A scene generalization method, system, terminal and medium for millimeter wave radar model

Abstract

The present invention discloses a method, system, terminal and medium for scene generalization of a millimeter wave radar model, and relates to the field of scene generalization technology. The method comprises: establishing an original radar model, generating a pre-training data set according to the original radar model, and performing model training on the created modified training network model according to the pre-training data set to obtain a modified radar model; combining the original radar model with the modified radar model to obtain a generalized radar model; obtaining the scene data to be generalized of the target scene, inputting the scene data to be generalized into the generalized radar model, and outputting the corresponding scene generalization result. The present invention modifies the result of the original model by using the modified radar model without changing the structure of the original model, thereby reducing the cost of training, thereby reducing the amount of data required for training when training the radar model, greatly improving the efficiency of model training, and reducing the demand for new scene data.

Description

Technical Field

[0001] The present invention relates to the field of scene generalization technology, and in particular to a scene generalization method, system, terminal and computer-readable storage medium for a millimeter-wave radar model. Background Art

[0002] The features between scenes have similar common features and large differences. When the model is trained with data from a certain scene and used in other scenes, the difference between the original scene and the new scene data leads to large errors in the prediction results of the model on the new scene. Data-driven radar modeling is a relatively complex application scenario, and there is currently a lack of research on improving the generalization ability of data-driven radar models.

[0003] At present, for this situation, the transfer learning method can be used to adjust the difference features between scenes to adapt to the new scene while retaining the original model's fitting effect on the common features between scenes. However, the existing transfer learning is to adjust part of the model structure, which is easy to retain the undesirable structure in the original model, thereby introducing the error of the original model into the existing model, resulting in inaccurate generalization of the new scene, and the amount of data required during training is large, resulting in low efficiency in generalizing the new scene. In addition, the training process of this method is relatively complicated, which affects the application efficiency in large-scale industrial development.

[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0005] The main purpose of the present invention is to provide a scene generalization method, system, terminal and medium for a millimeter-wave radar model, aiming to solve the problem that transfer learning in the prior art is to adjust part of the structure of the model, which easily retains the undesirable structure in the original model, thereby introducing the error of the original model into the existing model, resulting in inaccurate generalization of new scenes, and the amount of data required during training is large, resulting in low efficiency in generalizing new scenes.

[0006] To achieve the above object, the present invention provides a scene generalization method of a millimeter wave radar model, and the scene generalization method of the millimeter wave radar model comprises the following steps:

[0007] An original radar model is established, a pre-training data set is generated according to the original radar model, and a modified training network model is trained according to the pre-training data set to obtain a modified radar model;

[0008] Combining the original radar model with the modified radar model to obtain a generalized radar model;

[0009] The to-be-generalized scene data of the target scene is obtained, the to-be-generalized scene data is input into the generalized radar model, and a corresponding scene generalization result is output.

[0010] Optionally, in the scenario generalization method of the millimeter wave radar model, the step of establishing the original radar model specifically includes:

[0011] Acquire multiple original scene data of the original scene, and perform truth value extraction on all the original scene data to obtain multiple groups of scene target truth values;

[0012] Creating a radar training network model, inputting a first set of scene target true values ​​into the radar training network model, performing radar data prediction on the first set of scene target true values, and obtaining a first radar data prediction result;

[0013] Calculating a loss value for the first radar data prediction result and the real radar data corresponding to the first set of scene target true values ​​to obtain a first loss value, and adjusting the parameters of the radar training network model according to the first loss value;

[0014] The next set of scene target true values ​​is input into the radar training network model until the training status of the radar training network model meets the first preset condition, thereby obtaining the original radar model.

[0015] Optionally, the scenario generalization method of the millimeter wave radar model, wherein the generating a pre-training data set according to the original radar model specifically includes:

[0016] Acquire multiple scene data of the new scene, input all the scene data into the original radar model, and output corresponding second radar data prediction results;

[0017] A plurality of scene parameters of the new scene are obtained, and all the scene parameters are combined with the second radar data prediction result to obtain a pre-training data set.

[0018] Optionally, the scenario generalization method of the millimeter wave radar model, wherein the model training is performed on the modified training network model created according to the pre-training data set to obtain the modified radar model, specifically comprising:

[0019] Creating a revised training network model, inputting a first group of training samples of the pre-training data set into the revised training network model, performing radar data prediction on the first group of training samples, and obtaining a third radar data prediction result;

[0020] Calculating the loss value of the third radar data prediction result and the real radar data corresponding to the first group of training samples to obtain a second loss value, and adjusting the parameters of the original radar model according to the second loss value;

[0021] The first set of training samples is input into the modified training network model until the training condition of the modified training network model meets the second preset condition, thereby obtaining a modified radar model.

[0022] Optionally, the scene generalization method of the millimeter wave radar model, wherein the step of obtaining the scene data to be generalized of the target scene, inputting the scene data to be generalized into the generalized radar model, and outputting the corresponding scene generalization result, specifically includes:

[0023] Acquire the to-be-generalized scene data of the target scene, and perform truth value extraction on the to-be-generalized scene data to obtain the target truth value of the new scene;

[0024] The new scene target true value is input into the generalized radar model, the corresponding radar data result is output, and the radar data of the radar data result is scene generalized to obtain the corresponding scene generalization result.

[0025] Optionally, the scene generalization method of the millimeter wave radar model, wherein the step of inputting the true value of the new scene target into the generalized radar model and outputting the corresponding radar data result specifically includes:

[0026] The new scene target true value is input into the generalized radar model, and the original radar model of the generalized radar model performs radar data prediction on the new scene target true value to obtain a fourth radar data prediction result;

[0027] The radar prediction data of the fourth radar data prediction result is obtained, the radar prediction data and the target scene parameter of the target scene are input into the modified radar model of the generalized radar model, and the corresponding radar data result is output.

[0028] Optionally, in the scene generalization method of the millimeter wave radar model, the scene target true value includes a heading angle, a lateral velocity, a longitudinal velocity, a lateral distance and a longitudinal distance of the target object.

[0029] Optionally, in the scenario generalization method of the millimeter wave radar model, the scenario generalization system of the millimeter wave radar model includes:

[0030] A model training module is used to establish an original radar model, generate a pre-training data set according to the original radar model, and perform model training on the created revised training network model according to the pre-training data set to obtain a revised radar model;

[0031] A model combination module, used for combining the original radar model with the modified radar model to obtain a generalized radar model;

[0032] The scene generalization module is used to obtain the scene data to be generalized of the target scene, input the scene data to be generalized into the generalized radar model, and output the corresponding scene generalization result.

[0033] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a scene generalization program of a millimeter-wave radar model stored in the memory and executable on the processor, and when the scene generalization program of the millimeter-wave radar model is executed by the processor, the steps of the scene generalization method of the millimeter-wave radar model as described above are implemented.

[0034] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a scene generalization program of a millimeter-wave radar model, and when the scene generalization program of the millimeter-wave radar model is executed by a processor, the steps of the scene generalization method of the millimeter-wave radar model as described above are implemented.

[0035] In the present invention, an original radar model is established, a pre-training data set is generated according to the original radar model, and a modified training network model is trained according to the pre-training data set to obtain a modified radar model; the original radar model and the modified radar model are combined and processed to obtain a generalized radar model; the scene data to be generalized of the target scene is obtained, the scene data to be generalized is input into the generalized radar model, and the corresponding scene generalization result is output. The present invention uses the modified radar model to modify the results of the original model without changing the structure of the original model, thereby reducing the cost of training, thereby reducing the amount of data required for training when training the radar model, greatly improving the efficiency of model training, and reducing the demand for new scene data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a preferred embodiment of the scene generalization method of the millimeter wave radar model of the present invention;

[0037] Figure 2 It is a training diagram of the original radar model and the modified radar model in a preferred embodiment of the present invention;

[0038] Figure 3 is a schematic structural diagram of a generalized radar model of a preferred embodiment of the present invention;

[0039] Figure 4 1 is a comparative schematic diagram of radar scattering areas predicted by various models in a preferred embodiment of the present invention;

[0040] Figure 5 is a comparative schematic diagram of lateral velocity prediction by various models in a preferred embodiment of the present invention;

[0041] Figure 6 1 is a schematic diagram comparing the longitudinal speeds predicted by various models in a preferred embodiment of the present invention;

[0042] Figure 7 It is a comparative schematic diagram of the lateral distances predicted by various models in a preferred embodiment of the present invention;

[0043] Figure 8 It is a comparative schematic diagram of the longitudinal distances predicted by various models in a preferred embodiment of the present invention;

[0044] Fig. 9 It is a structural diagram of a preferred embodiment of the scene generalization system of the millimeter wave radar model of the present invention;

[0045] Fig.10 It is a schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0046] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0047] It should be noted that if the embodiments of the present invention involve directional indications (such as up, down, left, right, front, back...), the directional indications are only used to explain the relative position relationship, movement status, etc. between the components under a certain specific posture (as shown in the accompanying drawings). If the specific posture changes, the directional indication will also change accordingly.

[0048] In addition, if there are descriptions involving "first", "second", etc. in the embodiments of the present invention, the descriptions of "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or suggesting their relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the technical solutions between the various embodiments can be combined with each other, but they must be based on the ability of ordinary technicians in the field to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be deemed that such a combination of technical solutions does not exist and is not within the scope of protection required by the present invention.

[0049] The scene generalization method of the millimeter wave radar model described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the scenario generalization method of the millimeter wave radar model includes the following steps:

[0050] Step S10: establishing an original radar model, generating a pre-training data set according to the original radar model, and performing model training on the created revised training network model according to the pre-training data set to obtain a revised radar model.

[0051] Specifically, the features between scenes have similar common features and large difference features. When the model is trained with the data of a certain scene and used in other scenes, the difference features between the original scene and the new scene data cause the prediction results of the model on the new scene to have large errors. In this case, the transfer learning method can be used to retain the fitting effect of the original model on the common features between scenes and adjust the difference features between scenes to adapt to the new scene. In an embodiment of the present invention, the original data set with a large amount of data is first used to train the ORM (Original Radar Model). For the new scene to be generalized, the ORM is used as a pre-trained model. On this basis, the RRM (Revised Radar Model) is trained using the new scene data with a small sample, and the output of the pre-trained model is adjusted to retain the ORM's ability to fit the common features while correcting the fitting of the difference features. The reason why all the parameters output by the pre-trained model are adjusted is that the prediction accuracy of the original radar model in the new scene is not high. The purpose of the adjustment is to use a corrected radar model to correct the data output by the original radar model so that it can accurately predict the parameters in the new scene.

[0052] For the above model training Figure 2 As shown, specifically, first, an original radar model is established, and the corresponding process is to obtain multiple original scene data of the original scene, and perform true value extraction on all the original scene data to obtain multiple groups of scene target true values, wherein the scene target true values ​​include the heading angle, lateral speed, longitudinal speed, lateral distance and longitudinal distance of the target object; then, a radar training network model is created, a first group of scene target true values ​​are input into the radar training network model, radar data prediction is performed on the first group of scene target true values, and a first radar data prediction result is obtained; loss value is performed on the real radar data corresponding to the first radar data prediction result and the first group of scene target true values. Calculate to obtain a first loss value, and adjust the parameters of the radar training network model according to the first loss value; input the next set of scene target true values ​​into the radar training network model, and repeat the above process, which will not be repeated here, until the training status of the radar training network model meets the first preset condition, and the first preset condition includes that the first loss value meets the preset requirement or the number of training times reaches the preset number of times, wherein the preset requirement can be determined according to the accuracy of the radar training network model, which will not be described in detail here, and the preset number of times can be the maximum number of training times of the radar training network model, for example, 2000 times, etc., and finally obtain the trained original radar model.

[0053] However, since the original radar model cannot completely and accurately reflect the radar detection results, its output contains certain errors, especially when the original radar model incorrectly predicts the missed alarm state of the target, the output parameters of the original radar model differ greatly from the actual radar output parameters. However, this error will bring unexpected interference to the training of the modified radar model. Therefore, when training the modified radar model, the output of the original radar model cannot be used as the input of the modified radar model alone. The scene information and the output of the original radar model must be used as the input of the modified radar model together. Specifically, multiple scene data of the new scene are obtained, wherein the scene data is a small amount, all of the scene data are input into the original radar model, and the corresponding second radar data prediction result is output; multiple scene parameters of the new scene are obtained, and all of the scene parameters are compared with the second radar data. The prediction results are combined to obtain a pre-training data set; after obtaining the pre-training data set, model training is performed on the created modified training network model according to the pre-training data set, specifically, a first group of training samples of the pre-training data set is input into the modified training network model, radar data prediction is performed on the first group of training samples, and a third radar data prediction result is obtained; the loss value of the third radar data prediction result and the real radar data corresponding to the first group of training samples is calculated to obtain a second loss value, and the parameters of the modified training network model are adjusted according to the second loss value; the next group of training samples is input into the modified training network model, and the above process is repeated, which will not be repeated here, until the training status of the modified training network model meets the second preset condition, and finally a trained modified radar model is obtained.

[0054] Step S20: Combining the original radar model with the modified radar model to obtain a generalized radar model.

[0055] Specifically, after the training is completed to obtain the corrected radar model, in order to achieve generalization of the model to new scenes, in an embodiment of the present invention, the original radar model and the corrected radar model are combined to obtain a GRM-NS (Generalized Radar Model for New Scene); in the simulation requirements for any new scene, since the original radar model has completed the fitting of common features, the corrected radar model only needs to be adjusted for the different features, so the model complexity of the corrected radar model is low, the amount of new scene data required is small, and high modeling efficiency can be achieved.

[0056] Step S30: Obtain the scene data to be generalized of the target scene, input the scene data to be generalized into the generalized radar model, and output a corresponding scene generalization result.

[0057] The step S30 comprises:

[0058] Step S31, obtaining the to-be-generalized scene data of the target scene, and performing truth value extraction on the to-be-generalized scene data to obtain the target truth value of the new scene;

[0059] Step S32: inputting the true value of the new scene target into the generalized radar model, and the original radar model of the generalized radar model performs radar data prediction on the true value of the new scene target to obtain a fourth radar data prediction result;

[0060] Step S33, obtaining radar prediction data of the fourth radar data prediction result, inputting the radar prediction data and the target scene parameters of the target scene into the corrected radar model of the generalized radar model, and outputting the corresponding radar data result;

[0061] Step S34: performing scene generalization on the radar data of the radar data result to obtain a corresponding scene generalization result.

[0062] Specifically, after the generalized radar model is obtained, the target scene can be generalized by the generalized radar model. The specific process is as follows: Figure 3 As shown, in the embodiment of the present invention, taking the vehicle driving scene as an example, the to-be-generalized scene data of the target scene is obtained, and the to-be-generalized scene data is subjected to truth extraction to obtain the new scene target truth value, wherein the new scene target truth value includes the target vehicle heading angle, the target vehicle lateral speed, the target vehicle longitudinal speed, the target vehicle lateral distance and the target vehicle longitudinal distance; then, the new scene target truth value is input into the generalized radar model, and the corresponding radar data result is output, and the specific processing process of the generalized radar model is as follows: the new scene target truth value is input into the generalized radar model, and the original radar model of the generalized radar model performs radar data prediction on the new scene target truth value to obtain The fourth radar data prediction result; obtaining radar prediction data of the fourth radar data prediction result, inputting the radar prediction data and the target scene parameters of the target scene into the modified radar model of the generalized radar model, and outputting the corresponding radar data result; finally, performing scene generalization on the radar data of the radar data result, wherein the radar data includes the target radar scattering area, target lateral speed, target longitudinal speed, target lateral distance and target longitudinal distance of the target scene, and obtaining the corresponding scene generalization result. The present invention not only greatly shortens the amount of data required for generalization, but also can be plug-and-play, reduces the training cost, and also saves the computing resources required for model application.

[0063] Further, in an embodiment of the present invention, in order to verify the modeling method of the present invention, 50,000 SUV (Sport Utility Vehicle) data are used to train an ANN (Artificial Neural Network)-based radar model, RM-ANN, and an ANN-based correction model, CRM-ANN, respectively, wherein CRM-ANN is also used as the original radar model (ORM) of the generalized model; then, 5,000 car data and the output of ORM are used to train a corrected radar model (RRM) for car scenes, and the trained RRM is combined with ORM into a generalized radar model for new scenes, GRM-NS (Generalized Radar Model for New Scene); subsequently, 5,000 car data are trained using the corrected modeling method to obtain CRM-NSSD (Correct Radar Model for New Scene Based on Small Data); 50,000 car data are trained using the corrected method to obtain CRM-NSLD (Correct Radar Model for New Wcene Based on Small Data). LargeData, a modified radar model built using a large amount of new scene data). To verify the effectiveness of the modeling method, a three-layer hidden layer structure with 80 nodes is used for all models except the modified radar model. Because ANN contains one input layer, one output layer and three hidden layers, each hidden layer is set to have 80 nodes, and the activation function of each node is leaky RELU; and the three-layer 80-node structure is a structure that can obtain a better fitting effect after repeated experiments. The RRM hidden layer is 2 layers with 60 nodes because the characteristics of RRM are the use of a small amount of data and a simple structure. The setting of 2 layers with 60 nodes reflects the characteristics of RRM that it is easy to train, which is in line with the design settings; therefore, the hidden layer structures of each model are: RM-ANN is 3 layers with 80 nodes; CRM-ANN and ORM are both 3 layers with 80 nodes; RM-NSSD is 3 layers with 80 nodes; RM-NSLD is 3 layers with 80 nodes; RRM is 2 layers with 60 nodes.

[0064] Afterwards, the test set was used to verify the effect of the model. ORM was able to show certain prediction capabilities in new scenarios, but the error was very large, especially in the prediction of RCS (Radar Cross Section). Since the main difference between SUVs and sedans is reflected in the appearance, and RCS is mainly affected by the appearance, this also proves that the difference characteristics between the scenes will cause the model's prediction results in new scenes to deteriorate. CRM-NSSD, which has insufficient data, also showed poor prediction capabilities, but because the data it used was sedan scene data, it had a significant improvement in RCS prediction accuracy compared to ORM; and GRM-NS constructed by the present invention, with the same amount of data as CRM-NSSD, achieved accurate predictions of various radar output parameters, and the corresponding radar cross section predictions were as follows: Figure 4 As shown, the lateral velocity prediction is as follows Figure 5 As shown, the longitudinal velocity prediction is as follows Figure 6 As shown, the lateral distance prediction is as follows Figure 7 As shown and the longitudinal distance prediction as Figure 8 As shown by Figures 4 to 8 It can be seen that the MSE (Mean-Square Error) of GRM-NS for each parameter prediction is improved by 73.9%, 41.9%, 77.9%, 68.1%, and 42.9% respectively compared with ORM; and by 25.6%, 58.1%, 83.7%, 65.1%, and 92.6% respectively compared with CRM-NSSD. In addition, the model prediction accuracy of GRM-NS is very close to that of CRM-NSLD trained with ten times the amount of data. It can be seen that the target parameter change trends predicted by the two models are basically the same and very close to the true values. Therefore, the present invention solves the most important defect of the data-driven radar model, improves the practicality of the radar model, and can generalize the model to new scenes with less training data, which greatly improves the efficiency of model training, can greatly reduce the demand for new scene data, and saves time and money costs.

[0065] Furthermore, if Fig. 9 As shown, based on the above-mentioned scene generalization method of the millimeter wave radar model, the present invention also provides a scene generalization system of the millimeter wave radar model, wherein the scene generalization system of the millimeter wave radar model includes:

[0066] A model training module 51 is used to establish an original radar model, generate a pre-training data set according to the original radar model, and perform model training on the created modified training network model according to the pre-training data set to obtain a modified radar model;

[0067] A model combination module 52, used for combining the original radar model with the modified radar model to obtain a generalized radar model;

[0068] The scene generalization module 53 is used to obtain the scene data to be generalized of the target scene, input the scene data to be generalized into the generalized radar model, and output the corresponding scene generalization result.

[0069] Furthermore, if Fig.10 As shown, based on the scene generalization method of the above-mentioned millimeter wave radar model, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Fig.10 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0070] In some embodiments, the memory 20 may be an internal storage unit of the terminal, such as a hard disk or memory of the terminal. In other embodiments, the memory 20 may also be an external storage device of the terminal, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Furthermore, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a scene generalization program 40 of a millimeter-wave radar model is stored on the memory 20, and the scene generalization program 40 of the millimeter-wave radar model can be executed by the processor 10, thereby realizing the scene generalization method of the millimeter-wave radar model in the present application.

[0071] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the scene generalization method of the millimeter wave radar model.

[0072] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface. The components of the terminal communicate with each other via a system bus.

[0073] In one embodiment, when the processor 10 executes the scene generalization program 40 of the millimeter wave radar model in the memory 20, the following steps are implemented:

[0074] An original radar model is established, a pre-training data set is generated according to the original radar model, and a modified training network model is trained according to the pre-training data set to obtain a modified radar model;

[0075] Combining the original radar model with the modified radar model to obtain a generalized radar model;

[0076] The to-be-generalized scene data of the target scene is obtained, the to-be-generalized scene data is input into the generalized radar model, and a corresponding scene generalization result is output.

[0077] The establishing of the original radar model specifically includes:

[0078] Acquire multiple original scene data of the original scene, and perform truth value extraction on all the original scene data to obtain multiple groups of scene target truth values;

[0079] Creating a radar training network model, inputting a first set of scene target true values ​​into the radar training network model, performing radar data prediction on the first set of scene target true values, and obtaining a first radar data prediction result;

[0080] Calculating a loss value for the first radar data prediction result and the real radar data corresponding to the first set of scene target true values ​​to obtain a first loss value, and adjusting the parameters of the radar training network model according to the first loss value;

[0081] The next set of scene target true values ​​is input into the radar training network model until the training status of the radar training network model meets the first preset condition, thereby obtaining the original radar model.

[0082] The step of generating a pre-training data set according to the original radar model specifically includes:

[0083] Acquire multiple scene data of the new scene, input all the scene data into the original radar model, and output corresponding second radar data prediction results;

[0084] A plurality of scene parameters of the new scene are obtained, and all the scene parameters are combined with the second radar data prediction result to obtain a pre-training data set.

[0085] The step of performing model training on the created modified training network model according to the pre-training data set to obtain a modified radar model specifically includes:

[0086] Creating a revised training network model, inputting a first group of training samples of the pre-training data set into the revised training network model, performing radar data prediction on the first group of training samples, and obtaining a third radar data prediction result;

[0087] Calculating the loss value of the third radar data prediction result and the real radar data corresponding to the first group of training samples to obtain a second loss value, and adjusting the parameters of the modified training network model according to the second loss value;

[0088] The first set of training samples is input into the modified training network model until the training condition of the modified training network model meets the second preset condition, thereby obtaining a modified radar model.

[0089] The step of obtaining the scene data to be generalized of the target scene, inputting the scene data to be generalized into the generalized radar model, and outputting the corresponding scene generalization result specifically includes:

[0090] Acquire the to-be-generalized scene data of the target scene, and perform truth value extraction on the to-be-generalized scene data to obtain the target truth value of the new scene;

[0091] The new scene target true value is input into the generalized radar model, the corresponding radar data result is output, and the radar data of the radar data result is scene generalized to obtain the corresponding scene generalization result.

[0092] The step of inputting the true value of the new scene target into the generalized radar model and outputting the corresponding radar data result specifically includes:

[0093] The new scene target true value is input into the generalized radar model, and the original radar model of the generalized radar model performs radar data prediction on the new scene target true value to obtain a fourth radar data prediction result;

[0094] The radar prediction data of the fourth radar data prediction result is obtained, the radar prediction data and the target scene parameter of the target scene are input into the modified radar model of the generalized radar model, and the corresponding radar data result is output.

[0095] The scene target true value includes the heading angle, lateral speed, longitudinal speed, lateral distance and longitudinal distance of the target object.

[0096] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a scene generalization program of a millimeter-wave radar model, and when the scene generalization program of the millimeter-wave radar model is executed by a processor, the steps of the scene generalization method of the millimeter-wave radar model as described above are implemented.

[0097] In summary, the present invention provides a method, system, terminal and medium for scene generalization of a millimeter wave radar model, the method comprising: establishing an original radar model, generating a pre-training data set according to the original radar model, and performing model training on the created modified training network model according to the pre-training data set to obtain a modified radar model; combining the original radar model with the modified radar model to obtain a generalized radar model; obtaining the scene data to be generalized of the target scene, inputting the scene data to be generalized into the generalized radar model, and outputting the corresponding scene generalization result. The present invention uses the modified radar model to correct the results of the original model without changing the structure of the original model, thereby reducing the cost of training, thereby reducing the amount of data required for training when training the radar model, greatly improving the efficiency of model training, and reducing the demand for new scene data.

[0098] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.

[0099] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.

[0100] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.

Claims

1. A scene generalization method for a millimeter wave radar model, characterized in that: The scenario generalization method of the millimeter wave radar model includes: An original radar model is established, a pre-training data set is generated according to the original radar model, and a modified training network model is trained according to the pre-training data set to obtain a modified radar model; The establishing of the original radar model specifically includes: Acquire multiple original scene data of the original scene, and perform truth value extraction on all the original scene data to obtain multiple groups of scene target truth values; Creating a radar training network model, inputting a first set of scene target true values ​​into the radar training network model, performing radar data prediction on the first set of scene target true values, and obtaining a first radar data prediction result; Calculating a loss value for the first radar data prediction result and the real radar data corresponding to the first set of scene target true values ​​to obtain a first loss value, and adjusting parameters of the radar training network model according to the first loss value; Inputting the next set of scene target true values ​​into the radar training network model until the training status of the radar training network model meets the first preset condition, thereby obtaining an original radar model; The generating a pre-training data set according to the original radar model specifically includes: Acquire multiple scene data of the new scene, input all the scene data into the original radar model, and output corresponding second radar data prediction results; Acquire multiple scene parameters of the new scene, and combine all the scene parameters with the second radar data prediction result to obtain a pre-training data set; Combining the original radar model with the modified radar model to obtain a generalized radar model; The to-be-generalized scene data of the target scene is obtained, the to-be-generalized scene data is input into the generalized radar model, and a corresponding scene generalization result is output.

2. The scene generalization method of the millimeter wave radar model according to claim 1, characterized in that: According to the The pre-training data set is used to train the created modified training network model to obtain a modified radar model, which specifically includes: Creating a revised training network model, inputting a first group of training samples of the pre-training data set into the revised training network model, performing radar data prediction on the first group of training samples, and obtaining a third radar data prediction result; Calculating the loss value of the third radar data prediction result and the real radar data corresponding to the first group of training samples to obtain a second loss value, and adjusting the parameters of the modified training network model according to the second loss value; The first set of training samples is input into the modified training network model until the training condition of the modified training network model meets the second preset condition, thereby obtaining a modified radar model.

3. The scene generalization method of the millimeter wave radar model according to claim 1, characterized in that: The obtaining of the to-be-generalized scene data of the target scene, inputting the to-be-generalized scene data into the generalized radar model, and outputting the corresponding scene generalization result specifically includes: Acquire the to-be-generalized scene data of the target scene, and perform truth value extraction on the to-be-generalized scene data to obtain the target truth value of the new scene; The true value of the new scene target is input into the generalized radar model, the corresponding radar data result is output, and the radar data of the radar data result is scene-generalized to obtain the corresponding scene generalization result.

4. The scene generalization method of the millimeter wave radar model according to claim 3, characterized in that: The inputting the true value of the new scene target into the generalized radar model and outputting the corresponding radar data result specifically includes: The new scene target true value is input into the generalized radar model, and the original radar model of the generalized radar model performs radar data prediction on the new scene target true value to obtain a fourth radar data prediction result; The radar prediction data of the fourth radar data prediction result is obtained, the radar prediction data and the target scene parameter of the target scene are input into the modified radar model of the generalized radar model, and the corresponding radar data result is output.

5. The scene generalization method of the millimeter wave radar model according to claim 1, characterized in that: The scene target true value includes the heading angle, lateral speed, longitudinal speed, lateral distance and longitudinal distance of the target object.

6. A scene generalization system for a millimeter wave radar model, characterized in that: The scene generalization system of the millimeter wave radar model includes: A model training module is used to establish an original radar model, generate a pre-training data set according to the original radar model, and perform model training on the created revised training network model according to the pre-training data set to obtain a revised radar model; The establishing of the original radar model specifically includes: Acquire multiple original scene data of the original scene, and perform truth value extraction on all the original scene data to obtain multiple groups of scene target truth values; Creating a radar training network model, inputting a first set of scene target true values ​​into the radar training network model, performing radar data prediction on the first set of scene target true values, and obtaining a first radar data prediction result; Calculating a loss value for the first radar data prediction result and the real radar data corresponding to the first set of scene target true values ​​to obtain a first loss value, and adjusting parameters of the radar training network model according to the first loss value; Inputting the next set of scene target true values ​​into the radar training network model until the training status of the radar training network model meets the first preset condition, thereby obtaining an original radar model; The generating a pre-training data set according to the original radar model specifically includes: Acquire multiple scene data of the new scene, input all the scene data into the original radar model, and output corresponding second radar data prediction results; Acquire multiple scene parameters of the new scene, and combine all the scene parameters with the second radar data prediction result to obtain a pre-training data set; A model combination module, used for combining the original radar model with the modified radar model to obtain a generalized radar model; The scene generalization module is used to obtain the scene data to be generalized of the target scene, input the scene data to be generalized into the generalized radar model, and output the corresponding scene generalization result.

7. A terminal, characterized in that: The terminal includes: a memory, a processor, and a memory stored in the memory and capable of A scene generalization program of a millimeter-wave radar model is run on the processor, and when the scene generalization program of the millimeter-wave radar model is executed by the processor, the steps of the scene generalization method of the millimeter-wave radar model as described in any one of claims 1-5 are implemented.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores millimeter wave radar A scene generalization program for a millimeter-wave radar model, wherein when the scene generalization program for the millimeter-wave radar model is executed by a processor, the steps of the scene generalization method for the millimeter-wave radar model as described in any one of claims 1 to 5 are implemented.

Images