A radar image denoising processing method based on deep learning

Abstract

The present invention provides a method for radar image noise reduction processing based on deep learning, belonging to the technical field of machine learning, including obtaining historical echo signal data, determining a historical radar image data set based on the historical echo signal data; determining a training image set and a test image set based on the historical radar image data set, and constructing a noise reduction neural network model based on the training image set; evaluating the noise reduction neural network model based on the test image set, and performing a second optimization on the noise reduction neural network model; performing noise reduction processing on a real-time radar image data set determined by real-time radar echo data based on the optimized noise reduction neural network model. It can enhance the noise reduction learning ability of the model, improve the processing ability of the model for complex noise environments, realize the gradual refinement adjustment of the model to form a performance closed-loop, realize the high-precision and fast processing of real-time radar data, improve the denoising effect of radar images, and improve the noise reduction accuracy in complex environments.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and particularly to a method for radar image denoising processing based on deep learning. Background Art

[0002] In the field of radar image processing, traditional denoising methods mainly include filtering-based algorithms (such as mean filtering, median filtering, and Wiener filtering) and statistical methods (such as Gaussian noise modeling). These methods smooth or remove the noise in the radar image by making assumptions about the noise characteristics. However, these traditional methods have limitations when faced with complex and variable noise environments, and it is difficult to effectively distinguish the target signal from the background noise, which easily leads to the loss of target information or residual noise. At the same time, with the improvement of the resolution of modern radar systems and the complexity of application scenarios, the computational efficiency and adaptability of traditional methods can no longer meet the requirements of real-time and high-precision.

[0003] Therefore, the present invention provides a method for radar image denoising processing based on deep learning. Summary of the Invention

[0004] The present invention provides a method for radar image denoising processing based on deep learning. By analyzing the historical echo data signal to determine the historical radar image data set, constructing and optimizing the denoising neural network model, and using the optimized denoising neural network model to perform denoising processing on the real-time radar image data set determined by the real-time radar echo data, the denoising learning ability of the model can be enhanced, the processing ability of the model for complex noise environments can be improved, a performance closed-loop can be formed by gradually refining and adjusting the model, high-precision and fast processing of real-time radar data can be achieved, the denoising effect of the radar image can be improved, and the denoising accuracy in complex environments can be enhanced.

[0005] The present invention provides a method for radar image denoising processing based on deep learning, including:

[0006] 101: Obtain historical echo signal data, and determine the historical radar image data set based on the historical echo signal data;

[0007] 102: Determine the training image set and the test image set based on the historical radar image data set, and construct a denoising neural network model based on the training image set;

[0008] 103: Evaluate the denoising neural network model based on the test image set, and perform a second optimization on the denoising neural network model;

[0009] 104: Perform denoising processing on the real-time radar image data set determined by the real-time radar echo data based on the optimized denoising neural network model.

[0010] A method for radar image noise reduction processing based on deep learning provided by the present invention obtains historical echo signal data and determines a historical radar image data set based on the historical echo signal data, including:

[0011] Obtain historical echo signal data of the radar within a historical specified time period, where the historical echo signal data includes radar signals at each time point within the historical specified time period;

[0012] Determine segmented time windows of the historical specified time period based on the pulse period, and divide the historical echo signal into multiple time echo signals based on the segmented time windows;

[0013] Convert each time echo signal into corresponding range information based on the pulse compression algorithm, and determine the range dimension information of each time echo signal, where the range dimension information includes the echo intensity of multiple range cells;

[0014] Perform Fourier transform on the range dimension data of each time echo signal to determine the Doppler frequency information of each range dimension data;

[0015] Determine a three-dimensional matrix of the historical echo signal data based on all time echo signals, the range dimension information of all time echo signals, and the Doppler frequency information of all range cells in the range dimension information;

[0016] Determine a two-dimensional matrix at each time point within the historical specified time period based on the time dimension of the three-dimensional matrix;

[0017] Determine the historical range-Doppler image of each segmented time window based on the two-dimensional matrices at all time points within each segmented time window, and perform time marking on the historical range-Doppler image of each segmented time window;

[0018] Perform sequential combination on the historical range-Doppler images of all segmented time windows based on the time marking to determine the historical radar image data set.

[0019] A method for radar image noise reduction processing based on deep learning provided by the present invention determines a training image set and a test image set based on the historical radar image data set, including:

[0020] Preprocess the historical radar image data set, where the preprocessing includes image quality inspection, image normalization, and image size adjustment;

[0021] Determine the target noise-free image of each historical range-Doppler image in the preprocessed radar image data set;

[0022] Determine the division ratio of the historical radar image dataset, and based on the division ratio, the time stamps of all historical range-Doppler images in the historical radar image dataset, and the target noise-free images, determine the training image set and the test image set.

[0023] According to a radar image denoising processing method based on deep learning provided by the present invention, construct a denoising neural network model based on the training image set, including:

[0024] The input layer of the denoising neural network model receives all historical range-Doppler images in the training image set;

[0025] Based on multiple convolutional layers of the denoising neural network model and the non-linear activation function of each convolutional layer, perform spatial feature extraction and noise feature extraction on each historical range-Doppler image received by the input layer to determine the local feature vector of each historical range-Doppler image;

[0026] Based on multiple deep convolutional layers of the denoising neural network model, perform fine-grained feature extraction on each historical range-Doppler image in the training image set to determine the fine-grained feature vector of each historical range-Doppler image;

[0027] Use multiple deconvolutional layers of the denoising neural network model to reconstruct the local feature vector and the fine-grained feature vector of each historical range-Doppler image in the training image set respectively, and determine the predicted noise-free image of each historical range-Doppler image;

[0028] Determine the comprehensive loss value based on the target noise-free image and the predicted noise-free image of the historical range-Doppler image in the training image set;

[0029] Compare the comprehensive loss value with the preset loss threshold. If the comprehensive loss value is greater than or equal to the preset loss threshold, perform the first optimization on the denoising neural network model based on the optimization algorithm until the comprehensive loss value is less than the preset loss threshold.

[0030] According to a radar image denoising processing method based on deep learning provided by the present invention, determine the comprehensive loss value based on the target noise-free image and the predicted noise-free image of the historical range-Doppler image in the training image set, including:

[0031] ;

[0032] ;

[0033] ;

[0034] ;

[0035] Wherein, Represents the comprehensive loss value, Represents the loss value of the a-th historical range-Doppler image in the training image set, Respectively represent the maximum loss value and the minimum loss value of all historical range-Doppler images in the training image set. Nu represents the number of historical range-Doppler images in the training image set, Represents the pixel loss value of the a-th historical range-Doppler image in the training image set, Represents the frequency domain loss value of the a-th historical range-Doppler image in the training image set, Represents the probability prediction value that the discriminator D determines that the a-th historical range-Doppler image in the training image set belongs to a real image, Represents The expected value of, 、 Respectively represent the pixel values of the i-th pixel point in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set. N1 represents the number of pixel points in the predicted noise-free image and the target noise-free image of the historical range-Doppler image, Represents the structural similarity value between the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, Represents the weight of the structural similarity value, Respectively represent the k-th frequency component in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, Respectively represent the phases of the k-th frequency components in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set. N2 represents the number of frequency components in the predicted noise-free image and the target noise-free image of the historical range-Doppler image, Represents the weight of the frequency component, Represents the weight of the phase of the frequency component, Represents the loss constant, with a value of 0.1.

[0036] According to a radar image denoising processing method based on deep learning provided by the present invention, evaluating the denoising neural network model based on a test image set and performing a second optimization on the denoising neural network model, including:

[0037] Input each historical range-Doppler image in the test set image into the first optimized denoising neural network model, and determine the predicted noise-free image of each historical range-Doppler image in the test set image based on the output result of the denoising neural network model;

[0038] Compare the predicted noise-free images and the target noise-free images of all historical range-Doppler images in the comparison test set to determine the denoising effect of the denoising neural network model;

[0039] Judge whether the denoising effect meets the preset denoising expectation. If not, adjust the model parameters of the denoising neural network model based on the denoising effect, and perform a second optimization on the denoising neural network model until the denoising effect meets the preset denoising expectation.

[0040] According to a method for radar image denoising processing based on deep learning provided by the present invention, perform denoising processing on a real-time radar image data set determined from real-time radar echo data based on the optimized denoising neural network model, including:

[0041] Obtain real-time radar echo data, and determine a real-time radar image set based on the real-time radar echo data;

[0042] Input the real-time radar image set into the second optimized denoising neural network model, and determine the denoising processing result of the real-time radar image set based on the output result of the denoising neural network model.

[0043] According to a method for radar image denoising processing based on deep learning provided by the present invention, the real-time radar image set includes a plurality of real-time range-Doppler images;

[0044] The denoising processing result includes the predicted denoised images of each real-time range-Doppler image in the real-time radar image set.

[0045] Compared with the prior art, the beneficial effects of the present application are as follows:

[0046] By analyzing the historical echo data signal to determine the historical radar image data set, constructing and optimizing the denoising neural network model, and performing denoising processing on the real-time radar image data set determined from the real-time radar echo data based on the optimized denoising neural network model, the denoising learning ability of the model can be enhanced, the processing ability of the model for complex noise environments can be improved, a performance closed-loop can be formed by gradually refining and adjusting the model, high-precision and fast processing of real-time radar data can be realized, the denoising effect of radar images can be improved, and the denoising accuracy in complex environments can be enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1It is a schematic flowchart of a radar image denoising processing method based on deep learning provided by an embodiment of the present invention. Detailed implementation manners

[0049] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0050] Embodiment 1:

[0051] An embodiment of the present invention provides a radar image denoising processing method based on deep learning, as Figure 1 shown, including:

[0052] 101: Obtain historical echo signal data, and determine a historical radar image dataset based on the historical echo signal data;

[0053] 102: Determine a training image set and a test image set based on the historical radar image dataset, and construct a denoising neural network model based on the training image set;

[0054] 103: Evaluate the denoising neural network model based on the test image set, and perform a second optimization on the denoising neural network model;

[0055] 104: Perform denoising processing on a real-time radar image dataset determined from real-time radar echo data based on the optimized denoising neural network model.

[0056] In this embodiment, the historical echo signal data is the echo signal record collected by the radar in the past period of time, including the distance and speed information of the target object and environmental noise. By processing these historical echo signal data, the corresponding historical radar image dataset is generated.

[0057] In this embodiment, the historical radar image dataset is divided into a training image set and a test image set. The training set is used for the learning of the model, and the test set is used for verifying the model performance. Through deep learning, the training image set is used to train the neural network model so that it can learn how to extract target features from the noisy image and remove noise.

[0058] In this embodiment, the test image set is used to evaluate the denoising effect of the initial neural network model, judge its adaptability and processing accuracy to different noise environments, and perform a second round of optimization on the model based on the evaluation results. By adjusting the model architecture, hyperparameters, or training strategy, the robustness and generalization ability of the model are improved.

[0059] In this embodiment, after the model optimization is completed, the radar echo data received in real time is converted into a real-time radar image data set and input into the optimized noise reduction neural network model for real-time processing. The model can quickly generate clear noise reduction images, extract key features of the target, and remove environmental noise at the same time.

[0060] Beneficial effects of the above technical solution: By analyzing the historical echo data signal to determine the historical radar image data set, constructing and optimizing the noise reduction neural network model, and performing noise reduction processing on the real-time radar image data set determined by the optimized noise reduction neural network model for the real-time radar echo data, the noise reduction learning ability of the model can be enhanced, the processing ability of the model for complex noise environments can be improved, a performance closed-loop can be formed by gradually refining and adjusting the model, high-precision and fast processing of real-time radar data can be achieved, the denoising effect of radar images can be improved, and the noise reduction accuracy in complex environments can be enhanced.

[0061] Embodiment 2:

[0062] The embodiment of the present invention provides a method for radar image noise reduction processing based on deep learning. Obtain historical echo signal data, and determine a historical radar image data set based on the historical echo signal data, including:

[0063] Obtain the historical echo signal data of the radar within a historical specified time period, where the historical echo signal data includes the radar signals at each time point within the historical specified time period;

[0064] Based on the pulse period, determine the segmented time windows of the historical specified time period, and divide the historical echo signal into multiple time echo signals based on the segmented time windows;

[0065] Based on the pulse compression algorithm, convert each time echo signal into the corresponding distance information respectively, and determine the distance dimension information of each time echo signal, where the distance dimension information includes the echo intensities of multiple distance units;

[0066] Perform Fourier transform on the distance dimension data of each time echo signal to determine the Doppler frequency information of each distance dimension data;

[0067] Based on all the time echo signals, the distance dimension information of all the time echo signals, and the Doppler frequency information of all the distance units in the distance dimension information, determine the three-dimensional matrix of the historical echo signal data;

[0068] Based on the time dimension of the three-dimensional matrix, determine the two-dimensional matrix at each time point within the historical specified time period;

[0069] Determine the historical range-Doppler image for each segmented time window based on the two-dimensional matrices at all time points within each segmented time window, and attach a time stamp to the historical range-Doppler image of each segmented time window;

[0070] Based on the time stamps, perform sequential combination on the historical range-Doppler images of all segmented time windows to determine the historical radar image dataset.

[0071] In this embodiment, the echo signals of the radar system within a historical specified time period are collected, and the radar signals at each time point within this time period are recorded. The echo signals reflect the characteristic information of the radar transmitted signal reflected by the object, including the physical characteristics of the target object and the environment.

[0072] In this embodiment, according to the pulse period of the radar, the historical specified time period is divided into several time windows, each window representing a segmented time range, and the historical echo signals are sliced into multiple time echo signals according to these segmented time windows.

[0073] In this embodiment, the pulse compression algorithm is used to process each time echo signal, convert it from the time domain to the range domain, and obtain the range dimension information of each echo signal. The range dimension information represents the echo intensity on different range cells and is a direct reflection of the target range and the reflected signal intensity.

[0074] In this embodiment, Fourier transform is performed on the range dimension data of each time echo signal to extract the Doppler frequency information corresponding to each range cell. The Doppler frequency reflects the velocity characteristics of the target. Combining the range information can determine the position and motion state of the target.

[0075] In this embodiment, all time echo signals, their range dimension information, and the Doppler frequency information of the range cells are organized into a three-dimensional matrix. The dimensions of this three-dimensional matrix are time, range, and Doppler frequency, which completely describe the information observed by the radar within the historical specified time period.

[0076] In this embodiment, based on the time dimension of the three-dimensional matrix, the data at each time point is extracted as a two-dimensional matrix, representing the target distribution situation and its range and velocity information at that time point.

[0077] In this embodiment, the two-dimensional matrices at all time points within each segmented time window are synthesized to generate the historical range-Doppler image of this segmented time window. The range-Doppler image is a typical form of radar image, reflecting the range and velocity characteristics of the target. Attach a time stamp to the image of each segmented time window to indicate the time sequence of the image.

[0078] In this embodiment, the horizontal axis of the range-Doppler image represents the Doppler frequency (target velocity unit); the vertical axis represents the range unit (target range unit); and the color represents the echo intensity (the higher the intensity, the brighter the color, and a pseudo-color map can be used).

[0079] In this embodiment, based on the time stamps, the historical range-Doppler images of all segmented time windows are combined in chronological order to generate a complete historical radar image dataset, providing a basis for subsequent noise reduction and analysis.

[0080] In this embodiment, the pulse period represents the time interval of the radar transmitted signal and is used to determine the time window for signal sampling.

[0081] In this embodiment, the pulse compression algorithm represents converting the radar echo from the time domain to the range domain through signal processing techniques to improve the range resolution.

[0082] Beneficial effects of the above technical solution: By obtaining historical echo signal data and determining the historical radar image dataset based on the historical echo signal data, the orderly and efficient organization of historical radar data can be achieved, improving data integrity and accuracy, and providing a high-quality input dataset for constructing a noise reduction neural network model.

[0083] Embodiment 3:

[0084] An embodiment of the present invention provides a method for radar image noise reduction processing based on deep learning. Based on the historical radar image dataset, a training image set and a test image set are determined, including:

[0085] Preprocess the historical radar image dataset, where the preprocessing includes image quality inspection, image normalization, and image size adjustment;

[0086] Determine the target noise-free image of each historical range-Doppler image in the preprocessed radar image dataset;

[0087] Determine the division ratio of the historical radar image dataset. Based on the division ratio, the time stamps of all historical range-Doppler images in the historical radar image set, and the target noise-free image, determine the training image set and the test image set.

[0088] In this embodiment, the image quality inspection means performing quality assessment on each historical range-Doppler image in the historical radar image dataset, removing invalid images caused by signal loss or anomalies, and ensuring the reliability and integrity of the data.

[0089] In this embodiment, image normalization refers to standardizing the pixel values of an image (such as normalizing to the range [0, 1] or [-1, 1]), making the distribution of image data consistent, eliminating data biases caused by different sampling dynamic ranges, and facilitating the convergence and performance improvement of deep learning models.

[0090] In this embodiment, image size adjustment means uniformly adjusting the size of the image (such as cropping or interpolation processing) to a fixed size, which is convenient for the input and calculation of the neural network model and avoids calculation problems caused by inconsistent sizes.

[0091] In this embodiment, for historical range-Doppler images, through prior knowledge or algorithm processing, a noise-free version of the target for each image is extracted. The noise-free target image represents the ideal output after denoising and is used as the "label" data for supervised learning to guide the denoising neural network model to learn how to remove noise and retain target features. These noise-free images can be generated by traditional filtering methods, expert annotation, or high-quality radar systems to ensure that they are as close as possible to the true target reflection characteristics.

[0092] In this embodiment, the data is divided according to common ratios (which can be 80% training set, 20% test set or 70% training set, 30% test set). The training set is used for model learning, and the test set is used for model verification to ensure the independence of training and evaluation.

[0093] In this embodiment, combined with time stamps, the data is divided in chronological order to ensure that the training set and the test set contain representative and diverse image samples and avoid overfitting.

[0094] In this embodiment, when dividing the training set and the test set, it is ensured that the noise-free images corresponding to each range-Doppler image are also correctly divided to ensure data alignment for supervised learning.

[0095] The beneficial effects of the above technical solutions: Determining the training image set and the test image set based on the historical radar image dataset can provide high-quality supervision signals, enhance the noise reduction learning ability of the model, and improve the robustness and accuracy of the noise reduction model in complex radar image processing tasks.

[0096] Embodiment 4:

[0097] An embodiment of the present invention provides a method for radar image noise reduction processing based on deep learning. A denoising neural network model is constructed based on the training image set, including:

[0098] The input layer of the denoising neural network model receives all historical range-Doppler images in the training image set;

[0099] Based on multiple convolutional layers of the denoising neural network model and the non-linear activation functions of each convolutional layer, spatial feature extraction and noise feature extraction are performed on each historical range-Doppler image received by the multi-input layer, and the local feature vector of each historical range-Doppler image is determined;

[0100] Based on multiple deep convolutional layers of the denoising neural network model, fine-grained feature extraction is performed on each historical range-Doppler image in the training image set, and the fine-grained feature vector of each historical range-Doppler image is determined;

[0101] Using multiple deconvolutional layers of the denoising neural network model, the local feature vector and the fine-grained feature vector of each historical range-Doppler image in the training image set are respectively reconstructed to determine the predicted noise-free image of each historical range-Doppler image;

[0102] Based on the target noise-free image and the predicted noise-free image of the historical range-Doppler image in the training image set, the comprehensive loss value is determined;

[0103] Compare the comprehensive loss value with the preset loss threshold. If the comprehensive loss value is greater than or equal to the preset loss threshold, the denoising neural network model is first optimized based on the optimization algorithm until the comprehensive loss value is less than the preset loss threshold.

[0104] In this embodiment, the input layer of the denoising neural network model directly receives all historical range-Doppler images in the training image set. These images contain target features and noise information. The role of the input layer is to take the original image as input and perform preliminary processing for subsequent convolutional layers to extract features.

[0105] In this embodiment, multiple convolutional layers of the denoising neural network are used to extract features from the image. The convolutional layer performs local operations on the image through a convolutional kernel to capture spatial features (target distribution) and noise features. The non-linear activation function (such as ReLU) introduces non-linear transformation to enhance the model's ability to express complex patterns and generate local feature vectors to describe the feature distribution in a small range of the image.

[0106] In this embodiment, multiple deep convolutional layers are used to further perform fine-grained feature extraction on the image. These deep convolutional layers have more convolutional kernels and deeper layers, and can capture higher-level and more detailed features, generate fine-grained feature vectors, and further separate target and noise features.

[0107] In this embodiment, through multiple deconvolutional layers, the local feature vector and the fine-grained feature vector are gradually decoded and reconstructed to generate the predicted noise-free image. The deconvolutional layer restores the low-dimensional feature vector to the high-dimensional image form and uses the extracted target features to restore the noise-free version of the image.

[0108] In this embodiment, if the comprehensive loss value is greater than or equal to the preset loss threshold, it indicates that the prediction ability of the model is insufficient and needs to be further optimized. By using an optimization algorithm (such as gradient descent or Adam algorithm), the parameters of the model (such as the weights of the convolutional kernels) are adjusted to continuously reduce the loss value until the comprehensive loss value is less than the preset threshold.

[0109] The beneficial effects of the above technical solution: By constructing a noise reduction neural network model based on the training image set, the processing ability of the model for complex noise environments can be improved, and the denoising effect of radar images can be significantly enhanced.

[0110] Embodiment 5:

[0111] The embodiment of the present invention provides a method for processing radar image denoising based on deep learning. The comprehensive loss value is determined based on the target noise-free image and the predicted noise-free image of the historical range-Doppler images in the training image set, including:

[0112] ;

[0113] ;

[0114] ;

[0115] ;

[0116] Among them, represents the comprehensive loss value, represents the loss value of the a-th historical range-Doppler image in the training image set, respectively represent the maximum loss value and the minimum loss value of all historical range-Doppler images in the training image set, Nu represents the number of historical range-Doppler images in the training image set, represents the pixel loss value of the a-th historical range-Doppler image in the training image set, represents the frequency domain loss value of the a-th historical range-Doppler image in the training image set, represents the probability prediction value that the discriminator D determines that the a-th historical range-Doppler image in the training image set belongs to a real image, represents the expected value of 、 respectively represent the pixel values of the i-th pixel point in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, N1 represents the number of pixel points in the predicted noise-free image and the target noise-free image of the historical range-Doppler image, represents the structural similarity value between the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, Represents the weight of the structural similarity value, respectively represent the k-th frequency component in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, respectively represent the phases of the k-th frequency component in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, N2 represents the number of frequency components of the predicted noise-free image and the target noise-free image of the historical range-Doppler image, Represents the weight of the frequency component, Represents the weight of the phase of the frequency component, Represents the loss constant, with a value of 0.1.

[0117] In this embodiment, The value range of is 0 - 1, After taking the logarithm, it emphasizes that the penalty for small probability prediction values (such as the case close to 0) for the loss will be greater. If is too low, the loss value will increase rapidly.

[0118] In this embodiment, represents the adversarial loss value of the a-th range-Doppler image in the training image set.

[0119] Beneficial effects of the above technical solution: Determining the comprehensive loss value based on the target noise-free image and the predicted noise-free image of the historical range-Doppler image in the training image set can improve the high precision and robustness of the denoising result, enabling the denoising neural network model to generate high-quality predicted noise-free images.

[0120] Embodiment 6:

[0121] The embodiment of the present invention provides a radar image denoising processing method based on deep learning, evaluating the denoising neural network model based on the test image set, and performing a second optimization on the denoising neural network model, including:

[0122] Input each historical range-Doppler image in the test set image into the first optimized denoising neural network model, and determine the predicted noise-free image of each historical range-Doppler image in the test set image based on the output result of the denoising neural network model;

[0123] Compare the predicted noise-free images and the target noise-free images of all historical range-Doppler images in the test set image to determine the denoising effect of the denoising neural network model;

[0124] Judge whether the denoising effect meets the preset denoising expectation. If not, adjust the model parameters of the denoising neural network model based on the denoising effect, and perform a second optimization on the denoising neural network model until the denoising effect meets the preset denoising expectation.

[0125] In this embodiment, each historical range-Doppler image in the test set is taken as an input and fed into the noise reduction neural network model optimized in the first round. The model extracts features from the input image and generates the corresponding predicted noise-free image by combining the parameters trained and optimized previously, that is, the noise reduction output of the model for the test data.

[0126] In this embodiment, the predicted noise-free images output by the model are compared one by one with the target noise-free images in the test set. The target noise-free images represent the ideal denoising results, while the predicted noise-free images are the actual denoising results of the model. By comparing the two, the performance of the model in the denoising task is evaluated.

[0127] In this embodiment, the denoising effect evaluation can use quantitative metrics (such as mean squared error MSE, signal-to-noise ratio SNR, or structural similarity SSIM) to evaluate the overall denoising effect of the test set, quantify the noise reduction ability of the model, determine whether the model accurately restores the target characteristics, and effectively suppress noise at the same time.

[0128] In this embodiment, a comparison is made with the preset denoising expectation, which is the lower limit of the performance index set according to the actual application requirements (such as MSE being less than a certain value, SNR reaching a certain threshold, etc.). If the denoising effect of the model is lower than this expectation, it indicates that the current model performance is still insufficient and further optimization is required.

[0129] In this embodiment, if the denoising effect does not meet the expectation, the model parameters (such as convolutional kernel weights, learning rate, etc.) are adjusted, and the noise reduction neural network model is optimized in the second round by combining the optimization algorithm. The optimization process will further reduce the loss value and improve the denoising effect of the model on the test set. This process is iterative until the model performance reaches or exceeds the preset denoising expectation, ensuring the reliability and accuracy of the model in actual applications.

[0130] Advantages of the above technical solution: By evaluating the noise reduction neural network model based on the test image set and performing the second optimization on the noise reduction neural network model, the denoising effect can be accurately quantified, a performance closed-loop can be formed by gradually refining the adjustment of the model, and the noise reduction accuracy in complex environments can be improved.

[0131] Embodiment 7:

[0132] An embodiment of the present invention provides a method for radar image noise reduction processing based on deep learning, which performs noise reduction processing on a real-time radar image data set determined from real-time radar echo data based on an optimized noise reduction neural network model, including:

[0133] Obtain real-time radar echo data and determine a real-time radar image set based on the real-time radar echo data;

[0134] Input the real-time radar image set into the second optimized noise reduction neural network model, and determine the noise reduction processing result of the real-time radar image set based on the output result of the noise reduction neural network model.

[0135] In this embodiment, the radar device continuously receives the real-time echo signals of the target area. These echo data are converted into visual two-dimensional range-Doppler images after signal processing, which can reflect the distance, speed of the target object and the noise distribution in the surrounding environment in real time.

[0136] In this embodiment, the continuously acquired radar echo data are divided according to the pulse period of the radar to generate a real-time radar image set.

[0137] In this embodiment, the real-time radar image set is input into the noise reduction neural network model that has completed the second round of optimization.

[0138] In this embodiment, the model processes each input real-time range-Doppler image one by one to generate the corresponding predicted noise-free image. These images remove the environmental noise and retain the target features, and the final output is the noise reduction processing result of the real-time radar image set.

[0139] The beneficial effects of the above technical solution: By performing noise reduction processing on the real-time radar image data set determined by the optimized noise reduction neural network model, the noise reduction adaptability of the radar echo data in dynamic scenes can be improved.

[0140] Embodiment 8:

[0141] An embodiment of the present invention provides a method for radar image noise reduction processing based on deep learning. The real-time radar image set includes multiple real-time range-Doppler images;

[0142] The noise reduction processing result includes the predicted noise reduction images of each real-time range-Doppler image in the real-time radar image set.

[0143] In this embodiment, the noise reduction neural network model processes each real-time range-Doppler image in the real-time radar image set one by one and outputs the corresponding predicted noise reduction image. The predicted noise reduction image retains the distance and speed features of the target object, and at the same time removes the environmental noise, interference signals and system errors, making the radar image clearer and more accurate. Finally, the noise reduction processing result consists of all the predicted noise reduction images of the real-time radar image set, providing a complete noise reduction view.

[0144] The beneficial effects of the above technical solution: Determining the real-time radar image set and the noise reduction processing result can ensure the accuracy and continuity of the noise reduction process and output a high-quality predicted noise reduction image set.

[0145] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without creative efforts.

[0146] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on such an understanding, the essence of the above technical solution, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0147] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for radar image noise reduction processing based on deep learning, characterized in that, Including: 101: Obtain historical echo signal data and determine a historical radar image dataset based on the historical echo signal data; 102: Determine a training image set and a test image set based on the historical radar image dataset, and determine a comprehensive loss value based on the target noise-free image of the historical range-Doppler image in the training image set and the predicted noise-free image determined by the noise reduction neural network model; Compare the comprehensive loss value with a preset loss threshold. If the comprehensive loss value is greater than or equal to the preset loss threshold, perform a first optimization on the noise reduction neural network model based on an optimization algorithm until the comprehensive loss value is less than the preset loss threshold; 103: Evaluate the noise reduction neural network model based on the test image set and perform a second optimization on the noise reduction neural network model; 104: Perform noise reduction processing on the real-time radar image dataset determined from the real-time radar echo data based on the optimized noise reduction neural network model; Among them, determining the comprehensive loss value based on the target noise-free image and the predicted noise-free image of the historical range-Doppler image in the training image set includes: ； ； ； ； Among them, represents the comprehensive loss value, represents the loss value of the a-th historical range-Doppler image in the training image set, respectively represent the maximum loss value and the minimum loss value of all historical range-Doppler images in the training image set, Nu represents the number of historical range-Doppler images in the training image set, represents the pixel loss value of the a-th historical range-Doppler image in the training image set, represents the frequency-domain loss value of the a-th historical range-Doppler image in the training image set, represents the probability prediction value that the discriminator D determines that the a-th historical range-Doppler image in the training image set belongs to a real image, represents the expected value of , respectively represent the pixel values of the i-th pixel point in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, N1 represents the number of pixel points in the predicted noise-free image and the target noise-free image of the historical range-Doppler image, represents the structural similarity value between the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, represents the weight of the structural similarity value, respectively represent the k-th frequency component in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, respectively represent the phases of the k-th frequency component in the predicted noise-free image and the target noise-free image of the a-th historical range-Doppler image in the training image set, N2 represents the number of frequency components in the predicted noise-free image and the target noise-free image of the historical range-Doppler image, represents the weight of the frequency component, represents the weight of the phase of the frequency component, represents the loss constant, with a value of 0.

1.

2. The method for radar image noise reduction processing based on deep learning according to claim 1, characterized in that, Obtaining historical echo signal data and determining a historical radar image dataset based on the historical echo signal data includes: Obtain historical echo signal data of the radar within a historical specified time period, where the historical echo signal data includes radar signals at each time point within the historical specified time period; Determine the segmented time windows of the historical specified time period based on the pulse period, and divide the historical echo signal into multiple time echo signals based on the segmented time windows; Based on the pulse compression algorithm, convert each time echo signal into the corresponding range information respectively, and determine the range dimension information of each time echo signal, where the range dimension information includes the echo intensity of multiple range cells; Perform Fourier transform on the range dimension data of each time echo signal to determine the Doppler frequency information of each range dimension data; Determine the three-dimensional matrix of the historical echo signal data based on all the time echo signals, the range dimension information of all the time echo signals, and the Doppler frequency information of all range cells in the range dimension information; Determine the two-dimensional matrix at each time point within the historical specified time period based on the time dimension of the three-dimensional matrix; Determine the historical range-Doppler image of each segmented time window based on the two-dimensional matrices at all time points within each segmented time window, and perform time marking on the historical range-Doppler image of each segmented time window; Based on the time marking, perform sequential combination on the historical range-Doppler images of all segmented time windows to determine the historical radar image dataset.

3. A method for radar image noise reduction processing based on deep learning according to claim 2, characterized in that Determining the training image set and the test image set based on the historical radar image dataset includes: Perform preprocessing on the historical radar image dataset, where the preprocessing includes image quality inspection, image normalization, and image size adjustment; Determine the target noise-free image of each historical range-Doppler image in the preprocessed radar image dataset; Determine the division ratio of the historical radar image dataset, and determine the training image set and the test image set based on the division ratio, the time marking of all historical range-Doppler images in the historical radar image set, and the target noise-free image.

4. A method for radar image noise reduction processing based on deep learning according to claim 3, characterized in that, Determining a predicted noise-free image based on a denoising neural network model, including: The input layer of the denoising neural network model receives all historical range-Doppler images in the training image set; Based on multiple convolutional layers of the denoising neural network model and the non-linear activation function of each convolutional layer, spatial feature extraction and noise feature extraction are performed on each historical range-Doppler image received by the multi-input layer to determine the local feature vector of each historical range-Doppler image; Based on multiple deep convolutional layers of the denoising neural network model, fine-grained feature extraction is performed on each historical range-Doppler image in the training image set to determine the fine-grained feature vector of each historical range-Doppler image; Using multiple deconvolutional layers of the denoising neural network model to reconstruct the local feature vector and the fine-grained feature vector of each historical range-Doppler image in the training image set respectively, and determine the predicted noise-free image of each historical range-Doppler image.

5. A method for radar image noise reduction processing based on deep learning according to claim 4, characterized in that, Determining a comprehensive loss value based on the target noise-free image and the predicted noise-free image of the historical range-Doppler images in the training image set, including: Input each historical range-Doppler image in the test set image into the first optimized denoising neural network model, and determine the predicted noise-free image of each historical range-Doppler image in the test set image based on the output result of the denoising neural network model; Compare the predicted noise-free images and the target noise-free images of all historical range-Doppler images in the test set image to determine the denoising effect of the denoising neural network model; Judge whether the denoising effect meets the preset denoising expectation. If not, adjust the model parameters of the denoising neural network model based on the denoising effect, and perform the second optimization on the denoising neural network model until the denoising effect meets the preset denoising expectation.

6. A radar image denoising processing method based on deep learning according to claim 1, characterized in that Performing denoising processing on the real-time radar image data set determined by the real-time radar echo data based on the optimized denoising neural network model, including: Obtain real-time radar echo data, and determine a real-time radar image set based on the real-time radar echo data; Input the real-time radar image set into the second optimized denoising neural network model, and determine the denoising processing result of the real-time radar image set based on the output result of the denoising neural network model.

7. A method for radar image noise reduction processing based on deep learning according to claim 1, characterized in that The real-time radar image set includes multiple real-time range-Doppler images; The denoising processing result includes the predicted denoised image of each real-time range-Doppler image in the real-time radar image set.

Images