Vehicle-mounted SAR speed estimation method, system and terminal based on deep learning

Abstract

The present invention discloses a vehicle-mounted SAR speed estimation method, system and terminal based on deep learning, the method comprising: obtaining the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor; using the sub-aperture speed as the reference speed, dividing the speed range of the vehicle into multiple sub-speed intervals, and marking them as different speed classes; obtaining samples and corresponding labels, building a convolutional neural network model, inputting the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtaining a speed estimation model; predicting the sub-speed interval based on the speed estimation model, and converting the predicted sub-speed interval into a final speed estimation value. The present invention realizes the speed estimation of the sub-aperture echo, ensures the accuracy of the speed estimation value while ensuring efficiency, fully meets the application of vehicle-mounted scenarios, gets rid of the reliance on inertial navigation or similar speed measurement devices, and reduces the system hardware system cost.

Description

Technical Field

[0001] The present invention relates to the field of signal processing technology, and in particular to a vehicle-mounted SAR speed estimation method, system, terminal and computer-readable storage medium based on deep learning. Background Art

[0002] The rapid development of intelligent driving has greatly reduced traffic accidents caused by human factors and has become one of the most promising modes of transportation. One of the key issues of intelligent driving systems is the ability of the car to perceive the surrounding environment. Therefore, a safe and efficient intelligent driving system cannot be separated from a robust and effective environmental perception system as a basic support.

[0003] Smart driving vehicles can better understand the situation of themselves and the surrounding environment, which is inseparable from the support of sensors. Therefore. Automotive sensors can be divided into: optical cameras, ultrasonic, laser radar, millimeter wave radar, etc. Specifically, due to the influence of road visibility, cameras cannot perceive targets all day long; laser radar has high detection accuracy and a large detection range, but its performance will also be severely reduced due to the influence of bad weather. In comparison, millimeter wave radar has the ability to penetrate smoke, fog, rain, and dust, so it has the ability to observe all day and all weather. In addition, in order to obtain two-dimensional high-resolution images, traditional airborne or satellite-borne SAR (Synthetic Aperture Radar) technology is used on the automotive platform, and finally high-resolution SAR imaging of the environment around the car is achieved.

[0004] In the SAR imaging algorithm, accurate radar motion speed information is one of the key parameters required, because inaccurate motion parameters will cause the position of the target in the SAR image to rotate or stretch to varying degrees compared with the actual scene, and even seriously damage the resolution of the image. At the same time, compared with the full-aperture imaging of airborne SAR that records all data, the vehicle-mounted SAR imaging system can only adopt the mode of recording data while imaging in order to meet the real-time requirements, that is, sub-aperture imaging of part of the data. Therefore, the real-time speed of the vehicle-mounted platform radar plays a vital role in the SAR imaging system.

[0005] At present, many research teams mainly use high-precision inertial navigation systems or similar motion sensors to obtain the motion parameters of vehicle-mounted SAR. However, real-time kinematic products such as high-precision inertial navigation are expensive and not practical for automotive scenarios. Many researchers have turned to multi-sensor fusion solutions, that is, using existing automotive sensors to estimate the real-time speed of the platform. Compared with inertial navigation systems, this solution reduces hardware costs, but it still cannot get rid of hardware constraints.

[0006] Therefore, the prior art still needs to be improved and developed. Summary of the invention

[0007] The main purpose of the present invention is to provide a vehicle-mounted SAR speed estimation method, system, terminal and computer-readable storage medium based on deep learning, aiming to solve the problem that the existing technology of obtaining the motion parameters of the vehicle-mounted SAR relies on the limitation of the inertial navigation speed measurement sensor and cannot get rid of the hardware constraints.

[0008] To achieve the above object, the present invention provides a vehicle-borne SAR speed estimation method based on deep learning, and the vehicle-borne SAR speed estimation method based on deep learning comprises the following steps:

[0009] Obtaining the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor;

[0010] Using the sub-aperture speed as a reference speed, dividing the speed range of the vehicle into a plurality of sub-speed intervals, and marking them into different speed classes;

[0011] Obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model;

[0012] The sub-speed interval is predicted based on the speed estimation model, and the predicted sub-speed interval is converted into a final speed estimation value.

[0013] The vehicle-mounted SAR speed estimation method based on deep learning, wherein the speed sensor is a high-precision speed sensor;

[0014] The obtaining of the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor specifically includes:

[0015] The echo aperture obtained within a preset time is regarded as a sub-aperture;

[0016] Get the sub-aperture velocity corresponding to the radar sub-aperture collected by the high-precision velocity sensor.

[0017] The vehicle-mounted SAR speed estimation method based on deep learning, wherein the sub-aperture speed is used as the reference speed, the speed range of the vehicle is divided into multiple sub-speed intervals, and marked as different speed classes, specifically includes:

[0018] The reference speed is shifted according to the shift speed, and the optimal shift speed is determined according to the image evaluation index;

[0019] The speed range of the vehicle is divided into a plurality of sub-speed intervals according to the optimal offset speed, each sub-aperture speed corresponds to each sub-speed interval, and each sub-speed interval is marked as a different speed class.

[0020] In the vehicle-mounted SAR speed estimation method based on deep learning, the image evaluation indicators include equivalent views and entropy, the equivalent views are used to measure the speckle suppression of different SAR images, and entropy is a measure of the amount of image information.

[0021] The vehicle-mounted SAR speed estimation method based on deep learning, wherein the acquiring samples and corresponding labels, building a convolutional neural network model, inputting the acquired samples and corresponding labels into the convolutional neural network model for supervised learning, and obtaining a speed estimation model, specifically includes:

[0022] Get the samples and corresponding labels to be input into the convolutional neural network model, the labels are the velocity classes corresponding to the sub-aperture velocity, and the samples are the range Doppler domain data matrix;

[0023] A convolutional neural network model is built, and the acquired samples and corresponding labels are input into the convolutional neural network model for supervised learning. When the accuracy of the training set and the validation set meets the requirements, a speed estimation model is obtained.

[0024] The vehicle-mounted SAR speed estimation method based on deep learning, wherein the sub-speed interval is predicted based on the speed estimation model, and the predicted sub-speed interval is converted into a final speed estimation value, specifically comprising:

[0025] When the velocity estimation model predicts the velocity category of the sample, each sub-aperture velocity includes multiple sub-velocity categories, and the middle value of the velocity interval is used as the velocity value of the current category;

[0026] The average of the predicted categories of multiple samples is calculated to obtain a final speed estimation value, which is the vehicle-mounted SAR speed.

[0027] In the vehicle-mounted SAR velocity estimation method based on deep learning, the range Doppler domain data matrix is ​​obtained by performing a two-dimensional fast Fourier operation on the near-ground echo.

[0028] In addition, to achieve the above-mentioned purpose, the present invention also provides a vehicle-borne SAR speed estimation system based on deep learning, wherein the vehicle-borne SAR speed estimation system based on deep learning includes:

[0029] A speed acquisition module, used to acquire the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor;

[0030] A speed classification module, used to use the sub-aperture speed as a reference speed, divide the speed range of the vehicle into a plurality of sub-speed intervals, and mark them into different speed classes;

[0031] A model training module is used to obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model;

[0032] The speed estimation module is used to predict the sub-speed interval based on the speed estimation model and convert the predicted sub-speed interval into a final speed estimation value.

[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 deep learning-based vehicle-mounted SAR speed estimation program stored in the memory and runnable on the processor, and when the deep learning-based vehicle-mounted SAR speed estimation program is executed by the processor, the steps of the deep learning-based vehicle-mounted SAR speed estimation method 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 vehicle-mounted SAR speed estimation program based on deep learning, and when the vehicle-mounted SAR speed estimation program based on deep learning is executed by a processor, the steps of the vehicle-mounted SAR speed estimation method based on deep learning as described above are implemented.

[0035] The present invention obtains the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor; uses the sub-aperture speed as the reference speed, divides the speed range of the vehicle into multiple sub-speed intervals, and marks them as different speed classes; obtains samples and corresponding labels, builds a convolutional neural network model, inputs the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtains a speed estimation model; predicts the sub-speed interval based on the speed estimation model, and converts the predicted sub-speed interval into a final speed estimation value. The present invention realizes speed estimation of sub-aperture echoes, ensures the accuracy of speed estimation values ​​while ensuring efficiency, fully meets the application of vehicle-mounted scenarios, gets rid of reliance on inertial navigation or similar speed measurement devices, and reduces system hardware system costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of a preferred embodiment of the vehicle-mounted SAR speed estimation method based on deep learning of the present invention;

[0037] Figure 2 It is a schematic diagram of loss curves of a training set and a validation set in a preferred embodiment of a vehicle-mounted SAR velocity estimation method based on deep learning of the present invention;

[0038] Figure 3 It is a schematic diagram of a speed estimation time curve in a preferred embodiment of a vehicle-mounted SAR speed estimation method based on deep learning of the present invention;

[0039] Figure 4 is a schematic diagram of an optical image of a scene in a preferred embodiment of the vehicle-mounted SAR velocity estimation method based on deep learning of the present invention;

[0040] Figure 5 It is a preferred embodiment of the vehicle-mounted SAR speed estimation method based on deep learning of the present invention. Figure 4 Schematic diagram of SAR image in;

[0041] Figure 6 It is a schematic diagram of the principle of a preferred embodiment of a vehicle-mounted SAR speed estimation system based on deep learning of the present invention;

[0042] Figure 7 Schematic diagram of the operating environment of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION

[0043] 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.

[0044] The vehicle-mounted SAR speed estimation method based on deep learning described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the vehicle-mounted SAR speed estimation method based on deep learning includes the following steps:

[0045] Step S10: Obtain the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor.

[0046] Specifically, the step S10 includes:

[0047] S11, taking the echo aperture obtained within a preset time as a sub-aperture;

[0048] S12. Obtain a sub-aperture velocity corresponding to a radar sub-aperture acquired by a high-precision velocity sensor.

[0049] Wherein, the speed sensor is a high-precision speed sensor.

[0050] Specifically, the echo aperture recorded in a shorter time (for example, the preset time is 0.1 second) is used as a sub-aperture, and the sub-aperture velocity corresponding to the radar sub-aperture is collected by a high-precision velocity sensor and recorded as v c The sub-aperture velocity is used as the reference velocity of the corresponding sub-aperture. For example, scenes 1, 2, and 3 are three different sub-aperture echoes recorded, as shown in Table 1, which lists the sub-aperture velocity values ​​measured by the high-precision velocity sensor.

[0051] Scenario speed Scene 1 3.59m / s Scene 2 3.96m / s Scene 3 3.46m / s

[0052] Table 1: Subaperture velocities for three scenarios

[0053] Step S20: Taking the sub-aperture speed as a reference speed, dividing the speed range of the vehicle into a plurality of sub-speed intervals, and marking them into different speed classes.

[0054] Specifically, the step S20 includes:

[0055] S21, shifting the reference speed according to the shift speed, and determining the optimal shift speed according to the image evaluation index;

[0056] S22. Divide the speed range of the vehicle into a plurality of sub-speed intervals according to the optimal offset speed, wherein each sub-aperture speed corresponds to each sub-speed interval, and mark each sub-speed interval as a different speed class.

[0057] The image evaluation indexes include equivalent appearance number and entropy. The equivalent appearance number is used to measure the speckle suppression of different SAR images, and entropy is a measure of the amount of image information.

[0058] Specifically, the sub-aperture speed in step S10 is used as the reference speed, and the speed offset on the imaging quality is analyzed based on the reference speed for the same sub-aperture. For the same sub-aperture, for example, the high-precision speed sensor measures v1, and this speed is used for imaging. The speed offset Δv is used to analyze the impact of v1±Δv on imaging, that is, to determine the accuracy of the speed. The appropriate value of Δv is determined through actual measured data verification, and then the speed range is divided into multiple reasonable speed intervals and marked as different speed categories.

[0059] Assume that the deviation Δv on the reference speed obtained in step S10 has little effect on the SAR imaging quality, that is, the vehicle speed range can be divided into multiple sub-speed intervals with Δv as the interval. The vehicle travels continuously and has a vehicle speed range within a period of time, such as 2-6m / s. After analyzing the value of Δv, the vehicle speed can be divided into sub-speed intervals with Δv as the interval; each sub-aperture speed can correspond to each sub-speed interval.

[0060] Therefore, three measured sub-aperture echo data (echoes admitted by the sub-aperture) are taken as examples to analyze the impact of the value of Δv on imaging. The image evaluation indicators are equivalent number of views (ENL) and entropy. ENL is usually used to measure the speckle suppression of different SAR images, and Entropy is a measure of the amount of image information.

[0061] As shown in Table 2, the influence of Δv = ±0.1m / s, ±0.2m / s, and ±0.3m / s on the SAR imaging quality is listed. It can be seen from Table 2 that in step S10, the velocity value v c Based on the offset of ±0.3m / s, the ENL and Entropy of the image are very close, so it can be considered that Δv has little effect on the imaging quality within the range of 0.3m / s.

[0062] Therefore, the speed range is divided into intervals of 0.3m / s, considering that the vehicle speed in the actual experiment is between 2m / s and 6m / s, and most of them are concentrated between 2.9m / s and 4.7m / s. Therefore, the speed is divided into 8 intervals: <2.9m / s, 2.9m / s~3.2m / s, 3.2m / s~3.5m / s, 3.5m / s~3.8m / s, 3.8m / s~4.1m / s, 4.1m / s~4.4m / s, 4.4m / s~4.7m / s, and >4.7m / s, and marked as a, b, c, d, e, f, g, and h.

[0063]

[0064] Table 2: ENL and Entropy evaluate the impact of Δv on imaging quality

[0065] Step S30: Obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model.

[0066] Specifically, the step S30 includes:

[0067] S31, obtaining samples and corresponding labels to be input into the convolutional neural network model, where the labels are velocity classes corresponding to the sub-aperture velocity, and the samples are range Doppler domain data matrices;

[0068] S32. Build a convolutional neural network model, input the acquired samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model when the accuracy of the training set and the validation set meets the requirements.

[0069] The range Doppler domain data matrix is ​​obtained by performing a two-dimensional fast Fourier operation on the near-ground echo.

[0070] Specifically, the input of CNN (Convolutional Neural Networks) includes samples and labels, the label is the velocity classification corresponding to the sub-aperture echo velocity value obtained in step S10 (step S10 obtains the sub-aperture velocity, and divides the sub-aperture velocity into each sub-velocity interval), and the sample is the echo of the near-ground area (i.e., the azimuth dimension is -3 to 3 meters, and the distance dimension is 0 to 3 meters) and a 2D (two-dimensional) FFT (Fast Fourier transform) operation is performed to obtain a range Doppler domain data matrix.

[0071] In order to improve the classification performance of CNN, the range-Doppler domain data matrix is ​​divided into 10 small matrices in order, and the samples of the 10 small matrices are labeled with the same label. According to the above operation, as shown in Table 3, the training set and validation set of 8 types of samples are prepared respectively, and the number ratio of the two types of samples is approximately 8:2. The training of the CNN network requires a training set, a validation set, and a test set. Table 3 shows the number of samples prepared respectively.

[0072] category a b c d e f g h Training set 7928 12504 11640 10944 12176 12208 8240 8864 Validation set 1982 3126 2910 2736 3044 3052 2060 2222 Test Set 1670 1590 5040 5280 5720 5470 1440 2310

[0073] Table 3: Number of training sets, validation sets, and test sets

[0074] Furthermore, a CNN model (convolutional neural network model) is built, and the samples and their corresponding labels are sent to the network for supervised learning, and finally a speed estimation model is obtained.

[0075] The CNN model (convolutional neural network model) constructed by the present invention comprises six layers, the first three layers are convolutional layers, the last three layers are fully connected layers, and the last layer connects 8 multi-classifiers through a softmax activation function. Table 4 lists the architecture of the CNN model (convolutional neural network model). The training set and the validation set are fed into the network to fit the model and adjust the hyperparameters, such as Figure 2 As shown in the figure, after 60 epochs (epochs is defined as a single training iteration of all batches in forward and backward propagation, which means that 1 cycle is a single forward and backward pass of the entire input data. Simply put, epochs refers to how many times the data will be "rounded" during training), the validation set loss function began to oscillate and converge, indicating that the convolutional neural network model has been basically trained. In the end, the accuracy of the training set and validation set reached 96.8% and 95.1%. At the same time, in order to evaluate the generalization ability of the model, as shown in Table 3, a new test set was prepared for prediction by the trained CNN model, and the accuracy of the test set was 88.9%.

[0076]

[0077]

[0078] Table 4: Composition and related parameters of the built CNN model

[0079] Step S40: predicting the sub-speed interval based on the speed estimation model, and converting the predicted sub-speed interval into a final speed estimation value.

[0080] Specifically, the step S40 includes:

[0081] S41, when the velocity estimation model predicts the velocity category of the sample, each sub-aperture velocity includes multiple sub-velocity categories, and the middle value of the velocity interval is used as the velocity value of the current category;

[0082] S42, calculating the average value of the predicted categories of multiple samples to obtain a final speed estimation value, where the speed estimation value is the vehicle-mounted SAR speed.

[0083] Specifically, when the speed estimation model predicts the speed category of the sample, a sub-aperture contains ten sub-speed categories, and the middle value of the speed interval is taken as the speed value of the category, for example, a is 3.05m / s. Then, the average value of the predicted category of the ten samples is calculated to be the final estimated speed (i.e., the final speed estimation value). The samples generated by the 2D FFT of the scene one, two, and three sub-apertures are sent to the trained speed estimation model. The prediction results of the speed estimation model and the final speed estimation value are shown in Table 5. Step S10 is the speed measured by the speed sensor, and this Table 5 is the speed estimated by the CNN model. Each sub-aperture echo has 10 samples, and 10 speed classifications are obtained. 10c means that all 10 samples are predicted to be in category c, that is, 3.2m / s~3.5m / s.

[0084] CNN model prediction results Speed ​​estimate Scene 1 10c 3.65m / s Scene 2 10d 4.05m / s Scene 3 9b 1a 3.32m / s

[0085] Table 5: Speed ​​estimation results based on deep learning (scenario 1, 2, 3)

[0086] The effect of the present invention can be illustrated by processing the echo number collected by the 77GHz-81GHz TI AWR1243 radar system. The speed estimation performance of the system is verified by two aspects: speed estimation time and imaging results.

[0087] Among them, the speed estimation time mainly includes the sample generation time and the CNN model speed prediction time. Figure 3 The speed estimation time curve of these 495 frames of images was plotted. The average speed estimation time for each frame of image was calculated to be 0.0538 seconds. It can be seen that the time curve is relatively flat, which is very suitable for vehicle-mounted scenarios.

[0088] The ultimate goal of SAR echo velocity estimation is to make the imaging result have better focusing performance and to show the real actual imaging scene. The sub-aperture echo of a scene is imaged by BP (BackProjection) based on the velocity estimation result. Figure 4 is the actual optical scene, Figure 5 The corresponding SAR imaging results are shown in Figure 2, where the horizontal axis is the distance dimension and the vertical axis is the azimuth dimension. The objects contained in the scene can be clearly identified and matched. In the SAR image, (A) is the near-ground imaging, (B) is the round manhole cover, (D) is the bushes, and (E) is the metal frame of the window. (C) is the bicycles placed side by side and the distance between them is clearly visible. In the optical scene, the angle of the bicycle placement cannot be visually observed, but in the SAR image, more details of the bicycle placement can be seen.

[0089] The present invention provides a vehicle-mounted SAR speed estimation method based on deep learning. In SAR imaging, inaccurate radar speed will seriously affect the SAR imaging quality, so it is necessary to obtain accurate radar speed. The present invention uses sub-aperture echoes and existing speed information, combined with a convolutional neural network model, to convert the speed estimation problem into a speed multi-classification problem. Specifically, the input of the convolutional neural network model includes samples and labels. The samples are subjected to a 2D FFT operation on the sub-aperture near-ground echo to obtain range Doppler domain data, and the labels are derived from the sub-aperture speed measured by the speed sensor. The samples and labels are sent to the convolutional neural network model for supervised learning, and finally a speed estimation model is obtained; the speed estimation model can predict the speed interval of the sub-aperture echo, and then convert the predicted speed interval into a specific speed value, which is the final speed estimation value; the speed estimation of the radar sub-aperture echo is realized, and the calculation efficiency is high; the scheme makes the acquisition of speed parameters free from the limitation of relying on speed measuring sensors such as inertial navigation, greatly reduces the system hardware system, simplifies the system implementation, and reduces the system hardware complexity. At the same time, the results of imaging the measured echo data with the estimated velocity value also prove that the scheme has good performance, that is, it realizes the velocity estimation of sub-aperture echoes, ensures the accuracy of the velocity estimation value while ensuring efficiency, and fully meets the application of vehicle-mounted scenarios.

[0090] Furthermore, the speed classification interval used for CNN model training can be narrowed to further improve the speed estimation accuracy; in the experiment, the speed range of the SAR echo acquisition platform can be expanded to enrich the speed sample classes; and a better deep learning model framework can be built to make the model more generalizable and the speed estimation more accurate.

[0091] Furthermore, if Figure 6As shown, based on the above-mentioned vehicle-borne SAR speed estimation method based on deep learning, the present invention also provides a vehicle-borne SAR speed estimation system based on deep learning, wherein the vehicle-borne SAR speed estimation system based on deep learning includes:

[0092] The speed acquisition module 51 is used to acquire the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor;

[0093] A speed classification module 52, for taking the sub-aperture speed as a reference speed, dividing the speed range of the vehicle into a plurality of sub-speed intervals, and marking them into different speed classes;

[0094] A model training module 53 is used to obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model;

[0095] The speed estimation module 54 is used to predict the sub-speed interval based on the speed estimation model and convert the predicted sub-speed interval into a final speed estimation value.

[0096] Furthermore, if Figure 7 As shown, based on the above-mentioned deep learning-based vehicle-mounted SAR speed estimation method and system, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 7 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.

[0097] 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 (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, 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 vehicle-mounted SAR speed estimation program 40 based on deep learning is stored on the memory 20, and the vehicle-mounted SAR speed estimation program 40 based on deep learning can be executed by the processor 10, thereby realizing the vehicle-mounted SAR speed estimation method based on deep learning in the present application.

[0098] 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 deep learning-based vehicle-mounted SAR speed estimation method.

[0099] 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 10-30 of the terminal communicate with each other via a system bus.

[0100] In one embodiment, when the processor 10 executes the deep learning-based vehicle-mounted SAR speed estimation program 40 in the memory 20, the following steps are implemented:

[0101] Obtaining the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor;

[0102] Using the sub-aperture speed as a reference speed, dividing the speed range of the vehicle into a plurality of sub-speed intervals, and marking them into different speed classes;

[0103] Obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a speed estimation model;

[0104] The sub-speed interval is predicted based on the speed estimation model, and the predicted sub-speed interval is converted into a final speed estimation value.

[0105] Wherein, the speed sensor is a high-precision speed sensor;

[0106] The obtaining of the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor specifically includes:

[0107] The echo aperture obtained within a preset time is regarded as a sub-aperture;

[0108] Get the sub-aperture velocity corresponding to the radar sub-aperture collected by the high-precision velocity sensor.

[0109] The sub-aperture speed is used as the reference speed, and the speed range of the vehicle is divided into a plurality of sub-speed intervals, which are marked as different speed classes, specifically including:

[0110] The reference speed is shifted according to the shift speed, and the optimal shift speed is determined according to the image evaluation index;

[0111] The speed range of the vehicle is divided into a plurality of sub-speed intervals according to the optimal offset speed, each sub-aperture speed corresponds to each sub-speed interval, and each sub-speed interval is marked as a different speed class.

[0112] The image evaluation indexes include equivalent appearance number and entropy. The equivalent appearance number is used to measure the speckle suppression of different SAR images, and entropy is a measure of the amount of image information.

[0113] The acquiring of samples and corresponding labels, building a convolutional neural network model, inputting the acquired samples and corresponding labels into the convolutional neural network model for supervised learning, and obtaining a speed estimation model specifically includes:

[0114] Get the samples and corresponding labels to be input into the convolutional neural network model, the labels are the velocity classes corresponding to the sub-aperture velocity, and the samples are the range Doppler domain data matrix;

[0115] A convolutional neural network model is built, and the acquired samples and corresponding labels are input into the convolutional neural network model for supervised learning. When the accuracy of the training set and the validation set meets the requirements, a speed estimation model is obtained.

[0116] The predicting of the sub-speed interval based on the speed estimation model and converting the predicted sub-speed interval into a final speed estimation value specifically includes:

[0117] When the velocity estimation model predicts the velocity category of the sample, each sub-aperture velocity includes multiple sub-velocity categories, and the middle value of the velocity interval is used as the velocity value of the current category;

[0118] The average of the predicted categories of multiple samples is calculated to obtain a final speed estimation value, which is the vehicle-mounted SAR speed.

[0119] The range Doppler domain data matrix is ​​obtained by performing a two-dimensional fast Fourier operation on the near-ground echo.

[0120] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a vehicle-mounted SAR speed estimation program based on deep learning, and when the vehicle-mounted SAR speed estimation program based on deep learning is executed by a processor, the steps of the vehicle-mounted SAR speed estimation method based on deep learning as described above are implemented.

[0121] In summary, the present invention provides a vehicle-mounted SAR speed estimation method, system, terminal and computer-readable storage medium based on deep learning, the method comprising: obtaining the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor; using the sub-aperture speed as the reference speed, dividing the vehicle's speed range into multiple sub-speed intervals, and marking them as different speed classes; obtaining samples and corresponding labels, building a convolutional neural network model, inputting the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtaining a speed estimation model; predicting the sub-speed interval based on the speed estimation model, and converting the predicted sub-speed interval into a final speed estimation value. The present invention realizes the speed estimation of the sub-aperture echo, ensures the accuracy of the speed estimation value while ensuring efficiency, fully meets the application of vehicle-mounted scenarios, gets rid of the reliance on inertial navigation or similar speed measurement devices, and reduces the system hardware system cost.

[0122] 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.

[0123] 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. It should be understood that the application of the present invention is not limited to the above-mentioned examples. For those skilled in the art, it can be improved or transformed according to the above description, and all these improvements and transformations should belong to the protection scope of the claims attached to the present invention.

Claims

1. A vehicle-mounted SAR velocity estimation method based on deep learning, characterized in that: The vehicle-mounted SAR speed estimation method based on deep learning includes: Obtaining the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor; The sub-aperture speed is used as a reference speed, and the speed range of the vehicle is divided into a plurality of sub-speed intervals, and the sub-speed intervals are marked as different speed classes; the reference speed is shifted according to the shift speed, and an optimal shift speed is determined according to an image evaluation index; the speed range of the vehicle is divided into a plurality of sub-speed intervals according to the optimal shift speed, each sub-aperture speed corresponds to each sub-speed interval, and each sub-speed interval is marked as a different speed class; Acquire samples and corresponding labels, build a convolutional neural network model, input the acquired samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a velocity estimation model; acquire samples and corresponding labels to be input into the convolutional neural network model, the labels are velocity classes corresponding to the sub-aperture velocity, and the samples are range Doppler domain data matrices; build a convolutional neural network model, input the acquired samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a velocity estimation model when the accuracy of the training set and the validation set meet the requirements; divide the range Doppler domain data matrix into 10 small matrices in sequence, and label the 10 small matrix samples with the same labels; The sub-speed interval is predicted based on the speed estimation model, and the predicted sub-speed interval is converted into a final speed estimation value; when the speed estimation model predicts the speed category of the sample, each sub-aperture speed contains multiple sub-speed categories, and the middle value of the speed interval is used as the speed value of the current category; the average value of the predicted categories of multiple samples is calculated to obtain the final speed estimation value, and the speed estimation value is the vehicle-mounted SAR speed.

2. The vehicle-mounted SAR velocity estimation method based on deep learning according to claim 1, characterized in that: The speed sensor is a high-precision speed sensor; The obtaining of the sub-aperture velocity corresponding to the radar sub-aperture collected by the velocity sensor specifically includes: The echo aperture obtained within a preset time is regarded as a sub-aperture; Get the sub-aperture velocity corresponding to the radar sub-aperture collected by the high-precision velocity sensor.

3. The vehicle-mounted SAR velocity estimation method based on deep learning according to claim 1, characterized in that: The image evaluation indexes include equivalent view number and entropy. The equivalent view number is used to measure the speckle suppression of different SAR images, and entropy is a measure of the amount of image information.

4. The vehicle-mounted SAR velocity estimation method based on deep learning according to claim 1, characterized in that: The range Doppler domain data matrix is ​​obtained by performing a two-dimensional fast Fourier operation on the near-ground echo.

5. A vehicle-mounted SAR speed estimation system based on deep learning, characterized in that: The vehicle-mounted SAR speed estimation system based on deep learning includes: A speed acquisition module, used to acquire the sub-aperture speed corresponding to the radar sub-aperture collected by the speed sensor; A speed classification module, used to use the sub-aperture speed as a reference speed, divide the speed range of the vehicle into a plurality of sub-speed intervals, and mark them into different speed classes; shift the reference speed according to the shift speed, and determine the optimal shift speed according to the image evaluation index; divide the speed range of the vehicle into a plurality of sub-speed intervals according to the optimal shift speed, each sub-aperture speed corresponds to each sub-speed interval, and mark each sub-speed interval as a different speed class; The model training module is used to obtain samples and corresponding labels, build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a velocity estimation model; obtain samples and corresponding labels to be input into the convolutional neural network model, the labels are velocity classes corresponding to the sub-aperture velocity, and the samples are range Doppler domain data matrices; build a convolutional neural network model, input the obtained samples and corresponding labels into the convolutional neural network model for supervised learning, and obtain a velocity estimation model when the accuracy of the training set and the verification set meets the requirements; divide the range Doppler domain data matrix into 10 small matrices in sequence, and label the 10 small matrix samples with the same label; The speed estimation module is used to predict the sub-speed interval based on the speed estimation model and convert the predicted sub-speed interval into a final speed estimation value; when the speed estimation model predicts the speed category of the sample, each sub-aperture speed contains multiple sub-speed categories, and the middle value of the speed interval is used as the speed value of the current category; the average value of the predicted categories of multiple samples is calculated to obtain the final speed estimation value, and the speed estimation value is the vehicle-mounted SAR speed.

6. A terminal, characterized in that: The terminal includes: a memory, a processor, and a deep learning-based vehicle-mounted SAR speed estimation program stored in the memory and running on the processor. When the deep learning-based vehicle-mounted SAR speed estimation program is executed by the processor, the steps of the deep learning-based vehicle-mounted SAR speed estimation method according to any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a vehicle-mounted SAR speed estimation program based on deep learning. When the vehicle-mounted SAR speed estimation program based on deep learning is executed by a processor, the steps of the vehicle-mounted SAR speed estimation method based on deep learning are implemented as described in any one of claims 1 to 4.

Images