | Device and method for pedestrian recognition | |
| CN111521989A (en) | Deep learning for super-resolution in radar systems | |
| EP1290467A1 (en) | Track map generator | |
| CN110596696B (en) | Apparatus and method for improved radar beamforming | |
| US11783452B2 (en) | Traffic monitoring using distributed fiber optic sensing | |
| CN111126332B (en) | Frequency hopping signal classification method based on contour features | |
| CN111598142A (en) | Outdoor terrain classification method for wheeled mobile robot | |
| CN110703272B (en) | Surrounding target vehicle state estimation method based on vehicle-to-vehicle communication and GMPHD filtering | |
| CN112990317B (en) | Weak and small target detection method | |
| CN112099014B (en) | Road millimeter wave noise model detection estimation method based on deep learning | |
| CN209214563U (en) | Multi-information fusion spectrum of road surface roughness real-time testing system | |
| CN109977908B (en) | Vehicle driving lane detection method based on deep learning | |
| CN114218999A (en) | Millimeter wave radar target detection method and system based on fusion image characteristics | |
| CN117471457A (en) | Sparse SAR learning imaging method, device and medium based on deep expansion complex network | |
| CN108564585B (en) | Image change detection method based on self-organizing mapping and deep neural network | |
| CN101561879A (en) | Curvelet expression-based method for image underdetermined blind source separation | |
| CN115775378A (en) | Vehicle-road cooperative target detection method based on multi-sensor fusion | |
| Bystrov et al. | Automotive surface identification system | |
| CN113848532A (en) | FMCW radar signal noise reduction system and method based on noise reduction model | |
| CN109991657B (en) | Seismic data high-resolution processing method based on inverse binary recursive singular value decomposition | |
| DE102020132159A1 (en) | Method for training an autoencoder to reduce interference in a radar system | |
| Bilik et al. | Target classification using Gaussian mixture model for ground surveillance Doppler radar | |
| CN116109931B (en) | Automatic urban ground subsidence recognition and classification method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| TR01 | Transfer of patent right |
Effective date of registration: 20240410 Address after: Room 201, Building A, No. 318 Outer Ring West Road, University City, Panyu District, Guangzhou City, Guangdong Province, 510000 Patentee after: Guangzhou University Town (Guangong) Science and Technology Achievement Transformation Center Country or region after: China Address before: 510062 Dongfeng East Road, Yuexiu District, Guangzhou, Guangdong 729 Patentee before: GUANGDONG University OF TECHNOLOGY Country or region before: China |