Deep learning sea clutter identification method based on constant false alarm convolution kernel

Abstract

The invention discloses a deep learning sea clutter identification method based on a constant false alarm convolution kernel. Firstly, multi-frame radar signals are converted into graph data, sea clutters are preliminarily suppressed by using spatial-temporal characteristics of the graph data, then, a detection sliding window convolutional neural network model is constructed by taking a classical constant false alarm detection sliding window model in radar signal processing as a convolution kernel of the convolutional neural network model, and the detection sliding window convolutional neural network model is used for detecting the sea clutters. And feature extraction and classification are carried out on graph data nodes by using the provided detection sliding window convolutional neural network model, so that target identification under the sea clutter background is realized. Simulation experiments prove that the detection probability can reach 98% when the signal-to-clutter ratio is 9dB, the method is superior to traditional CFAR detection, a convolution kernel is designed through a detection method similar to a constant false alarm rate according to the working principle of a detection sliding window in CFAR, a network model is more interpretable, meanwhile, more data features are integrated, the parameter scale is smaller, and the operation speed is higher.

Description

Technical Field

[0001] The invention belongs to the field of radar target detection and recognition, and is suitable for solving the problem of radar target detection under sea clutter interference. Background Art

[0002] The ocean environment is different from the land environment. It is affected by weather, tides, the rotation of the earth and other factors. The randomness and complexity of the sea clutter generated are much greater than that of land clutter. The detection of sea targets is widely involved in both military and civilian fields. Anti-sea clutter interference is one of the main problems faced by the current sea surface target detection and recognition technology.

[0003] Therefore, how to effectively and reliably detect and identify maritime targets in the background of sea clutter is a key technology that needs to be studied. Constant False Alarm Rate (CFAR) is a common adaptive algorithm used in radar systems to detect target echoes in the background of noise, clutter and interference, but its adaptability to low observable target detection in the background of sea clutter needs to be further improved.

[0004] In recent years, deep learning and artificial intelligence technologies have flourished. Convolutional Neural Network (CNN), Deep Belief Network (DBN), Recurrent Neural Network (RNN), etc. are currently widely used to detect targets, providing a new idea for suppressing sea clutter.

[0005] The present invention discloses a sea clutter identification method based on a deep learning sea clutter identification method of a constant false alarm convolution kernel. The method is referred to as DSW-CNN (Convolutional Neural Network model based on Sliding Window Detection model). First, a multi-frame radar signal is converted into graph data, and the spatiotemporal characteristics of the graph data are used to preliminarily suppress sea clutter. Then, a detection sliding window convolution neural network model is constructed by using the classic constant false alarm detection sliding window model in radar signal processing as the convolution kernel of the convolution neural network model. The proposed detection sliding window convolution neural network model is used to extract features and classify graph data nodes, thereby realizing target identification under the background of sea clutter. Summary of the invention

[0006] The problem to be solved by the present invention is to provide a sea clutter identification method based on a detection sliding window convolutional neural network model in order to solve the adaptability problem of radar to low observable target detection under sea clutter background.

[0007] A deep learning sea clutter identification method based on constant false alarm convolution kernel includes the following steps:

[0008] Step 1: Input the two-dimensional signal S of radar range dimension and pulse dimension

[0009] Step 2: Low Threshold Detection Processing

[0010] For radar signal S (:,k) Each frame of echo signal is subjected to threshold detection. The detection threshold of each frame is determined by the specific data of the frame. The detection threshold can be calculated using the following formula

[0011]

[0012] Among them, p fa To set the constant false alarm rate, the sort() function arranges each frame of echo in descending order, and obtains a new set of data Ss after descending order. (:,k) , T k The value is Ss (:,k) The 0.01*R item, R represents the number of data after arrangement;

[0013] After each frame is processed, an array can be obtained

[0014]

[0015] Wherein, r and k are the indexes of the data unit, indicating that the data is the data of the rth distance unit in the kth frame;

[0016] Step 3: Connectivity detection processing

[0017] After low threshold detection, S' is a 0, 1 matrix. The data with a value of 1 is the data of interest, and its confidence in being the target is higher. Since the target echo signal has continuity, it is considered that multiple consecutive data with a value of 1 in one frame are all from the echo information of one target. Through connectivity detection, it can avoid judging continuous data with a value of 1 as different targets, improve its detection accuracy, and reduce unimportant information input.

[0018] For each frame S' (:,k) Connectivity detection can obtain n connected components, corresponding to the n nodes in the frame

[0019]

[0020] Among them, n i is the number of nodes in the i-th frame, n is the total number of nodes in the data set, and two feature values, spatiotemporal auxiliary features and signal features, can be obtained from each node;

[0021] Step 4: Spatiotemporal auxiliary feature extraction processing

[0022] Input range dimension and pulse dimension 2D radar signal, range auxiliary feature F a = {r, k, c}, the distance from the center of the node to the unit r i , frame number k i and the size c of each node i Composition, where r i for

[0023]

[0024] Perform spatiotemporal auxiliary feature extraction processing on all points to achieve morphological point condensation of target signals;

[0025] Step 5: Relevance detection processing

[0026] In multi-frame data, by associating nodes between frames, we can better identify target echo signals and sea clutter signals and associate two adjacent frames.

[0027]

[0028] V i,j = abs(r i -r j )

[0029] Among them, V i,j The upper limit is a, which means that the absolute value of the difference between the node center distance units of two adjacent frames is less than or equal to a;

[0030] Step 6: Deep learning sea clutter identification based on CFAR convolution kernel

[0031] The trained deep learning sea clutter identification model based on constant false alarm convolution kernel is used to perform sea clutter identification processing on the two-dimensional radar signal image, and the output is the signal image after filtering out the sea clutter.

[0032] Furthermore, the deep learning sea clutter identification model based on constant false alarm convolution kernel in step 6 is:

[0033] (1) After the deep learning network model based on the constant false alarm convolution kernel is built, it is first trained using the calibrated sea clutter and target data sets, and the trained network is used to identify and filter sea clutter;

[0034] (2) A convolutional neural network is constructed according to the U-Net network structure. In the design of the convolution kernel of the network model, the constant false alarm detection sliding window model in traditional signal processing is used as the convolution kernel. The detection unit, protection unit, and reference unit are set in the convolution kernel respectively. The weight of the protection unit is set to 0, the weight of the detection unit is set to -1, and the weight of the reference unit is used for iterative training to adapt to different sea conditions.

[0035] The beneficial effect of the present invention is that the proposed method for identifying sea clutter based on deep learning (DSW-CNN) based on constant false alarm rate convolution kernel, firstly, analyzes the characteristics of sea clutter to obtain features that can distinguish sea clutter from targets. Then, the multi-frame radar signal is converted into graph data, and the spatiotemporal characteristics of the graph data are used to preliminarily suppress sea clutter. Finally, the proposed detection sliding window convolution neural network model is used to extract and classify the graph data nodes, thereby realizing target identification under the background of sea clutter. Simulation experiments have shown that when the signal-to-noise ratio is 9dB, its detection probability can reach 98%, which is better than traditional CFAR detection. Through a detection method similar to the constant false alarm rate, the convolution kernel is designed according to the detection sliding window working principle in CFAR, so that the network model is more interpretable, and more data features are incorporated at the same time, and the parameter scale is smaller and the running speed is faster. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Attached Figure 1 is a flow chart of the method steps of the present invention;

[0037] Attached Figure 2 This is the simulation data when the signal-to-noise ratio is 6dB;

[0038] Attached Figure 3 This is the CFAR detection result when the signal-to-noise ratio is 6dB;

[0039] Attached Figure 4 This is the DSW-CNN detection result when the signal-to-noise ratio is 6dB. DETAILED DESCRIPTION

[0040] The present invention will be described in further detail below with reference to the accompanying drawings.

[0041] See attached Figure 1 , a sea clutter identification method based on detection sliding window convolutional neural network, comprising the following steps:

[0042] Step 1: Input the two-dimensional signal S of radar range dimension and pulse dimension;

[0043] Step 2: Low threshold detection processing;

[0044] For radar signal S (:,k) Each frame of echo signal is subjected to threshold detection. The detection threshold of each frame is determined by the specific data of the frame. The detection threshold can be calculated using the following formula

[0045]

[0046] Among them, p fa To set the constant false alarm rate, the sort() function arranges each frame of echo in descending order, and obtains a new set of data Ss after descending order. (:,k) , T k The value is Ss (:,k) The 0.01*R item, R represents the number of data after arrangement;

[0047] After each frame is processed, an array can be obtained

[0048]

[0049] Wherein, r and k are the indexes of the data unit, indicating that the data is the data of the rth distance unit in the kth frame;

[0050] Step 3: Connectivity detection processing;

[0051] After low threshold detection, S' is a 0, 1 matrix. The data with a value of 1 is the data of interest, and its confidence in being the target is higher. Since the target echo signal has continuity, it is considered that multiple consecutive data with a value of 1 in one frame are all from the echo information of one target. Through connectivity detection, it can avoid judging continuous data with a value of 1 as different targets, improve its detection accuracy, and reduce unimportant information input.

[0052] For each frame S' (:,k) Connectivity detection can obtain n connected components, corresponding to the n nodes in the frame

[0053]

[0054] Among them, n i is the number of nodes in the i-th frame, n is the total number of nodes in the data set, and two feature values, spatiotemporal auxiliary features and signal features, can be obtained from each node;

[0055] Step 4: spatiotemporal auxiliary feature extraction processing;

[0056] Input range dimension and pulse dimension 2D radar signal, range auxiliary feature F a = {r, k, c}, the distance from the center of the node to the unit r i , frame number k i and the size c of each node i Composition, where r i for:

[0057]

[0058] Perform spatiotemporal auxiliary feature extraction processing on all points to achieve morphological point condensation of target signals;

[0059] Step 5: correlation detection processing;

[0060] In multi-frame data, by associating nodes between frames, we can better identify target echo signals and sea clutter signals and associate two adjacent frames:

[0061]

[0062] V i,j = abs(r i -r j )

[0063] Among them, V i,j The upper limit is a, which means that the absolute value of the difference between the node center distance units of two adjacent frames is less than or equal to a;

[0064] Step 6: Deep learning sea clutter identification based on constant false alarm convolution kernel;

[0065] The trained deep learning sea clutter identification model based on constant false alarm convolution kernel is used to perform sea clutter identification processing on the two-dimensional radar signal image, and the output is the signal image after filtering out the sea clutter.

[0066] Step 61: Construct a convolutional neural network according to the U-Net network structure. In the design of the convolution kernel of the network model, the constant false alarm detection sliding window model in traditional signal processing is used as the convolution kernel. The detection unit, protection unit, and reference unit are respectively set in the convolution kernel. The weight of the protection unit is set to 0, the weight of the detection unit is set to -1, and the weight of the reference unit is used for iterative training to adapt to different sea conditions.

[0067] Step 62: After the detection sliding window convolutional neural network model is built, it is first trained using the calibrated sea clutter and target data sets, and the trained network is used for sea clutter identification and filtering;

[0068] Step 63: Use the trained convolutional neural network model based on the detection sliding window to perform sea clutter identification processing on the two-dimensional radar signal image, and output the signal image after filtering out the sea clutter.

[0069] The effect of the present invention can be further illustrated by the following simulation experiment:

[0070] Simulation environment: When SCR = 6dB, the simulation data is as shown in the attached Figure 2 When the signal-to-clutter ratio is 6dB, the target has only a few frames with high intensity, and there is no obvious edge between the target and the sea clutter background.

[0071] Simulation results and analysis: Appendix Figure 3 , Attachment Figure 4 The CFAR algorithm and DSW-CNN network are compared and demonstrated respectively. It can be seen that although the detection results obtained by the CFAR algorithm can analyze the location of the target through experience, its false detection values ​​are too high and the clutter suppression is not thorough. The proposed DSW-CNN network has more network layers and a larger parameter scale, so its detection results are more accurate and it has a better prediction and completion effect for the weaker target signal segments. The detection results of the DSW-CNN network are obviously better than the CFAR algorithm, but because of its lightweight abandonment of the deep network structure and its small parameter scale, the output results are more blurred than the teacher network output results, and the weaker target signal segments are lost to a certain extent.

Claims

1. A deep learning sea clutter identification method based on constant false alarm convolution kernel, characterized in that: The following steps are involved: Step 1: Input the two-dimensional signal S of radar range dimension and pulse dimension; Step 2: Low threshold detection processing; For radar signal S (:,k) Each frame of echo signal is subjected to threshold detection. The detection threshold of each frame is determined by the specific data of the frame. The detection threshold is calculated using the following formula: Among them, p fa To set the constant false alarm rate, the sort() function arranges each frame of echo in descending order, and obtains a new set of data Ss after descending order. (:,k) , T k The value is Ss (:,k) The 0.01*R item, R represents the number of data after arrangement; After each frame is processed, the array is obtained Wherein, r and k are the indexes of the data unit, indicating that the data is the data of the rth distance unit in the kth frame; Step 3: Connectivity detection processing; After low threshold detection, S' is a 0, 1 matrix. Data with a value of 1 is the data of interest, and its confidence in being the target is higher. Since the target echo signal is continuous, it is considered that multiple consecutive data with a value of 1 in one frame are all from the echo information of one target. Through connectivity detection, it is avoided that continuous data with a value of 1 are judged as different targets, thereby improving its detection accuracy and reducing the input of unimportant information. For each frame S' (:,k) Perform connectivity detection to obtain n connected components, corresponding to the n nodes in the frame Among them, n i is the number of nodes in the i-th frame, n is the total number of nodes in the data set, and two feature values, spatiotemporal auxiliary features and signal features, are obtained from each node; Step 4: spatiotemporal auxiliary feature extraction processing; Input range dimension and pulse dimension 2D radar signal, range auxiliary feature F a = {r, k, c}, the distance from the center of the node to the unit r i , frame number k i and the size c of each node i Composition, where r i for: Perform spatiotemporal auxiliary feature extraction on all points to achieve morphological point condensation of target signals; Step 5: correlation detection processing; In multi-frame data, by associating the nodes between frames, we can better identify the target echo signal and the sea clutter signal and associate two adjacent frames: V i,j =abs(r i -r j ) Among them, V i,j The upper limit is a, which means that the absolute value of the difference between the node center distance units of two adjacent frames is less than or equal to a; Step 6: Deep learning sea clutter identification based on constant false alarm convolution kernel; The trained deep learning sea clutter identification model based on constant false alarm convolution kernel is used to perform sea clutter identification processing on the two-dimensional radar signal image, and the output is the signal image after filtering out the sea clutter.

2. A deep learning sea clutter identification method based on constant false alarm convolution kernel according to claim 1, characterized in that: The deep learning sea clutter identification model based on constant false alarm convolution kernel in step 6 includes: (1) After the deep learning network model based on the constant false alarm convolution kernel is built, it is first trained using the calibrated sea clutter and target data sets, and the trained network is used to identify and filter sea clutter; (2) A convolutional neural network is constructed according to the U-Net network structure. In the design of the convolution kernel of the network model, the constant false alarm detection sliding window model in traditional signal processing is used as the convolution kernel. The detection unit, protection unit, and reference unit are set in the convolution kernel respectively. The weight of the protection unit is set to 0, the weight of the detection unit is set to -1, and the weight of the reference unit is used for iterative training to adapt to different sea conditions.

Images