Deep learning sea clutter discrimination method based on constant false alarm rate kernel

Abstract

The application discloses a deep learning sea clutter discrimination method based on a constant false alarm rate convolution kernel. Firstly, multi-frame radar signals are converted into graph data, and the space-time characteristics of the graph data are used to preliminarily suppress sea clutter. Then, a detection sliding window convolution neural network model is constructed by taking a classic constant false alarm rate detection sliding window model in radar signal processing as a convolution kernel of the convolution neural network model, and the detection sliding window convolution neural network model is used to extract features and classify graph data nodes, so that target discrimination under a sea clutter background is realized. Simulation experiments prove that the detection probability can reach 98% when the signal-to-clutter ratio is 9 dB, which is better than traditional CFAR detection. Through a similar constant false alarm rate detection method, the convolution kernel is designed according to the detection sliding window working principle in CFAR, so that the network model is more interpretable, more data characteristics are integrated, the parameter scale is smaller, and the running speed is faster.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of radar target detection and identification, and is suitable for solving the problem of radar target detection under sea clutter interference. BACKGROUND

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

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

[0004] In recent years, deep learning and artificial intelligence technology have developed rapidly. Convolutional neural network (CNN), deep belief network (DBN) and recurrent neural network (RNN) are currently widely used in target detection, which provides a new idea for the suppression of sea clutter.

[0005] The application discloses a sea clutter discrimination method based on a deep learning sea clutter discrimination method based on a constant false alarm convolution kernel. The abbreviation is DSW-CNN (Convolutional Neural Network model based on Sliding Window Detection model). First, the multi-frame radar signal is converted into graph data, and the space-time features of the graph data are used to preliminarily suppress the sea clutter. Then, the classic constant false alarm detection sliding window model in radar signal processing is used as the convolution kernel of the convolution neural network model, a detection sliding window convolution neural network model is constructed, the detection sliding window convolution neural network model is used to extract features and classify the graph data nodes, so that the target discrimination under the sea clutter background is realized. SUMMARY

[0006] The problem to be solved by the application is that the adaptability of radar to low observable target detection under the sea clutter background is provided, and a sea clutter discrimination method based on a detection sliding window convolution neural network model is provided.

[0007] A deep learning sea clutter discrimination method based on constant false alarm rate convolution kernel, comprising the following steps:

[0008] Step 1: inputting a two-dimensional signal S of radar distance dimension and pulse dimension

[0009] Step 2: low threshold detection processing

[0010] The radar signal S is detected by threshold detection, and the detection threshold of each frame is determined by the specific data of each frame. (:,k) The threshold detection is performed on each frame of echo signal, and the detection threshold of each frame is determined by the specific data of each frame.

[0011]

[0012] Wherein, p fa is the set constant false alarm rate, the function sort() is the descending arrangement of each frame of echo, and a new set of data Ss is obtained after the descending arrangement (:,k) , T k is the value of Ss (:,k) The 0.01*R term is R, which represents the number of arranged data.

[0013] After processing each frame, the array

[0014]

[0015] Wherein, r, k are the indexes of data units, indicating that the data is 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, and the data with value 1 is the data of interest, which has higher confidence of target. Since the target echo signal has continuity, it is considered that multiple continuous data with value 1 in a frame are all from the echo information of one target. Through connectivity detection, it can be avoided to judge the continuous data with value 1 as different targets, improve the detection accuracy, and reduce the input of unimportant information,

[0018] The connectivity detection is performed on each frame S' (:,k) , and n connected components are obtained, corresponding to n nodes in the frame

[0019]

[0020] Wherein, n i is the number of nodes in the i-th frame, and n is the total number of nodes in the group of data.

[0021] Step 4: Spatiotemporal auxiliary feature extraction processing

[0022] Input two-dimensional radar signal with range dimension and pulse dimension, and distance 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 on all points to achieve morphological point aggregation of target signals;

[0025] Step 5: Correlation 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 unit in two adjacent frames is less than or equal to a;

[0030] Step 6: Deep learning sea clutter identification based on constant false alarm convolution kernel

[0031] The sea clutter identification model based on the trained 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 the 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 based on 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 sea clutter identification method based on deep learning (DSW-CNN) with a constant false alarm rate convolution kernel first 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 convolutional neural network model is used to extract and classify features of the graph data nodes, thereby achieving target identification in the sea clutter background. Simulation experiments have shown that the detection probability can reach 98% at a signal-to-noise ratio of 9dB, which is superior to 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, making the network model more interpretable, while incorporating more data features, with a smaller parameter size and faster operation speed. BRIEF DESCRIPTION OF THE DRAWINGS

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

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

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

[0039] Attachment 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] Refer to the 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 radar signal S in range and pulse dimensions;

[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 each frame. The detection threshold can be calculated using the following formula

[0045]

[0046] where p fa is the set constant false alarm rate, the function sort() is to arrange each frame echo in descending order, and a new set of data Ss (:,k) is obtained after arranging in descending order k , and the value of T (:,k) is the 0.01*R term, and R represents the number of arranged data;

[0047] After each frame is processed, the array

[0048]

[0049] where r and k are the indexes of the data unit, indicating that the data is 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, and the data with a value of 1 is the data of interest, which has a higher confidence of the target. Since the target echo signal has continuity, it is considered that multiple continuous data with a value of 1 in a frame are all from the echo information of one target. Through connectivity detection, it can be avoided to judge the continuous data with a value of 1 as different targets, improve the detection accuracy, and reduce the input of unimportant information,

[0052] The connectivity detection is performed on each frame S' (:,k) , and n connected components can be obtained, corresponding to n nodes in the frame

[0053]

[0054] where n i is the number of nodes in the ith frame, and n is the total number of nodes in the set of data. The time-space auxiliary feature and signal feature values can be obtained from each node;

[0055] Step 4: time-space auxiliary feature extraction processing;

[0056] The input is a two-dimensional radar signal of distance dimension and pulse dimension, and the distance auxiliary feature F a ={r, k, c} is composed of the center distance unit r i of the node, the frame number k i , and the size c i of each node, where r i is:

[0057]

[0058] The time-space auxiliary feature extraction processing is performed on all points, and point track condensation is realized on the morphology of the target signal.

[0059] Step 5: correlation detection processing

[0060] In the multi-frame data, the target echo signal and the sea clutter signal can be better identified through the correlation of nodes between frames, and the adjacent two frames are correlated.

[0061]

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

[0063] Wherein, V i,j The upper limit is a, which represents that the absolute value of the difference between the node center distances of adjacent two frames is less than or equal to a.

[0064] Step 6: sea clutter discrimination based on constant false alarm rate convolution kernel deep learning

[0065] The sea clutter discrimination model based on constant false alarm rate convolution kernel deep learning is used for sea clutter discrimination processing of the two-dimensional radar signal diagram, and the output is the signal diagram after filtering the sea clutter.

[0066] Step 61: according to the U-Net network structure, the convolutional neural network is constructed, and in the network model convolution kernel design, the constant false alarm detection sliding window model in the traditional signal processing is used as the convolution kernel, and the detection unit, the protection unit and the reference unit are 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] After the detection sliding window convolutional neural network model is built, the calibrated sea clutter and target data set are used for training, and the trained network is used for sea clutter discrimination and filtering.

[0068] Step 63: the sea clutter discrimination processing of the two-dimensional radar signal diagram is performed by using the trained convolutional neural network model based on the detection sliding window, and the output is the signal diagram after filtering the sea clutter.

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

[0070] Simulation environment: when SCR=6dB, the simulation data is as shown in the accompanying Figure 2 Under the condition that the signal-to-clutter ratio is 6dB, the target has only individual frame intensity, and the overall sea clutter background has no obvious edge.

[0071] Simulation results and analysis: attached Figure 3 , attached Figure 4 The CFAR algorithm and the DSW-CNN network are respectively compared and displayed. It can be seen from them that although the detection result obtained by using the CFAR algorithm can be analyzed by experience to find the target, the error detection value is more, and the clutter suppression is not complete. The detection result of the proposed DSW-CNN network is more accurate due to more network layers and larger parameter size, and has a good prediction and completion effect on the weak target signal. The detection result of the DSW-CNN network is obviously better than that of the CFAR algorithm, but due to the lightweight discarding of the deep network structure, the parameter size is small, which leads to the output result being more blurred than the output result of the teacher network, and the weak target signal has a certain loss.

Claims

1. A deep learning sea clutter discrimination method based on constant false alarm rate convolution kernel, characterized in that, The method comprises the following steps: Step 1: inputting a two-dimensional signal S of radar distance and pulse; Step 2: low threshold detection processing; The radar signal S (:,k) Each frame of echo signal is threshold detected, the detection threshold of each frame is determined by the specific data of the frame, and the detection threshold is calculated by the following formula: where p fa is the set constant false alarm rate, the function sort() is to arrange each frame of echo in descending order, and a new set of data Ss (:,k) is obtained after the descending arrangement k , and T (:,k) is the 0.01*R term, where R represents the number of arranged data Each frame, after processing, results in an array Wherein, r, k are indexes of data units, indicating that the data is the rth distance unit in the kth frame; Step 3: connectivity detection processing; After low threshold detection, S' is a 0, 1 matrix, and the data with a value of 1 is the data of interest, which has a higher confidence of the target. Since the target echo signal has continuity, it is considered that multiple continuous data with a value of 1 in a frame are all from the echo information of one target. Through connectivity detection, continuous data with a value of 1 is avoided to be judged as different targets, the detection accuracy is improved, and the input of unimportant information is reduced. For each frame S' (:,k) Perform connectivity detection to obtain n connected components, corresponding to n nodes in the frame wherein n i is the number of nodes in the i-th frame, n is the total number of nodes in the group of data, and the spatio-temporal auxiliary feature and the signal feature are obtained from each node. Step 4: spatio-temporal auxiliary feature extraction processing; Input two-dimensional radar signal with range dimension and pulse dimension, and distance 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: Spatio-temporal auxiliary feature extraction processing is performed on all points to realize point trace condensation of the target signal in morphology; Step 5: correlation detection processing; In multiple frames of data, the nodes between frames are associated to better identify target echo signals and sea clutter signals. The correlation between adjacent two frames is performed: V i,j = abs(r i - r j ) V i,j The upper limit a represents that the absolute value of the difference between the node center distances of two adjacent frames is less than or equal to a. Step 6: deep learning sea clutter discrimination based on constant false alarm rate convolution kernel; The trained deep learning sea clutter discrimination model based on constant false alarm rate convolution kernel is used to perform sea clutter discrimination processing on the two-dimensional radar signal graph, and the output is a signal graph filtered of sea clutter.

2. The deep learning sea clutter discrimination method based on constant false alarm rate kernel according to claim 1, characterized in that, The deep learning sea clutter discrimination model based on constant false alarm rate convolution kernel in step 6 comprises: (1) After the deep learning network model based on constant false alarm rate convolution kernel is built, the calibrated sea clutter and target data set are used for training, and the trained network is used for sea clutter discrimination and filtering; (2) According to the U-Net network structure, a convolutional neural network is constructed. In the design of the network model convolution kernel, the constant false alarm detection sliding window model in traditional signal processing is used as the convolution kernel. The detection unit, the protection unit and the reference unit are 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.

Images