A sky wave radar transient interference classification method based on RD pattern texture features

Abstract

The application discloses a sky wave radar transient interference classification method based on RD pattern texture features, and comprises the following steps: processing an RD pattern obtained by a radar, designing and making a picture set for training of an SVM, wherein the picture set comprises three types of no interference, single pulse linear frequency modulation signal transient interference and multi-period linear frequency modulation signal transient interference; two types of texture features are extracted from the obtained training set respectively; a linear classifier is used as a classifier for training; a transient interference multi-classifier fusion judgment algorithm is designed to complete classification of an input file.The application has the advantages that: a signal processing problem is converted into an image classification problem, two types of image texture extraction are used for feature acquisition, so that a picture set available for classifier training is obtained, transient interference is better classified, and a solid foundation is laid for subsequent transient interference suppression and other signal processing links; the processing speed is fast, and the accuracy is high.

Description

Technical Field

[0001] This invention relates to the field of radar signal processing technology, and in particular to a method for classifying transient interference in skywave radar based on range-Doppler (RD) texture features. Background Technology

[0002] Over-the-Horizon Radar (OTHR) offers advantages such as being unaffected by the Earth's curvature, increasing detection range, and providing anti-stealth capabilities. Therefore, countries worldwide, including my country, have conducted extensive research and exploration into OHR. However, the high-frequency band in OHR applications suffers from significant noise and interference, severely impacting radar performance.

[0003] Transient interference can originate from meteor trail echoes, lightning, and man-made pulses. It is unpredictable and its presence increases the radar's original noise floor, sometimes even masking the true target signal, severely reducing the radar's signal-to-noise ratio and detection capability, and creating difficulties for subsequent signal processing. Linear frequency modulated (LFM) signals are particularly typical of transient interference because they exhibit frequency sweeping and modulation characteristics. Radars operating at any frequency band are susceptible to interference from LFM signals, and its impact on the noise floor is significant.

[0004] Therefore, the identification and suppression of transient interference has significant practical implications and is urgently needed in various fields such as national defense and engineering.

[0005] To suppress transient interference, it is first necessary to classify and identify the types of transient interference. Currently, research on transient interference mainly focuses on suppression, and the proposed methods for handling transient interference mainly include spatial domain processing, temporal domain processing, and spatiotemporal adaptive processing. There is still a lack of research on the classification of transient interference. Existing technologies [Xing Mengdao, Bao Zheng, Qiang Yong. Transient interference suppression of skywave over-the-horizon radar [J]. Acta Electronica Sinica, 2002(6):823-826] mainly rely on a directly set threshold to remove the interference exceeding the threshold and then restore the signal to achieve the purpose of interference suppression. Existing technologies [Zhen Hongfei. Research on transient interference detection and suppression of skywave over-the-horizon radar [D]. Master's thesis. University of Electronic Science and Technology of China, 2011:20-36] introduce fractal theory into transient interference detection, and separate transient interference in the clutter background based on the difference in fractal dimension. It also introduces a transient interference detection method based on the generalized frequency domain and generalized sinusoidal signal model. On the basis of this model, a number of power spectrum estimation methods are introduced for transient interference detection. An interference detection method based on the periodogram method, ESPRIT algorithm and TLS-ESPRIT algorithm under this model is proposed, and the feasibility of this method is verified. This method involves a large amount of computation and a long processing time. Existing technology [Wang Ni. Research on Transient Interference Suppression Algorithm for Skywave Radar [D]. Master's Thesis. Harbin Institute of Technology, 2021: 11-23] uses a time-frequency joint analysis method to detect transient interference, which is a short-duration, high-energy local abrupt signal, by applying time-frequency transforms such as wavelet transform, short-time Fourier transform, and S-transform, thus laying the foundation for interference suppression.

[0006] The existing technology [Luo Zhongtao, Xia Hang, Lu Kun, et al. Automatic identification method of transient interference from medium-range Doppler image of over-the-horizon radar [J]. Journal of Electronics, 2021, 49(07): 1279-1285] uses the simulated RD image to extract local binary pattern (LBP) features and then inputs them into a classifier, transforming the interference detection problem into an image classification problem to achieve the purpose of interference detection.

[0007] However, the aforementioned existing technologies have the following drawbacks: large computational load, slow judgment time, and poor compatibility and applicability. Summary of the Invention

[0008] This invention addresses the shortcomings of existing technologies by providing a skywave radar transient interference classification method based on RD map texture features. It introduces image processing techniques, transforming signal processing problems into image-based recognition and classification problems. By leveraging the differentiated representations of different types of transient interference on the RD map, texture features are extracted from the image and combined with Support Vector Machines (SVM) to complete the interference detection and classification.

[0009] To achieve the above-mentioned objectives, the technical solution adopted by the present invention is as follows:

[0010] A method for classifying transient interference in skywave radar based on RD map texture features includes the following steps:

[0011] S1: Perform image processing on the RD map obtained by radar, including: strong interference removal, grayscale enhancement and size transformation. Design and create an image training set for training a linear classifier. The image training set includes three categories: no interference, transient interference of single-pulse linear frequency modulation signal, and transient interference of multi-periodic linear frequency modulation signal.

[0012] S2: Extract two types of texture features from the obtained image training set: Histogram of Oriented Gradient (HOG) features and Gray Level Concurrence Matrix (GLCM) features.

[0013] S3: Training using a linear classifier;

[0014] S4: Design a transient interference multi-classifier fusion judgment algorithm to classify the input files;

[0015] Further, step S1 specifically includes:

[0016] S11: Strong Interference Removal: First, a threshold is set based on the intensity of the entire RD image that has been acquired. Rows and columns in the matrix that exceed the threshold value are directly deleted. The threshold is calculated as follows:

[0017] threshold = median(pic) + m

[0018] In the formula, the value of m needs to be selected based on the overall image situation, pic represents the input image, median represents taking the median value of the image intensity, and threshold represents the calculated threshold.

[0019] S12: Gray-scale enhancement: Histogram equalization is used to enhance the contrast of the image obtained after step S11.

[0020] S13: Size transformation: Based on the image size obtained in S12, flexibly select the number of images to be divided, ensuring that the image sizes are the same after division.

[0021] Further, step S2 specifically includes:

[0022] S21: Extract HOG features from the image training set obtained in S1 to form feature vector A;

[0023] S22: Extract GLCM features from the image training set obtained after step S1 to form feature vector B;

[0024] S23: The extracted features are formed by fusing feature vector A and feature vector B.

[0025] Furthermore, in S2, the histogram of oriented gradients (HOGs) divides an image into several units, calculates the gradient corresponding to each pixel in a unit, and constructs an HOG histogram. The specific value of the HOG histogram is determined by both the magnitude and direction of the gradient. The histograms obtained from each unit are combined. Within a larger block region, the histograms of the contained units are concatenated to perform contrast normalization on the gradient intensity. The specific gradient calculation is shown in the following formula.

[0026] G x (x,y)=I(x+1,y)-I(x-1,y)

[0027] G y (x,y)=I(x,y+1)-I(x,y-1)

[0028]

[0029]

[0030] In the formula, G x (x, y) and G y (x, y): represent the gradients in the horizontal and vertical directions, respectively, G x (x, y) is the difference between pixel (x+1, y) and pixel (x-1, y), while G y (x, y) is the difference between pixel (x, y+1) and pixel (x, y-1).

[0031] G(x, y): represents the gradient magnitude at pixel (x, y).

[0032] θ(x, y): Represents the gradient direction at pixel (x, y).

[0033] Furthermore, the generation process of GLCM in S2 is as follows:

[0034] 1) Determine the maximum and minimum gray levels L in the image. max L min Then the range of values ​​for any pair of points (i, k) in the image is (L... min L min ) and (L max L max )between;

[0035] 2) Based on the specific distance d and angle θ requirements, count the number of occurrences of each point pair (i, j) in the entire image and arrange them into a matrix;

[0036] 3) Normalize the number of occurrences of (i, j) to obtain the probability P(i, j), and form the GLCM matrix.

[0037] 4) Based on the obtained GLCM matrix, four statistical measures were constructed as features: contrast, correlation, energy, and homogeneity. The mean and standard deviation of the corresponding statistical measures were calculated as features.

[0038] Further, step S3 specifically includes:

[0039] By combining the image training set, HOG, GLCM, and a linear classifier, the linear classifier is trained to obtain the trained classifier.

[0040] Further, step S4 specifically includes:

[0041] Based on the trained classifier obtained in S3, the RD image to be detected is first processed according to the steps in S1, and then judged according to the designed transient interference detection judgment algorithm to obtain the final classification result of the RD image.

[0042] Furthermore, the linear classifier is a support vector machine (SVM).

[0043] Compared with the prior art, the advantages of the present invention are as follows:

[0044] 1. This invention transforms the signal processing problem into an image classification problem. By extracting the texture features of the RD image and combining them with a classification algorithm, transient interference can be classified more comprehensively and accurately.

[0045] 2. This invention extracts the texture features of an image and transforms them into a training set. Compared with the traditional method of directly processing the original signal, this invention requires less computational and storage resources and has a smaller size and lower overhead.

[0046] 3. After extracting image texture features, this invention can quickly classify the objects to be classified using a trained classifier, greatly improving the classification speed. This advantage is particularly important for skywave radar systems with high real-time requirements.

[0047] 4. This invention, by introducing Histogram of Oriented Gradients (HOG) and Gray-Level Co-occurrence Matrix (GLCM) features, integrates local gradient and texture information of the image, enabling a more comprehensive and accurate description of transient interference characteristics. It achieves high accuracy in interference classification.

[0048] 5. This invention is not only applicable to the transient interference classification problem of skywave radar, but also to image classification problems in other fields, and has strong versatility and scalability. Attached Figure Description

[0049] Figure 1 This is a general flowchart of an embodiment of the present invention;

[0050] Figure 2 This is the judgment algorithm for transient interference detection in this embodiment of the invention;

[0051] Figure 3 a is the RD diagram of the measured data after unprocessing in the embodiments of the present invention;

[0052] Figure 3 b is the transient interference RD diagram of a single-pulse linear frequency modulation signal converted from unprocessed measured data in an embodiment of the present invention.

[0053] Figure 3 c is the transient interference RD diagram of the multi-periodic linear frequency modulated signal converted from the unprocessed measured data in the embodiments of the present invention;

[0054] Figure 4 a is the interference-free RD image obtained after image processing in an embodiment of the present invention;

[0055] Figure 4 b is the transient interference RD diagram of the single-pulse linear frequency modulated signal obtained after image processing in an embodiment of the present invention;

[0056] Figure 4 c is the transient interference RD diagram of the multi-periodic linear frequency modulated signal obtained after image processing in an embodiment of the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and examples.

[0058] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0059] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0060] Example 1:

[0061] The specific process of a skywave radar transient interference classification method based on RD map texture features is as follows:

[0062] Step 1: Perform relevant image processing on the RD image obtained by radar. The processing steps include strong interference removal, grayscale enhancement, and size transformation. For strong interference removal, first determine a threshold based on the overall intensity of the acquired RD image. Rows and columns exceeding the threshold value in the matrix are directly deleted. The threshold is calculated as shown in the following formula, where the value of m needs to be selected based on the overall image situation:

[0063] threshold = median(pic) + m

[0064] After using histogram equalization to enhance grayscale, the number of images to be divided is flexibly selected based on the size of the obtained image, ensuring that the images are the same size after division. Repeating the above process multiple times can obtain an image training set for classifier training.

[0065] Based on the different types of interference received, the image training set can be classified into three categories: no interference, transient interference from single-pulse linear frequency modulation (LFM) signals, and transient interference from multi-periodic LFM signals. These are reflected in the images as follows: the texture particles of the background noise in the image without interference are very small; when there is LFM interference, the background will have parallel horizontal or diagonal stripes; and under periodic LFM interference, the background will appear as striped blocks.

[0066] Step 2: Extract two types of texture features from the obtained training set: Histogram of Oriented Gradient (HOG) features and Gray Level Co-currency Matrix (GLCM) features. As can be seen from the image training set obtained in Step 1, the three images corresponding to the three types of interference conditions contain rich texture features, and there are certain differences between them. Therefore, we can choose texture features to distinguish the types of RD images.

[0067] Texture Feature 1: HOG (Heading Oriented Gradient) constructs the feature vector used for recognition by calculating the gradient information of partial regions of an image and statistically analyzing the histogram of gradient information. Its operating principle is to divide an image into several cells, calculate the gradient corresponding to each pixel in each cell, divide the image into different intervals, and construct an oriented gradient histogram. The specific gradient calculation is shown in the following formula.

[0068] G x (x,y)=I(x+1,y)-I(x-1,y)

[0069] G y (x,y)=I(x,y+1)-I(x,y-1)

[0070]

[0071]

[0072] Texture Feature 2: GLCM, a statistical texture analysis method, uses the spatial distribution of gray levels to represent texture: the gray level distribution of coarse textures changes more slowly with distance than that of fine textures. GLCM can reflect comprehensive information about image gray levels regarding direction, adjacent spacing, and the magnitude of change. It is the basis for analyzing local patterns and their arrangement rules in images. Elements in GLCM are the joint distribution of gray levels of two pixels with a certain spatial relationship. Assuming the spatial relationship of a pair of pixels is P(i, k|d, θ), this represents the frequency of gray level j starting from i, given a distance d and a direction θ. The distance d can take positive integer values, and the angle θ can take values ​​of [0°, 45°, 90°, 135°].

[0073] The GLCM generation process includes the following three steps:

[0074] (1) Determine the maximum and minimum gray levels L in the image. max L min Then the range of values ​​for any pair of points (i, j) in the image is (L... min L min ) and (L max L max )between;

[0075] (2) Based on the specific distance d and angle θ requirements, count the number of occurrences of each point pair (i, j) in the entire image and arrange them into a matrix;

[0076] (3) Normalize the number of occurrences of (i,j) to obtain the probability P(i,j), and form the GLCM matrix.

[0077] Generally, due to the large dimension of GLCM, it is not suitable to use it directly as a feature for classification. Therefore, this study constructs four statistical measures as features based on it: contrast, correlation, energy, and homogeneity. This method selects four angles and five distances, calculates the four statistical measures under each combination of angle and distance, and then takes the mean and standard deviation of each combination to finally form the feature vector for classification.

[0078] Step 3: Based on the image training set obtained in S1 and the feature extraction methods in S2 (HOG and GLCM), create a feature set that can be used as input. Combine this with the selected classifier and train it to obtain a usable classifier. In this method, the classifier chosen is a Support Vector Machine (SVM). SVM is a few-shot learning method with a solid theoretical foundation, and the algorithm is simple. Its computational complexity mainly depends on the number of support vectors, and it has a certain degree of robustness. When using SVM to solve multi-class classification problems, the multi-class problem can be divided into multiple binary problems. That is, a binary SVM is designed between each pair of classes. This method has a total of 3 classes, so the number of classifiers is 3.

[0079] Step 4: Based on the trained classifier obtained in S3, for the RD image to be judged, first follow the image processing steps in S1, then judge it according to the designed transient interference multi-classifier fusion judgment algorithm to obtain the final classification result of the RD image. The specific process of the transient interference multi-classifier fusion judgment algorithm is as follows: for each RD image to be classified, the N sub-images are divided and judged by the trained classifier obtained in S3. The N results are analyzed as follows: if all N results are interference-free, the RD image to be classified is classified as interference-free; if other results appear, the ratio of the other two cases is calculated, and the result with the largest ratio is taken as the classification result of the whole image, and this result is output.

[0080] Example 2: This Example 2 applies to the skywave radar transient interference classification method based on RD map texture features in Example 1, and is implemented according to the following steps.

[0081] The RD graphs used to create the training set were all obtained by processing and transforming the actual measured data.

[0082] For the unprocessed RD map, the first step is to remove strong interference from the RD map obtained by radar, as shown in the attached figure. Figure 3 As shown in a, the unprocessed RD image has strong sea clutter in the middle. In order to prevent the sidelobe extension of strong clutter from affecting the texture features of the noise floor, it is necessary to remove strong interference. At the same time, there are also influences from meteor trails, stripe interference, etc. Figure 3 b shows the transient interference RD diagram of a single-pulse linear frequency modulated signal converted from unprocessed measured data; Figure 3c shows the transient interference RD diagram of the multi-periodic linear frequency modulated signal converted from the unprocessed measured data. Based on the data processed in Example 2, sea clutter is approximately 60 dB higher than the noise floor, so the m value for removing this strong interference is set to 60. The threshold calculation method is as follows:

[0083] threshold = median(pic) + 60;

[0084] After grayscale enhancement, the image was divided into four 100×100 sub-images. This process was repeated to create a training set of images. The training set included three categories: interference-free, transient interference from a single-pulse linear frequency modulated signal, and transient interference from a multi-periodic linear frequency modulated signal, with 200 images in each category. The training set and test set each contained 600 images. (See attached figure.) Figure 4 a is the interference-free RD image obtained after processing. Figure 4 b is the transient interference RD diagram of the single-pulse linear frequency modulated signal obtained after processing. Figure 4 c is the transient interference RD diagram of the multi-periodic linear frequency modulated signal obtained after processing.

[0085] Step 2: Feature extraction from the training set: The unit size in HOG is 32×32, the block size is 2×2, and the histogram interval is 9, corresponding to 9 gradient direction intervals, each interval being 20°; GLCM takes four directions [0°, 45°, 90°, 135°], with distances of [1, 2, 3, 4, 5]. The mean and standard deviation of contrast, correlation, energy, and homogeneity of each gray-level co-occurrence matrix are calculated respectively. The results are combined with the HOG features to form a feature vector for classification. Table 1 shows the first 20 rows of the feature vector of size 184×4.

[0086] Table 1 shows a partial display of feature vectors.

[0087]

[0088]

[0089] Step 3: Based on the training set obtained in S1 and the feature extraction methods in S2 (HOG and GLCM), and combined with the selected classifier SVM, train a usable classifier. Design a binary SVM between every two classes, for a total of 3 classifiers. The performance of the classifier can be tested using the test set, and the resulting confusion matrix is ​​shown in Table 2.

[0090] Table 2 Confusion Matrix for SVM Classification

[0091]

[0092] It can be seen that the overall accuracy of the classifier is 88.17%, with a recognition rate of 95.5% for interference-free cases and 99% for periodic linear frequency modulation interference, proving the effectiveness and feasibility of this method in interference classification.

[0093] Step 4: Based on the trained classifier obtained in S3, the RD image to be detected is first processed according to the image processing steps in S1, and then judged according to the transient interference multi-classifier fusion judgment algorithm to obtain the final classification result of the RD image. Table 3 shows the judgment algorithm process.

[0094] Table 3. Algorithm Flow

[0095]

[0096] The methods described above according to the invention can be implemented in hardware, firmware, or as software or computer code that can be stored in a recording medium (such as CD-ROM, RAM, floppy disk, hard disk, or magneto-optical disk), or as computer code originally stored on a remote recording medium or a non-transitory machine-readable medium and subsequently stored on a local recording medium, downloaded via a network. Thus, the methods described herein can be stored as software processing on a recording medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware (such as an ASIC or FPGA). It is understood that the computer, processor, microprocessor controller, or programmable hardware includes storage components (e.g., RAM, ROM, flash memory, etc.) capable of storing or receiving software or computer code. When said software or computer code is accessed and executed by the computer, processor, or hardware, the skywave radar transient interference classification method based on RD map texture features described herein is implemented. Furthermore, when a general-purpose computer accesses the code used to implement the processing shown herein, the execution of the code transforms the general-purpose computer into a dedicated computer for performing the processing shown herein.

[0097] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the implementation methods of the present invention, and should be understood that the scope of protection of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of the present invention.

Claims

1. A method for classifying transient interference in skywave radar based on RD graph texture features, characterized in that, Includes the following steps: S1: Perform relevant image processing on the RD map obtained by radar, including strong interference removal, grayscale enhancement and size transformation. Design and create an image training set for training a linear classifier. The image training set includes three categories: no interference, transient interference of single-pulse linear frequency modulation signal, and transient interference of multi-periodic linear frequency modulation signal. S2: Extract two types of texture features from the obtained image training set, namely Histogram of Oriented Gradients (HOG) features and Gray-Level Co-occurrence Matrix (GLCM) features. Step S2 is as follows: S21: Extract HOG features from the image training set obtained in S1 to form feature vector A; S22: Extract GLCM features from the image training set obtained after step S1 to form feature vector B; S23: The extracted features are formed by fusing feature vector A and feature vector B. S3: Training using a linear classifier; S4: Design a transient interference multi-classifier fusion judgment algorithm to classify the input files; Step S4 is as follows: Based on the trained classifier obtained in S3, the RD image to be detected that needs to be judged is first processed according to the steps in S1, and then judged according to the designed transient interference detection judgment algorithm to obtain the final classification result of the RD image; the linear classifier is a support vector machine (SVM). The specific process of the transient interference multi-classifier fusion judgment algorithm is as follows: For each RD image to be classified, the N sub-images are divided and judged by the trained classifier obtained in S3. The N results are then analyzed. The process is as follows: If all N results are interference-free, the RD image to be classified is classified as interference-free; if other results appear, the ratio of the other two cases is calculated, and the result with the largest ratio is taken as the classification result of the whole image, and this result is output.

2. The skywave radar transient interference classification method according to claim 1, characterized in that: Step S1 specifically involves: S11: Strong Interference Removal: First, a threshold is set based on the intensity of the entire RD image that has been acquired. Rows and columns in the matrix that exceed the threshold value are directly deleted. The threshold is calculated as follows: ， In the formula, the value of m needs to be selected based on the overall image situation, pic represents the input image, median represents taking the median value of the image intensity, and threshold represents the calculated threshold. S12: Gray-scale enhancement: Histogram equalization is used to enhance the contrast of the image obtained after step S11. S13: Size transformation: Based on the image size obtained in S12, flexibly select the number of images to be divided, ensuring that the image sizes are the same after division.

3. The skywave radar transient interference classification method according to claim 1, characterized in that: In S2, the Histogram of Oriented Gradients (HOG) divides an image into several units, calculates the gradient corresponding to each pixel in a unit, and constructs an HOG histogram. The specific value of the HOG histogram is determined by both the magnitude and direction of the gradient. The histograms obtained from each unit are combined. In a larger block region, the histograms of the contained units are concatenated to perform contrast normalization on the gradient intensity. The specific gradient calculation is shown in the following formula. ， In the formula, and : These represent the gradients in the horizontal and vertical directions, respectively. It is the difference between pixel (x+1, y) and pixel (x-1, y), while It is the difference between pixel (x, y + 1) and pixel (x, y - 1); : Represents the gradient magnitude at pixel (x, y); : Indicates the gradient direction at pixel (x, y).

4. The skywave radar transient interference classification method according to claim 1, characterized in that: The generation process of GLCM in S2 is as follows: 1) Determine the maximum and minimum gray levels in the image. , Then any pair of points in the image The range of values ​​is within and between; 2) Based on the specific distance d and angle The requirement is to count the number of each type of point pair in the entire image. The frequency of occurrence is arranged into a matrix; 3) To The probability is obtained by normalizing the number of occurrences. This forms a GLCM matrix; 4) Based on the obtained GLCM matrix, four statistical measures were constructed as features: contrast, correlation, energy, and homogeneity. The mean and standard deviation of the corresponding statistical measures were calculated as features.

5. The skywave radar transient interference classification method according to claim 1, characterized in that: Step S3 specifically involves: By combining the image training set, HOG, GLCM, and a linear classifier, the linear classifier is trained to obtain the trained classifier.

Images