Radar single target detection method, system and equipment based on convolutional neural network

Abstract

The invention discloses a radar single target detection method, system and device based on a convolutional neural network, and belongs to the technical field of radar target detection. The radar single target detection method comprises the following steps: firstly, preprocessing radar echo data, including orthogonal phase detection pulse pressure and the like, to obtain a one-dimensional fast time distance sequence of a target; and then the image is properly deformed to serve as an input data image, and all-zero vectors and zero-one vectors serve as output labels to carry out classification training. A convolutional neural network is used for training, hyper-parameters are properly adjusted, so that input and output label errors meet requirements, and training network parameters are obtained. And finally, carrying out target detection and distance parameter estimation on the received data by utilizing the network parameters. Simulation experiments prove that compared with a traditional radar target detection method, the method has obvious advantages in the aspects of accuracy, real-time performance and the like.

Description

Technical Field

[0001] The present invention relates to the field of radar target detection technology, and in particular to a radar single target detection method, system and device based on a convolutional neural network. Background Art

[0002] A neural network is a highly complex nonlinear learning system that can be used for predictive modeling, adaptive control, and training through data sets. Artificial neural networks have been successfully applied to speech recognition, image analysis, and adaptive control in the fields of autonomous robots, radar, and communications. Common neural networks include BP neural networks, radial basis neural networks, perceptron neural networks, self-organizing neural networks, and feedback neural networks. A deep convolutional neural network (DCNN) is a special neural network that consists of multiple convolutional layers, pooling layers, and fully connected layers. This network structure performs well in processing high-dimensional data such as images and videos because it has the ability to automatically learn image features. The LeNet-5 model is the pioneer of deep convolutional neural networks, and it is widely used in tasks such as handwritten digit recognition. The AlexNet model is the foundation of deep convolutional neural networks, which has led to a new round of development in the field of deep learning; the design ideas and network structures of VGG, GoogleNet / Inception, YOLO, ResNet, and other models are different, but they have all made important contributions to the development of convolutional neural networks.

[0003] Radar target detection is one of the core functions of radar systems. Its task is to process the received signals, extract the target's position, speed, shape and other features, and achieve target recognition and tracking. In order to solve these problems, researchers have proposed many radar target detection algorithms, such as constant false alarm rate detection algorithm, target detection in complex environments, sea surface target detection algorithm, target detection method for passive multi-base radar, and cooperative target detection method based on FDA array. In addition, with the development of artificial intelligence and machine learning technology, researchers have also begun to apply neural network technologies such as deep learning to radar target detection, and have achieved good results.

[0004] Su Ningyuan et al. used the long short-term memory network (LSTM) and convolutional neural network (CNN) to process the amplitude information of the signal time series from the two perspectives of signal prediction and feature classification, respectively, for the detection of one-dimensional sequence radar signals of maritime targets. This method does not need to assume the background statistical characteristics in advance and has stronger generalization ability. Liu Junwei et al. proposed a deep learning radar target detection method based on conditional generative adversarial network, which achieved good detection results. Chen Xiaolong et al. designed a dual-channel convolutional neural network detector to jointly extract amplitude features and time-frequency features from the amplitude sequence and time-frequency spectrum of the echo unit to improve the performance of the detector, and the false alarm is controllable. Wang Jin gang et al. designed a maritime target detector based on the attention mechanism, which extracted the high-dimensional difference features of the target and clutter from the echo to achieve detection. Chen Simin converted the phase of the echo unit into a graph and used a graph convolutional neural network to achieve target detection. Wang Zhifei et al. used the data features between echo frames and deep learning to achieve radar target detection. In order to enhance the detection performance by using the sequence feature differences between the target and clutter, Wan Hao proposed a detection method based on LSTM network. WangYumiao used the complex-valued UNet network to implement a small target detection method on the sea surface with controllable false alarm rate.

[0005] Although the above methods can all achieve single target detection by radar, due to the complexity of the network structure, the inference time of the deep learning model may be long, and the problem of insufficient real-time performance may occur when performing radar target detection. The radar system has extremely high requirements for real-time performance, which is bound to affect the response speed and real-time decision-making ability of the radar system. In addition, detection and estimation are separated in the above methods, and the radar target detection and estimation cannot be achieved synchronously, which will result in a longer total time required for target detection and estimation. Summary of the invention

[0006] In view of the above-mentioned problems, the present invention aims to provide a radar single target detection method, system and device based on convolutional neural network. After performing orthogonal phase detection processing on the radar echo data, a CNN with a multi-layer two-dimensional convolution structure is used to detect the radar single target and estimate the target distance, which has obvious advantages in terms of accuracy and real-time performance.

[0007] In order to achieve the above object, the technical solution adopted by the present invention is as follows:

[0008] The radar single target detection method based on convolutional neural network includes the following steps:

[0009] S1: Preprocess the radar echo data to obtain the one-dimensional fast time distance series of the target;

[0010] S2: Build a convolutional neural network for detecting single radar targets and train it;

[0011] S3: Use the network parameters trained in step S2 to perform target detection and distance parameter estimation on the data preprocessed in step S1.

[0012] Furthermore, the specific operation of step S1 includes the following steps:

[0013] S101: Establishing a return signal model after the radar transmitting signal encounters the target;

[0014] S102: Down-convert the radar return signal, and then perform orthogonal phase detection to obtain an echo signal model, that is, a one-dimensional fast time distance sequence of the target.

[0015] Furthermore, the echo signal model described in step S102 is expressed as

[0016] x′(t)=x(t)+n(t)

[0017] Among them, x′(t) is the echo signal, n(t) is the noise; A is the amplitude of the echo; u(tt r ) is the delay of the echo signal; The delay caused by the target, R t is the target distance, c is the speed of light; is the Doppler frequency of the target, v r is the target radial velocity, and λ is the radar operating wavelength.

[0018] Furthermore, the convolutional neural network described in step S2 includes six convolutional layers and two fully connected layers, each convolutional layer has 256 convolution kernels, the convolution kernel size of the first two convolutional layers is 3×3, and the convolution kernel size of the last four convolutional layers is 2×2, and each convolutional layer is followed by a batch normalization layer and a ReLU layer; after all the convolutional layers, there are two fully connected layers, after the first fully connected layer, there is a ReLU activation function and a dropput layer, and after the second fully connected layer, a sigmoid activation function is used to induce the classification probability and output it.

[0019] Furthermore, the convolutional layer network parameters of the convolutional neural network are Where W e Indicates dimension c e ×f e ×f e ×n e The convolution kernel, c e Indicates the number of input channels, f e Represents the length and width of the convolution kernel, n e represents the number of convolution kernels, that is, the number of output channels, be Indicates dimension n e The bias vector of .

[0020] Furthermore, the network parameters of the fully connected layer of the convolutional neural network are Among them, W 0f Indicates dimension c f ×n f The weight matrix, c f Indicates the number of input channels, n f represents the number of neurons in the fully connected layer, that is, the number of output channels, b 0f Indicates dimension n f The bias vector of .

[0021] Furthermore, the input data and output label pair of the convolutional neural network are expressed as

[0022] D train ={(k 1 , Y 1 ), (k 2 , Y 2 ),…,(k l , Y l )}

[0023] Among them, Y represents the training sample capacity network input, k is the label, l is the number of samples, if the target is detected, k is a zero-one vector, and the position of the non-zero element is the target distance, if no target is detected, k is an all-zero vector.

[0024] Furthermore, the present invention also includes a radar single target detection system based on a convolutional neural network, the detection system includes a radar data preprocessing module and a convolutional neural network module, the radar data preprocessing module is used to preprocess the radar echo data to obtain a one-dimensional fast time distance sequence of the target, and the convolutional neural network module is used to detect the radar single target;

[0025] The radar data preprocessing module and the convolutional neural network module are implemented based on the detection method as described above.

[0026] Furthermore, the present invention also includes an electronic device, which includes at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can perform the detection method as described above.

[0027] The beneficial effects of the present invention are:

[0028] 1. This paper proposes a radar single target detection method based on convolutional neural network. First, the radar echo data is preprocessed, including orthogonal phase detection pulse pressure, etc. to obtain the one-dimensional fast time distance sequence of the target. Then it is appropriately deformed as the input data image, and the all-zero vector and zero-one vector are used as output labels for classification training. The convolutional neural network is used for training, and the hyperparameters are appropriately adjusted so that the input and output label errors meet the requirements, and the training network parameters are obtained. Finally, the network parameters are used to perform target detection and distance parameter estimation on the received data. The radar target detection method is experimentally verified in the present invention. The results show that the method has high performance in radar target detection, can automatically learn features in the image, and avoid the tediousness and subjectivity of manual feature extraction. At the same time, the method can also improve the accuracy and recall rate of radar target detection, as well as real-time performance. Therefore, the method has broad application prospects and can provide radar systems with more accurate and faster target detection capabilities.

[0029] 2. The radar target single target detection method in the present invention can realize the detection and estimation of radar targets at the same time, which greatly saves the total time of target detection and estimation, and the accuracy is also improved accordingly. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is the single-target fast time-dimensional radar signal echo diagram in the present invention.

[0031] Figure 2 Schematic diagram of the convolutional neural network structure in the present invention.

[0032] Figure 3 This is a graph of learning rate changes during convolutional neural network training in the simulation experiment of the present invention.

[0033] Figure 4 This is a graph showing the change in the loss function during convolutional neural network training in the simulation experiment of the present invention.

[0034] Figure 5 It is the output result of the probability of no target induction in the simulation experiment of the present invention.

[0035] Figure 6 It is the target induction probability output result in the simulation experiment of the present invention.

[0036] Figure 7 These are the target distance estimation results for 50 times in the simulation experiment of the present invention. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the technical solution of the present invention, the technical solution of the present invention is further described below in conjunction with the accompanying drawings and embodiments.

[0038] Embodiment 1:

[0039] Embodiment 1 provides a radar single target detection method based on convolutional neural network, and the detection method comprises the following steps:

[0040] S1: Preprocess the radar echo data to obtain the one-dimensional fast time distance series of the target;

[0041] The radar uses a transmitter to transmit radar signals. In the process of the signal propagating from the radar to the target and returning from the target to the radar, the signal will be affected by factors such as atmospheric refraction, attenuation, and multipath effects, causing the amplitude and phase of the signal to change. After the radar receiver receives the echo signal, it needs to perform amplification, filtering, mixing, and other processing to extract information such as the distance and speed of the target. Specifically, the preprocessing of radar echo data in the present invention includes the following steps:

[0042] S101: Establishing a return signal model after the radar transmitting signal encounters the target;

[0043] Let the radar transmit signal e(t) be

[0044]

[0045] Wherein, u(t) is the baseband transmission signal, which can be a linear frequency modulation signal, a phase coded signal, and the most commonly used linear frequency modulation signal pulse train; f0 is the carrier frequency; is the initial phase. Then the radar return signal r(t) after the transmitted signal encounters the target is:

[0046]

[0047] Among them, A is the amplitude of the echo; u(tt r ) is the delay of the echo signal; The delay caused by the target is exactly the parameter that needs to be solved, R t is the target distance, c is the speed of light; is the Doppler frequency of the target, which is also a parameter of concern to the radar system. r is the target radial velocity, and λ is the radar operating wavelength.

[0048] S102: down-convert the radar return signal, and then perform orthogonal phase detection to obtain an echo signal model, that is, a one-dimensional fast time distance sequence of the target;

[0049] After obtaining the radar return signal, the radar system needs to go through one or more stages of down-conversion processing, and then through orthogonal phase detection to obtain the echo signal x(t):

[0050]

[0051] If noise is added, the echo signal is further expressed as

[0052] x′(t)=x(t)+n(t) (4)

[0053] It should be noted that if x′(t) here is a compressible pulse train, then the traditional range-Doppler map can be obtained by pulse compression, matching filter group, and then finding the modulus value. For simplicity, the present invention only considers the results after one-dimensional fast time pulse compression. For the convenience of processing, it is assumed that the target occupies only one distance unit and the noise obeys Gaussian white noise. Assuming that the distance unit is 100 and the target position is in the 50th distance unit, the fast time dimension signal received and sampled is as shown in the attached figure. Figure 1 The object of the present invention is to determine whether a target exists, and if a target exists, to estimate the distance unit of the target.

[0054] Step S2: construct a convolutional neural network for detecting a single radar target and perform training;

[0055] The input layer of the convolutional neural network is used to obtain the input data of the convolutional neural network, and its own features are retained. In the convolutional neural network that processes images, the input layer converts the image data into matrix pixel data and inputs it into the neural network. The first two parameters in the matrix represent the length and width of the input image, and the third parameter is the depth, which represents the number of channels of the image input. The function of the input layer is to perform some preliminary processing operations on the input image, including normalization, de-averaging, etc. The output layer is the last layer of the convolutional neural network, which is used to output the prediction results of the model. In general, the input and output of the convolutional neural network are to achieve the training and prediction tasks of the model.

[0056] The present invention adopts a supervised learning method (network training based on data and labels given in advance), and samples the classic one-dimensional fast time echo data according to formula (4) to obtain a data vector r. Assuming that the detection signal is M distance units, the input data dimension is M×1×1, that is, the length×width is M×1, and the number of channels is 1. It should be noted that the present invention adopts a convolutional neural network, and the length and width of the data are preferably in the form of a matrix. Therefore, the present invention converts its data into an N×N×1 square matrix data Y (in fact, this data can be a distance-Doppler two-dimensional image) as a network multi-channel input. The output is an M×1-dimensional signal vector. If there is no target, the output probability is relatively small. Here we define that if it does not exceed 0.5, there is no target. If it exceeds 0.5, the position of the maximum element is the distance value of the target. The target distance traverses all distance units, and randomly extracts training sets and test sets at a ratio of 8:2, with a sample number of l.

[0057] During training, the position of the non-zero element in the signal vector (target distance) is used as the label, denoted as k; for example, the received signal sampling data of the radar fast time dimension is 10 distance units. If there is a target in the 7th distance unit, then k = [0,0,0,0,0,0,1,0,0]; if there is no target, then the label k = [0,0,0,0,0,0,0,0,0,0].

[0058] The network's training set input data and output label pairs are represented as follows:

[0059] D train ={(k 1 ,Y 1 ),(k 2 ,Y 2 ),…,(k l ,Y l )} (5)

[0060] Among them, Y represents the training sample capacity network input, k is the label, and l is the number of samples.

[0061] The construction of the target detection network usually includes the following steps: Determine the network structure: Usually a transfer learning network is used, based on some existing classic network structures, such as AlexNet, VGG, ResNet, etc., to add a detection head at the end of the network. Loss function design: Commonly used loss functions for target detection tasks include cross entropy loss and L1 loss. Training and optimization: Use training data to train the target detection network, and continuously optimize the parameters of the network through the back propagation algorithm. During the training process, some optimization algorithms, such as gradient descent, Adam, etc., can be used to accelerate the convergence of the network. Testing and evaluation: After the training is completed, the target detection network needs to be tested and evaluated.

[0062] The convolutional neural network structure used to detect a single radar target in the present invention is shown in the attached figure. Figure 2 As shown in the figure, a CNN with a multi-layer two-dimensional convolutional structure is used for single target detection and target distance estimation. The network has a total of six convolutional layers and two fully connected layers. Each convolutional layer has 256 convolution kernels. The convolution kernel size of the first two convolutional layers is 3×3, while the size of the next four is 2×2. Each convolutional layer is followed by a batch normalization layer and a ReLU layer; after all convolutional layers, there are two fully connected layers. After the first fully connected layer, there is a ReLU activation function and a dropput layer. After the second fully connected layer, a sigmoid activation function is used to induce the classification probability and output it. Since each convolutional layer uses edge zero padding, only the number of channels is changed, while the length and width of the data remain unchanged.

[0063] The network parameters of the convolutional layer in the present invention are Where W e Indicates dimension ce ×f e ×f e ×n e The convolution kernel, c e Indicates the number of input channels, f e Represents the length and width of the convolution kernel, n e represents the number of convolution kernels (i.e. the number of output channels), b e Indicates dimension n e The bias vector of . Represents the convolution operation, and the operation of each convolution layer is:

[0064] ① Convolutional layer Conv-1 is used to extract features from input data. It uses ReLU activation function. e =3,n e =256. The specific operation can be expressed as:

[0065]

[0066] Where Y represents the training sample capacity network input.

[0067] ② Convolutional layers Conv-2 to Conv-5 realize nonlinear mapping of features, and all use ReLU activation function. e =256,n e =256. The specific operation can be expressed as:

[0068]

[0069] ③Convolutional layer Conv-6 is a data reconstruction layer, which obtains data output with a dimension of M×M×256. e =256,n e =256. The specific operation can be expressed as:

[0070]

[0071] The network parameters of the fully connected layer are Among them, W 0f Indicates dimension c f ×n f The weight matrix, c f Indicates the number of input channels, n f represents the number of neurons in the fully connected layer (i.e., the number of output channels), b 0f Indicates dimension n f The bias vector of . * represents the matrix multiplication operation, and the operation of each fully connected layer is:

[0072] ① The fully connected layer FC-1 is used to integrate the convolutional layer information, which uses the ReLU activation function. f=M×M×256, n f =1024. The specific operation can be expressed as:

[0073]

[0074] ② The fully connected layer FC-2 is used to induce the classification probability and output it, using the sigmoid activation function. f =1024,n f =N. The specific operation can be expressed as:

[0075]

[0076] In the formula, is the actual network output obtained by the training data Y through the DL network, There are two cases for the value of , one is the label without target and the other is the label with target. During the network training process, the cross entropy between the label and the network output is used as the loss function, that is:

[0077]

[0078] In addition, since the DL network has many parameters and the optimization problem is complex, during the network training process, as the number of iterations increases, the learning rate should be gradually reduced so that the network parameters gradually converge to the optimal or near-optimal state.

[0079] In order to prevent the network from overfitting, the present invention adopts the Dropout strategy to randomly remove neurons with a certain probability to alleviate the network's dependence on the distribution of training set data. It is getting closer and closer to k. When the network training is completed, it is tested. Similar to the way training samples are produced during the training process, the test samples are also input into the DL network in the form of multi-channel data Y to obtain the network output and check whether it is close to the true label k, so as to judge the network training level.

[0080] In addition, the present invention uses the back propagation algorithm to minimize the loss function to train the network parameters, and adopts the Adam optimizer to update the network parameters. After each forward propagation of the neural network, the error is back propagated and the network weights are updated, and the updates are repeated many times until the network objective function converges.

[0081] Further, step S3: using the network parameters trained in step S2 to perform target detection and distance parameter estimation on the data preprocessed in step S1.

[0082] Simulation experiment:

[0083] In the simulation experiment of the present invention, the deep learning toolbox in Matlab R2020b software implements the deep learning algorithm proposed in the present invention and compares it with the unit average constant false alarm detector (CFAR). Compare and analyze the target detection performance under different conditions.

[0084] Simulation conditions: 100 distance units are generated in the fast time dimension, so the size of the network input is 10×10×1. The number of targets is 1, the noise is additive Gaussian white noise, and the signal-to-noise ratio of the training samples is 10dB. Each signal-to-noise ratio generates 6,000 samples, of which 6,000 samples have targets and 6,000 samples have no targets, so the total number of samples is 12,000 samples, and the training set and test set are generated at a ratio of 10 to 2. The detection probability and false alarm probability of deep learning are statistically analyzed. The learning hyperparameters of the convolutional neural network are shown in Table 1 below.

[0085] Table 1. Learning hyperparameters of convolutional neural networks

[0086]

[0087] Simulation results:

[0088] (1) Learning rate curve

[0089] The learning rate is an important parameter for controlling the step size of the model weight update. During the training process, the learning rate will be gradually adjusted as the training progresses. The learning rate change diagram during the convolutional neural network training in the present invention is shown in the attached figure. Figure 3 As shown, from Figure 3 It can be clearly observed that the learning rate gradually decreases as the training rounds deepen.

[0090] (2) Loss function curve

[0091] During the training process, the loss function will gradually decrease as the training progresses. The loss value during the training process of the convolutional neural network in the present invention is plotted into a curve, and the result is shown in the attached figure. Figure 4 As shown in the attached Figure 4 The downward trend of the loss function can be intuitively observed, and it is finally trained to a state with good fitting.

[0092] (3) Target output results

[0093] Use the trained network to detect the target and estimate the target position of the incoming wave data. Figure 5 and attached Figure 6 The output results of the network with and without a target are given respectively. Figure 5 and attached Figure 6It can be seen that under the condition of 10dB, the output probability of no target is less than 0.5, and no false alarm occurs. The output probability of target is consistent with the target distance estimation spectrum, and the target can be detected smoothly. Figure 7 The result graph of 50 targeted estimations is given, and it can be seen that the target value can be estimated correctly.

[0094] The detection results of the radar single target detection method in the present invention are compared with the CA-CFAR detection algorithm. Based on statistical signal theory, the detection threshold commonly used in engineering is about 13.5 dB, that is, the threshold factor K of CA-CFAR detection is 4.5, so 10log10(4.5)

[0095] = 13.06 dB. Assuming that the signal-to-noise ratio received by the radar is 10 dB, this simulation experiment conducted 1000 detections and statistically analyzed the detection probability (Pd) and false alarm rate (Pfa). The comparison results of the two methods are shown in Table 2 below.

[0096] Table 2 Comparison results between the detection method of the present invention and the CA-CFAR detection method

[0097] CA-CFAR The method of the present invention Probability of detection (Pd) 63.60％ 91.02％ False alarm rate (Pfa) 7.70％ 3.46％

[0098] It can be seen from Table 2 that the detection probability of the detection method of the present invention for radar targets is much greater than that of the CA-CFAR method, and the false alarm rate is lower than that of the CA-CFAR method.

[0099] Embodiment 2:

[0100] Embodiment 2 provides a radar single target detection system based on convolutional neural network, the target detection system includes a radar data preprocessing module and a convolutional neural network module, the radar data preprocessing module is used to preprocess radar echo data to obtain a one-dimensional fast time distance sequence of the target, and the convolutional neural network module is used to detect the radar single target;

[0101] The radar data preprocessing module and the convolutional neural network module are implemented based on the detection method described in Example 1.

[0102] Embodiment three:

[0103] Embodiment three provides an electronic device, which includes at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the detection method described in embodiment one.

[0104] The above shows and describes the basic principles, main features and advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A radar single target detection method based on convolutional neural network, characterized in that: The following steps are involved: S1: Preprocess the radar echo data to obtain the one-dimensional fast time distance series of the target; S2: Build a convolutional neural network for detecting single radar targets and train it; S3: Use the network parameters trained in step S2 to perform target detection and distance parameter estimation on the data preprocessed in step S1.

2. The radar single target detection method based on convolutional neural network according to claim 1, characterized in that: The specific operation of step S1 includes the following steps: S101: Establishing a return signal model after the radar transmitting signal encounters the target; S102: Down-convert the radar return signal, and then perform orthogonal phase detection to obtain an echo signal model, that is, a one-dimensional fast time distance sequence of the target.

3. The radar single target detection method based on convolutional neural network according to claim 2, characterized in that: The echo signal model described in step S102 is expressed as x′(t)=x(t)+n(t) Among them, x′(t) is the echo signal, n(t) is the noise; A is the amplitude of the echo; u(tt r ) is the delay of the echo signal; The delay caused by the target, R t is the target distance, c is the speed of light; is the Doppler frequency of the target, vr is the radial velocity of the target, and λ is the operating wavelength of the radar.

4. The radar single target detection method based on convolutional neural network according to claim 3 is characterized in that: The convolutional neural network described in step S2 includes six convolutional layers and two fully connected layers. Each convolutional layer has 256 convolution kernels. The convolution kernel size of the first two convolutional layers is 3×3, and the convolution kernel size of the last four convolutional layers is 2×2. Each convolutional layer is followed by a batch normalization layer and a ReLU layer. After all the convolutional layers, there are two fully connected layers. After the first fully connected layer, there is a ReLU activation function and a dropput layer. After the second fully connected layer, a sigmoid activation function is used to induce the classification probability and output it.

5. The radar single target detection method based on convolutional neural network according to claim 4, characterized in that: The convolutional layer network parameters of the convolutional neural network are Where W e Indicates dimension c e ×f e ×f e ×n e The convolution kernel, c e Indicates the number of input channels, f e Represents the length and width of the convolution kernel, n e represents the number of convolution kernels, that is, the number of output channels, b e Indicates dimension n e The bias vector of .

6. The radar single target detection method based on convolutional neural network according to claim 5, characterized in that: The network parameters of the fully connected layer of the convolutional neural network are Among them, W 0f Indicates dimension c f ×n f The weight matrix, c f Indicates the number of input channels, n f represents the number of neurons in the fully connected layer, that is, the number of output channels, b 0f Indicates dimension n f The bias vector of .

7. The radar single target detection method based on convolutional neural network according to claim 6, characterized in that: The input data and output label pair of the convolutional neural network are expressed as D train ={(k 1 ,AND 1 ),(k 2 ,AND 2 ),…,(k l ,AND l )} Among them, Y represents the training sample capacity network input, k is the label, l is the number of samples, if the target is detected, k is a zero-one vector, and the position of the non-zero element is the target distance, if no target is detected, k is an all-zero vector.

8. Radar single target detection system based on convolutional neural network, characterized by: It includes a radar data preprocessing module and a convolutional neural network module, wherein the radar data preprocessing module is used to preprocess the radar echo data to obtain a one-dimensional fast time distance sequence of the target, and the convolutional neural network module is used to detect a single radar target; The radar data preprocessing module and the convolutional neural network module are implemented based on the detection method described in any one of claims 1-7.

9. An electronic device, characterized in that: It comprises at least one processor; and a memory connected to the processor in communication; wherein the memory stores instructions executable by the processor, and the instructions are executed by the processor so that the processor can execute the detection method described in any one of claims 1 to 7.

Images