Ship target detection method based on distance compression domain and electronic equipment

Abstract

The invention discloses a ship target detection method based on a distance compressed domain and electronic equipment, and relates to the technical field of target detection. The ship target detection method based on the distance compression domain comprises the following steps: acquiring an SAR echo signal, and performing distance compression on the SAR echo signal to obtain a distance compression domain image; performing feature enhancement on the distance compressed domain image, and performing image segmentation on the distance compressed domain image based on a feature enhancement result to remove land interference; and performing time-frequency analysis on the distance compressed domain image after the land interference is removed by using wavelet transform, and detecting a ship target based on a time-frequency analysis result. Therefore, by combining the multi-scale characteristic of wavelet transform and the target feature extraction advantage of the distance compressed domain, the ship target can be quickly and effectively detected under the condition that the distance migration correction is not completely carried out.

Description

Technical Field

[0001] This invention relates to the field of target detection technology, and in particular to a ship target detection method and electronic equipment based on distance compression domain. Background Technology

[0002] Synthetic Aperture Radar (SAR), a high-resolution remote sensing imaging technology, is widely used in maritime surveillance, environmental monitoring, and geological exploration. With the continuous advancement of radar technology, SAR plays an increasingly important role in ship target detection, especially in the surveillance of large sea areas. Ship targets typically appear as relatively small point targets in SAR images, and these targets are often interfered with by complex sea surface echoes, background noise, and weather conditions. Furthermore, with the development of remote sensing technology, SAR systems have higher spatial resolution, which means that spaceborne SAR will face greater challenges in terms of real-time performance and effectiveness in ship target detection. Summary of the Invention

[0003] The purpose of this invention is to propose a ship target detection method and electronic device based on distance compression domain, so as to achieve rapid and effective detection of ship targets.

[0004] In a first aspect, embodiments of the present invention propose a ship target detection method based on range compression domain. The method includes: acquiring SAR echo signals and performing range compression on the SAR echo signals to obtain a range compression domain image; performing feature enhancement on the range compression domain image and performing image segmentation on the range compression domain image based on the feature enhancement results to remove land interference; performing time-frequency analysis on the range compression domain image after removing land interference using wavelet transform, and detecting ship targets based on the time-frequency analysis results.

[0005] In some embodiments, the distance compressed domain image is represented by the following formula: ; in, This represents the distance compression domain image. Represents a complex constant. Represents the distance compression response function. Indicates distance in advance time. Indicates direction in slow time. Indicates the instantaneous distance of the ship target. This represents the speed of light and is related to the target's backscattering coefficient. Indicates the direction to the envelope. Indicates the time of beam center deviation. Represents the imaginary number symbol, Indicates the radar center frequency.

[0006] In some embodiments, the feature enhancement of the distance compression domain image includes: performing a dilation operation on the distance compression domain image; and performing edge detection on the dilated distance compression domain image using a preset first-order differential operator.

[0007] In some embodiments, the preset first-order differential operator is one of the Prewitt operator, the Sobel operator, and the Roberts operator.

[0008] In some embodiments, the expansion formula is: ; Where A represents the edge detection result, and B represents the structuring element. This means that B moves within plane A with its center point as the point of motion. Let A represent the set of all moving points that make A and B intersect. This indicates that the intersection of B and A after the reflection is translated by z is not empty.

[0009] In some embodiments, the image segmentation of the distance compressed domain image based on the feature enhancement result includes: reading the distance compressed domain image along the distance direction row by row after edge detection; dividing the read distance direction data into multiple sub-segments using the gradient prior knowledge obtained from edge detection for each row of distance direction data read; sequentially reading the multiple sub-segments, and determining that the sub-segment is a land signal when the length of the sub-segment is greater than a preset threshold and the data is a first value for each sub-segment read; after the row-by-row reading is completed, obtaining a land mask based on the sub-segments corresponding to all land signals, and using the land mask to perform image segmentation on the distance compressed domain image.

[0010] In some embodiments, Haar wavelet transform is used to perform time-frequency analysis on the range-compressed domain image after land interference removal.

[0011] In some embodiments, detecting ship targets based on time-frequency analysis results includes: identifying a high-frequency region in the time-frequency analysis results; when a target signal is detected in the high-frequency region based on the amplitude, determining the ship target based on the target signal; wherein the data length between the two endpoints of the target signal is greater than the synthetic aperture length.

[0012] In some embodiments, determining the ship target based on the target signal includes: when there are multiple target signals and they are adjacent, merging the multiple target signals; and determining the ship target based on the merging result.

[0013] In a second aspect, embodiments of the present invention provide an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein when the computer program is executed by the processor, it implements the method described in the first aspect embodiment.

[0014] The ship target detection method and electronic equipment based on range compression domain of this invention first acquires SAR echo signals and performs range compression on the SAR echo signals to obtain a range compression domain image; then, feature enhancement is performed on the range compression domain image, and image segmentation is performed based on the feature enhancement results to remove land interference; subsequently, wavelet transform is used to perform time-frequency analysis on the land interference-free range compression domain image, and ship targets are detected based on the time-frequency analysis results. Therefore, by combining the multi-scale characteristics of wavelet transform with the target feature extraction advantages of range compression domain, ship targets can be detected quickly and effectively without complete range migration correction. Attached Figure Description

[0015] Figure 1 This is a schematic diagram of an example SAR detection scenario according to the present invention; Figure 2 This is a flowchart of a target detection process in related technologies; Figure 3 This is a flowchart of a ship target detection method based on distance compression domain according to an embodiment of the present invention; Figure 4 This is a schematic diagram of an example of the SAR echo signal and range compressed domain image of the present invention; Figure 5 This is a schematic diagram of an example LFM (Linear Frequency Modulated) signal according to the present invention; Figure 6 This is a comparative schematic diagram of edge detection in an example of the present invention; Figure 7 This is a flowchart illustrating land segmentation as an example of the present invention; Figure 8 This is an example of a land mask and the effect of removing land in this invention; Figure 9 This is a schematic diagram of the simulated target detection result of an example of the present invention; Figure 10 This is a schematic diagram of the actual target detection result of an example of the present invention; Figure 11 This is a flowchart of the RD (range-Doppler) algorithm in related technologies; Figure 12 This is a schematic diagram illustrating sinc interpolation and feature enhancement time consumption as an example of the present invention. Detailed Implementation

[0016] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0017] Traditional SAR target detection workflows mainly include data acquisition, SAR imaging, and target detection. SAR imaging images the scattering characteristics of a region, which is generally characterized by complex terrain. Objects at different locations within the region exhibit varying scattering characteristics. Ultimately, the SAR receiver receives the superposition of backscattered signals from all objects within the detection area. Target detection methods include Constant False Alarm Rate (CFAR) detection methods and deep learning-based target detection methods.

[0018] For SAR imaging, Figure 1 A SAR detection scenario is shown. See also Figure 1 For a red dot target, the SAR beam illuminates it directly, starting from point A, reaching its closest point at point P, and just moving away from the red dot target at point B. The radar platform receives the echo signal from the red dot target from point A to point B. During operation, the distance between the radar platform and the target changes, resulting in the range pulse compression peak value not occurring at the same time for the same target at different radar time points. Therefore, without range migration correction, the range-compressed signal is an arc, and further azimuth compression will inevitably have a significant impact on image quality. Traditional target detection methods based on the imaging domain unavoidably require range migration correction, which is a major factor affecting real-time detection performance.

[0019] The CFAR detection method is widely used in one-dimensional radar signal detection. It primarily determines the detection threshold based on the statistical characteristics of reference cell clutter and a set false alarm rate. The detection process is as follows: Figure 2 As shown. In one-dimensional radar echo signal detection, when the clutter is uniform, the detection threshold determined by this method is only related to the false alarm rate and the number of clutter reference cells, and is independent of the clutter power level. Therefore, it is called constant false alarm rate detection.

[0020] Currently, researchers have proposed many efficient CFAR detection algorithms for various clutter environments, which can be broadly divided into two categories: one is the mean-based CFAR (CA-CFAR) algorithm, which assumes that the background clutter is uniformly distributed; the other is the ordered statistical CFAR (OS-CFAR) algorithm, which is designed to handle situations with multiple targets in the neighborhood. As the cornerstone of SAR target detection, CFAR detection methods still have certain shortcomings in their detection logic, and these methods often only exhibit optimal performance in specific environments. Most improved detection methods select a particular detection method based on certain criteria, which undoubtedly increases the computational load of the algorithm and lacks a method that truly integrates the advantages of each method.

[0021] Deep learning-based object detection methods are mainly divided into anchor-based methods and anchor-free methods. Anchor-based methods include two-stage object detection algorithms based on candidate regions and one-stage object detection algorithms based on regression.

[0022] Two-stage object detection algorithms based on candidate regions first extract candidate regions from the image, and then generate predicted bounding boxes for the target from these regions. Two-stage detection algorithms generally have high detection accuracy but are slow. One-stage object detection algorithms based on regression do not need to generate candidate regions; they directly predict the target's class probability and location information. Compared to two-stage algorithms, the detection speed is significantly improved. Anchor-bound object detection algorithms become complex due to the excessive number of anchor boxes generated, and the large number of hyperparameters also affects detector performance. Anchor-bound object detection algorithms, on the other hand, greatly reduce the number of hyperparameters by determining keypoints instead of anchor boxes.

[0023] Because remote sensing images often have complex backgrounds and ship targets with large scale variations, deep learning-based target detection methods generally need to use methods such as frequency domain enhancement, feature pyramid network structures, and attention mechanisms to enhance target features in order to improve the detection accuracy of multi-scale ship targets. This results in high computational complexity, making it difficult to meet the requirements of real-time detection.

[0024] To address the shortcomings and deficiencies of the aforementioned technologies, this invention proposes a ship target detection method and electronic device based on range compression domain. The aim is to rapidly and effectively detect ship targets without fully performing range migration correction by combining the multi-scale characteristics of wavelet transform with the advantages of target feature extraction in range compression domain. Specifically, ship target detection is first performed within the range compression domain to extract the target's feature information. Then, imaging processing is applied only to small regions containing the target, thereby significantly improving the real-time performance of the detection process while maintaining detection accuracy.

[0025] The ship target detection method and electronic equipment based on the distance compression domain according to embodiments of the present invention are described below with reference to the accompanying drawings.

[0026] Figure 3 This is a flowchart of a ship target detection method based on distance compression domain according to an embodiment of the present invention.

[0027] like Figure 3 As shown, the ship target detection method based on range compression domain includes: S11: Acquire SAR echo signals and perform range compression on the SAR echo signals to obtain a range-compressed domain image.

[0028] In some embodiments of the present invention, the distance compressed domain image is represented by the following equation (1): ; in, Represents the distance compression domain image. This represents a complex constant (related to the target backscattering coefficient). Represents the distance compression response function. Indicates distance in advance time. Indicates direction in slow time. Indicates the instantaneous distance of the ship target. This represents the speed of light and is related to the target's backscattering coefficient. Indicates the direction to the envelope. Indicates the time of beam center deviation. Represents the imaginary number symbol, Indicates the radar center frequency.

[0029] Specifically, from the range perspective, a ship target in a SAR image can generally be modeled as a point target. For a point target, the raw echo signal obtained by SAR (i.e., the SAR echo signal) can be modeled as follows: ; in, This represents the original echo signal. Indicates the distance to the envelope. =B⁄T, where B represents the LFM rate, B represents the signal bandwidth, and T represents the time width.

[0030] Since the echo signal received by radar in reality is not just the scattering of a single point target, it can be ignored. Thus, we can obtain the following formula (3): ; in, This indicates the original echo signal after ignoring A0. The approximate value obtained by Taylor expansion of the expression for slow time is: ; in, This indicates the shortest reference distance for the radar to reach the target. Indicates the radar's flight speed. Indicates slow time. Substituting the original echo signal after ignoring A0, we get: ; in, The formulas for the echoes expanded in the range and azimuth directions of SAR are given. , where represents the azimuth frequency modulation, and c represents the speed of light. It can be seen that there are two linear frequency modulation signals in the azimuth and range directions respectively. Regarding equation (5)... After the signal is matched and filtered, we can obtain the above equation (1).

[0031] As can be seen from equation (1), after distance compression, from distance to fast time... The signal shows Characteristics. From azimuth to slower time. The signal exhibits LFM characteristics (ignoring distance migration).

[0032] In one example, the SAR echo signal is as follows: Figure 4 As shown in (a), the distance compression domain image is as follows Figure 4 As shown in (b).

[0033] S12, feature enhancement is performed on the range compression domain image, and image segmentation is performed on the range compression domain image based on the feature enhancement results to remove land interference.

[0034] In some embodiments of the present invention, feature enhancement of the distance compression domain image includes: performing dilation operation on the distance compression domain image; and performing edge detection on the dilated distance compression domain image using a preset first-order differential operator.

[0035] For example, the first-order differential operator is preset to be one of the prewitt operator, the Sobel operator, or the Roberts operator.

[0036] In some examples, the expansion formula is: ; Where A represents the edge detection result, and B represents the structuring element. This means that B moves within plane A with its center point as the point of motion. Let A represent the set of all moving points that make A and B intersect. This indicates that the intersection of B and A after the reflection is translated by z is not empty.

[0037] Specifically, after transforming the SAR echo signal to the range compression domain, the energy in the range direction is concentrated, while the energy in the azimuth direction remains dispersed. Ideally, there would only be sea surface echoes and ship echoes. However, in reality, ships may approach land. Since the backscattering characteristics of land echoes and ship echoes are higher than those of sea surface echoes, signal feature enhancement is necessary. From the azimuth perspective, since no range migration correction is performed, ship echoes can be considered a type of LFM signal. Land echoes, ship echoes, and sea echoes can be considered as a superposition of many types of LFM signals. The LFM signal is defined as follows: ; in, Indicates LFM signal, Indicates the distance to the envelope. Indicates the radar center frequency. =B⁄T, representing LFM rate. B represents fast time (distance-to-time), B represents signal bandwidth, and T represents time width.

[0038] Figure 5 The real part waveforms of an ideal LFM signal and LFM-like signals such as land echo, ship echo, and ocean echo are shown.

[0039] Based on the backscattering characteristics of SAR echo signals, solids often exhibit greater backscattering than liquids. This manifests as brighter areas in images and larger amplitude regions in the signal. Therefore, by pre-setting first-order differential operators, such as the Prewitt, Sobel, and Roberts operators, relatively bright areas in the image can be detected from the image perspective for feature enhancement. Taking the Sobel operator as an example, the Sobel operator is defined as follows: ; ; By convolving these two kernels with the image, we can obtain the gradient values ​​Gx and Gy in the horizontal and vertical directions, respectively, representing the rate of change of pixel brightness in these directions. By calculating the magnitude and direction of the gradients, the Sobel operator can identify regions of dramatic brightness changes in the image, thus locating the boundary between ocean and land. Directly performing edge detection on the image can result in small holes appearing in areas that should be identified as connected land, leading to poor edge detection performance. Figure 6As shown in (a). Therefore, this invention first employs morphological erosion to dilate the distance-compressed domain image, filling small holes or cracks while preserving important structural information, thus connecting objects. Then, the Sobel operator is used on the morphologically eroded image to detect gradient information at the land-sea boundary, as shown in (a). Figure 6 As shown in (b) (the lines inside the dashed box are a partial magnification of the black box). The expansion operation formula is the above formula (7), that is, taking the center point of B as the moving point, making B move in the plane of A, and the operation result is the set of all such moving points that make A and B intersect.

[0040] In some embodiments of the present invention, image segmentation of the distance compressed domain image based on feature enhancement results includes: reading the distance compressed domain image after edge detection along the distance direction row by row; for each row of distance data read, dividing the read distance data into sub-segments using the gradient prior knowledge obtained from edge detection to obtain multiple sub-segments; reading multiple sub-segments sequentially, and for each read sub-segment, determining that the sub-segment is a land signal when the length of the sub-segment is greater than a preset threshold (e.g., 50) and the data is a first value (e.g., 1); after the row-by-row reading is completed, obtaining a land mask based on the sub-segments corresponding to all land signals, and using the land mask to perform image segmentation of the distance compressed domain image.

[0041] Each sub-segment contains either all 0s or all 1s.

[0042] Specifically, after the aforementioned feature enhancement operations, interference from ocean signals can be effectively separated. Since the backscattering characteristics of ship targets and land are both higher than those of ocean, further land removal is needed to locate ship targets. Along the range direction, ship targets occupy only a few to a dozen pixels, while land typically occupies hundreds or even thousands of pixels. Based on this, uninteresting land space can be filtered out. The filtering algorithm is as follows: 1) Read the binarized SAR range compression domain image along the range direction row by row after the dilation operation; 2) Using the gradient prior knowledge obtained from the Sobel operator detection results after dilation, the data is divided into several sub-segments; 3) Set a threshold based on the LFM characteristics of the ship target; here, an empirical threshold of 50 is selected. 4) When the length of the current segment is greater than the threshold and the data is 1, this segment can be considered as a land echo; 5) The result after removing the land is used as a mask to block the original SAR echo range compression domain data, thus obtaining the result after removing the land.

[0043] Specifically, such as Figure 7As shown, the input binary distance compression data (i.e., the result of the dilation operation) is used to determine if the data has been completely read. If the data has been completely read, a landmask is obtained, and the land removal effect can be obtained based on the landmask. If the data has not been completely read, a line of distance data is read, and it is divided into segments based on gradient prior knowledge. One segment is read, and it is determined whether the segment has been completely read. If the segment has been completely read, another line of distance data is read and processing continues. If the segment has not been completely read, it is determined whether the segment data is 1. If it is 1, it is further determined whether the segment data length is greater than a threshold (i.e., the preset threshold mentioned above). If it is greater than the threshold, the segment is determined to be a land signal, and another segment is read and the determination continues. If it is not greater than the threshold, the segment is determined to be a suspected ship signal, and another segment is read and the determination continues. If the segment data is not 1, the segment is determined to be a marine signal, and another segment is read and the determination continues.

[0044] Figure 8 The land mask is shown (see Figure 8 (a) and the effect of removing land from the range-compressed domain image using a land mask (see [reference]). Figure 8 (b)

[0045] S13 utilizes wavelet transform to perform time-frequency analysis on the range compressed domain image after removing land interference, and detects ship targets based on the time-frequency analysis results.

[0046] For example, Haar wavelet transform is used to perform time-frequency analysis on the range-compressed domain image after land interference removal.

[0047] In some embodiments, detecting ship targets based on time-frequency analysis results includes: identifying a high-frequency region in the time-frequency analysis results; when a target signal (i.e., the LFM signal corresponding to the ship target) is detected in the high-frequency region based on the amplitude, determining the ship target based on the target signal; wherein the data length between the two endpoints of the target signal is greater than the synthetic aperture length.

[0048] For example, high-frequency regions can be identified in time-frequency analysis results through a variable sliding window; the two endpoints can be two points with larger amplitudes in the high-frequency region.

[0049] As one implementation method, determining a ship target based on target signals includes: when there are multiple target signals that are adjacent, merging the multiple target signals; and determining the ship target based on the merging result.

[0050] Since actual ship targets are not ideal point targets, several LFM signals may be adjacent. By merging these signals, a single ship target can be considered detected. For example, the signal determined by the farthest two endpoints of adjacent target signals can be used as the merging result, and this merging result can be considered the detected ship target.

[0051] Specifically, after feature enhancement and image segmentation, SAR ship targets can be considered as searching for vertical lines against an interference-free background. Ideally, the echo of a point target in the range compression domain is at the single-pixel level from the range direction. After feature enhancement and image segmentation, Hough transform can be used to easily detect it. However, ship targets are only approximated as point targets and are not actually at the single-pixel level. Therefore, wavelet transform is chosen to detect the target from the signal.

[0052] Because energy is concentrated after range compression, time-frequency analysis of the range-compressed domain signal allows for the identification of high-energy regions in the high-frequency range to pinpoint the endpoints of the LFM signal. These endpoints help determine the signal length. When the signal length approaches the length of the synthetic aperture, it can be considered a characteristic signal of the ship target in the range-compressed domain. Simultaneously, the removal of land echo interference effectively improves detection accuracy.

[0053] Common time-frequency analysis methods for signals include Fourier transform, short-time Fourier transform, wavelet transform, and mode decomposition. This invention not only requires finding the high-frequency space but also the time of occurrence of the high-frequency space to determine the endpoints of the LFM signal in the SAR image. Therefore, Haar wavelet transform can be used to detect ship targets.

[0054] Based on the above description, the overall technical solution of this invention is as follows: For the raw echo signal obtained by SAR, the signal is first transformed to the range compression domain through range compression. In the range compression domain, due to the focusing process, the features of the ship target are enhanced. Edge detection is performed on the SAR image in the range compression domain to initially locate the coastline. Then, a filtering algorithm is used to segment the land and ocean images to prevent land echoes in high-energy areas from affecting the detection results. Finally, after removing the influence of land interference, the ship target is detected.

[0055] To demonstrate the effectiveness of this invention, simulated detection results for point targets and actual SAR imaging detection results including the coastline are presented. The actual SAR image is derived from a Sentinel-1A focused amplitude image, obtained in VV polarization and C-band. The imaging method is RD imaging, with a resolution of approximately 6 meters and a swath width of approximately 25 km. The simulated detection results are as follows: Figure 9 As shown, the actual test results are as follows: Figure 10 As shown.

[0056] As mentioned earlier, traditional imaging domain-based detection methods inevitably require range migration correction, which is a major factor affecting real-time detection performance. Common SAR imaging algorithms include the RD algorithm (range-Doppler algorithm) and the CS algorithm (frequency modulation scaling algorithm). Taking the RD algorithm as an example, its process is as follows: Figure 11 As shown.

[0057] exist Figure 11In the distance migration correction shown, common interpolation methods include the nearest neighbor approximation method and the sinc interpolation method. Firstly, since the nearest neighbor approximation method does not require multiplication and addition operations, it only requires 2 × Nr × Na data shift operations (Nr and Na are the number of range points and azimuth points, respectively). However, because sinc interpolation involves weighted calculations on neighboring points, the computational load increases significantly. sinc interpolation starts from... From The process is actually an FIR filtering process, and the P-point interpolation method requires... Multiplication operations and This is the second addition operation. The nearest neighbor approximation method is only a coarse correction because it uses the nearest neighbor... To approximate with that integer For the values ​​at a given point, the nearest neighbor approximation method only roughly corrects the curvature. The sinc interpolation method produces better images than the nearest neighbor approximation method, but the interpolation time varies depending on the number of points. The 4-point sinc interpolation method achieves better correction results than the nearest neighbor approximation method, but the correction time is longer. When more points are used, such as 8 points, the accuracy of sinc interpolation improves, but the computation time increases significantly. Pre-detection before distance migration correction can greatly reduce the magnitude of Nr and Na, thereby reducing the time loss caused by distance migration correction and improving the real-time performance of detection.

[0058] The following analysis examines the distance migration correction time for different numbers of points. The computer used was an i5-12400f CPU, an NVIDIA GeForce RTX 3060ti GPU, and 16GB of RAM. The time consumption analysis of distance migration correction and the proposed method is as follows: Figure 12 As shown.

[0059] from Figure 12 As can be seen, the time required for range migration correction increases significantly with the increase in imaging data size, while the relative time required for feature enhancement in the range compression domain is very small. Taking the true echo of Sentinel-1A data as an example, the number of data points after range compression is a complex number of 4100*7100, and the range migration correction time is 237.426511s. Detection in the range compression domain takes 6.610068s. Subsequently, imaging a small detected region of 1000*1000 points takes 9.718583s. It can be seen that detecting the target before range migration correction reduces the imaging area and significantly reduces the time required for range migration correction, thus enabling rapid target detection.

[0060] In summary, the novel SAR image ship target detection method proposed in this invention, based on the range compression domain rather than the traditional two-dimensional image, detects ship targets before range migration correction, breaking through the traditional concept that target detection must be performed in focused SAR products. This allows for the rapid elimination of the impact of broad SAR imaging on the real-time performance of ship target detection, achieving rapid detection of ship targets.

[0061] Based on the ship target detection method based on distance compression domain described in the above embodiments, this invention proposes an electronic device.

[0062] In this embodiment, the electronic device includes a memory, a processor, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the ship target detection method based on the distance compression domain described in the above embodiment.

[0063] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A ship target detection method based on distance compression domain, characterized in that, The method includes: Acquire SAR echo signals and perform range compression on the SAR echo signals to obtain a range-compressed domain image; Feature enhancement is performed on the distance compressed domain image, and image segmentation is performed on the distance compressed domain image based on the feature enhancement results to remove land interference; Wavelet transform is used to perform time-frequency analysis on the range compressed domain image after land interference removal, and ship targets are detected based on the time-frequency analysis results.

2. The method according to claim 1, characterized in that, The distance-compressed domain image is represented by the following formula: ； in, This represents the distance compression domain image. Represents a complex constant. Represents the distance compression response function. Indicates distance in advance time. Indicates direction in slow time. Indicates the instantaneous distance of the ship target. This represents the speed of light and is related to the target's backscattering coefficient. Indicates the direction to the envelope. Indicates the time of beam center deviation. Represents the imaginary number symbol, Indicates the radar center frequency.

3. The method according to claim 1, characterized in that, The feature enhancement of the distance-compressed domain image includes: Perform dilation operation on the distance compression domain image; Edge detection is performed on the distance compression domain image after dilation operation using a preset first-order differential operator.

4. The method according to claim 3, characterized in that, The preset first-order differential operator is one of the Prewitt operator, Sobel operator, or Roberts operator.

5. The method according to claim 3, characterized in that, The expansion formula is: ； Where A represents the edge detection result, and B represents the structuring element. This means that B moves within plane A with its center point as the point of motion. Let A represent the set of all moving points that make A and B intersect. This indicates that the intersection of B and A after the reflection is translated by z is not empty.

6. The method according to claim 3, characterized in that, The image segmentation of the distance-compressed domain image based on the feature enhancement result includes: Read the distance-compressed domain image after edge detection row by row along the distance direction; For each line of distance data read, the gradient prior knowledge obtained from edge detection is used to divide the read distance data into sub-segments, resulting in multiple sub-segments; The plurality of sub-segments are read sequentially. For each sub-segment read, if the length of the sub-segment is greater than a preset threshold and the data is a first value, the sub-segment is determined to be a land signal. After reading line by line, a land mask is obtained based on the sub-segments corresponding to all land signals, and the land mask is used to perform image segmentation on the distance compressed domain image.

7. The method according to claim 1, characterized in that, We use Haar wavelet transform to perform time-frequency analysis on the range-compressed domain image after removing land interference.

8. The method according to any one of claims 1-7, characterized in that, The detection of ship targets based on time-frequency analysis results includes: Identify the high-frequency region in the time-frequency analysis results; When a target signal is detected in the high-frequency region based on the amplitude, the ship target is determined based on the target signal. Wherein, the data length between the two endpoints of the target signal is greater than the synthetic aperture length.

9. The method according to claim 8, characterized in that, The step of determining the ship target based on the target signal includes: When there are multiple target signals that are adjacent to each other, the multiple target signals are merged. The ship targets are determined based on the merge results.

10. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory, which, when executed by the processor, implements the method as described in any one of claims 1-9.

Images