Marine ship target imaging detection method, electronic equipment and storage medium

By screening and compressing the distance and orientation of the ocean SAR images, and preliminary screening and detection combined with neural network models, the problem of large amount of calculation of ocean SAR image imaging processing is solved, and efficient ship target detection is achieved.

CN118604823BActive Publication Date: 2024-11-29BEIHANG UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410328312.9
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-21
Publication Date
2024-11-29
Estimated Expiration
2044-03-21

AI Technical Summary

Technical Problem

The calculation amount of marine SAR image imaging processing is huge, resulting in low detection efficiency, especially in imaging detection without target areas, and resource waste is severe.

Method used

By performing matching filtering and one-dimensional compression in the distance direction and orientation direction, the trained neural network model is used for preliminary screening and object detection, reducing imaging calculations for targetless areas and improving detection efficiency.

Benefits of technology

It significantly improves the efficiency of marine ship target imaging detection, saves computing resources, and achieves efficient ship target detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118604823B_ABST
    Figure CN118604823B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of target detection technology, and in particular to a method, electronic device and storage medium for detecting marine ship target imaging, wherein the method comprises: obtaining demodulated satellite SAR echo data; performing matched filtering in the distance direction to obtain a corresponding preliminary imaging in the range compression domain; performing one-dimensional compression in the azimuth direction to obtain a corresponding one-dimensional range signal; using a trained one-dimensional range signal screening model to perform preliminary screening on the obtained one-dimensional range signal, and selecting a one-dimensional range signal corresponding to a target area; obtaining a preliminary image of a corresponding target response area through indexing and imaging; performing block screening through regional dynamic quantization to obtain a target area image; and using a trained target detection model to detect each obtained target area image to obtain a corresponding detection result. The present invention can improve the efficiency of imaging detection for marine SAR images.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of target detection, and in particular to a method for detecting marine ship target imaging, electronic equipment and storage medium. Background Art

[0002] Marine ship target imaging detection has always been an important development direction of remote sensing marine observation technology. Through the all-day and all-weather operation of remote sensing SAR (Synthetic Aperture Radar) satellites, the imaging resolution is improved and the imaging width is increased, accumulating a large amount of effective data of high-resolution and wide-width marine observation. The acquired data is imaged and processed to obtain the corresponding marine observation image, and then the intelligent target detection model is used to detect the distant sea ship targets and obtain the ship target information, which is conducive to the realization of marine area management and security.

[0003] However, for spaceborne SAR data covering thousands or even tens of thousands of square kilometers, the amount of computation required for imaging processing is huge. Under the premise of ensuring a certain imaging resolution, it is difficult to achieve fast and efficient imaging, and then perform target detection operations on the acquired images. At the same time, since ocean SAR images have the characteristics of large scenes and sparse distribution of ship targets, there is a large proportion of target-free areas. Imaging and detection of target-free areas will consume a lot of computing resources, resulting in waste.

[0004] Therefore, according to the characteristics of ocean SAR images, it is necessary to propose an imaging detection method with high detection efficiency. Summary of the invention

[0005] Based on the problem that the existing technology for imaging detection of ocean SAR images is low in efficiency and easily causes waste of resources, the present invention provides a marine ship target imaging detection method, electronic equipment and storage medium, which can effectively improve the efficiency of imaging detection of ocean SAR images.

[0006] In a first aspect, the present invention provides a method for detecting marine ship target imaging, comprising:

[0007] Obtain demodulated and processed spaceborne SAR echo data;

[0008] Based on the spaceborne SAR echo data, a corresponding preliminary imaging in the range compression domain is obtained by performing matched filtering upward in the range;

[0009] The obtained range compression domain is initially imaged, and the corresponding one-dimensional range signal is obtained by performing one-dimensional compression in the azimuth direction;

[0010] Using the trained one-dimensional range signal screening model, the obtained one-dimensional range signals are preliminarily screened, and the one-dimensional range signals corresponding to the target areas are screened in batches according to a fixed sliding window length, and the one-dimensional range signal screening model is a neural network model trained with the sample one-dimensional range signal and the marked corresponding range target area as input;

[0011] According to the range of the selected one-dimensional range signal, a preliminary image of the corresponding target response area is obtained through indexing and imaging;

[0012] Based on the obtained preliminary image of the target response area, performing block screening through regional dynamic quantization to obtain a target area image;

[0013] The obtained target area images are detected using a trained target detection model to obtain corresponding detection results; wherein the target detection model is trained with sample target area images as input and corresponding detection results as output, and the detection results include target positions.

[0014] Optionally, the step of performing matched filtering in the range upward direction to obtain a corresponding preliminary imaging in the range compression domain includes:

[0015] Performing a range-direction Fourier transform on the satellite-borne SAR echo data to obtain a Fourier transform result;

[0016] According to the obtained Fourier transform result, matched filtering is performed in the frequency domain;

[0017] Based on the matched filtering results, preliminary imaging in the range compression domain is obtained through inverse Fourier transform.

[0018] Optionally, the obtaining a corresponding one-dimensional range signal by performing one-dimensional compression in the azimuth direction includes:

[0019] Based on the obtained preliminary imaging in the range compression domain, the corresponding one-dimensional range signal is obtained by integrating and superimposing the fast time along the azimuth direction.

[0020] Optionally, the step of obtaining a corresponding preliminary image of the target response area by indexing and imaging according to the range of the selected one-dimensional range signal includes:

[0021] According to the range range of the selected one-dimensional range signal, corresponding range compression domain data is extracted from the range compression domain preliminary imaging through indexing;

[0022] Based on the extracted distance compression domain data, a preliminary image of the corresponding target response area is obtained by imaging.

[0023] Optionally, the performing block screening by regional dynamic quantization to obtain a target region image includes:

[0024] The obtained preliminary image of the target response area is divided into blocks without duplication in azimuth to obtain multiple slice images, and the following operation is performed on each slice image:

[0025] Log stretching is performed on the grayscale matrix of the slice image. The expression is:

[0026] I′(x,y)=log(I(x,y)+1)

[0027] Where I represents the original grayscale matrix, I(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the grayscale matrix I, I′ represents the grayscale matrix after log stretching, and I′(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the grayscale matrix I′;

[0028] Perform denoising filtering on the stretched grayscale matrix I′ to obtain a filtered grayscale matrix;

[0029] Map the filtered grayscale matrix to a uniform grayscale range of 0 to 255 to obtain the final grayscale matrix;

[0030] Based on the final grayscale matrix obtained, the grayscale variance of the slice image is determined;

[0031] Screening is performed according to a preset variance threshold, and only slice images whose grayscale variance exceeds the threshold are retained as target area images.

[0032] Optionally, the denoising filter is performed on the stretched grayscale matrix using Lee filtering, and the expression is:

[0033]

[0034] Where s represents the size of the filter, k and l are the horizontal and vertical coordinates of each pixel in the filter, d(k-k0,l-l0) represents the distance between each pixel in the filter grayscale matrix and the center of the filter, k0 and l0 represent the horizontal and vertical coordinates of the center of the filter, respectively. represents the sum of the squares of the distances d(k-k0,l-l0), represents the estimated noise variance, I″ represents the grayscale matrix after filtering;

[0035] The filtered grayscale matrix I″ is mapped to a uniform grayscale range of 0 to 255 using the following formula:

[0036]

[0037] Among them, I″(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the filtered grayscale matrix I″, I″ min and I″ max Respectively represent the minimum and maximum grayscale values ​​of the pixel points in the filtered grayscale matrix I″, I tar (x,y) represents the final grayscale matrix I tar The grayscale value of the pixel with coordinates (x, y). n is a parameter used to control the stretch mapping and is a positive number greater than 1.

[0038] Optionally, the one-dimensional distance signal screening model is implemented using an RC-textCNN model.

[0039] Optionally, the target detection model is implemented using a yolov7 model.

[0040] In a second aspect, the present invention further provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, any of the methods described above is implemented.

[0041] In a third aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed in a computer, causes the computer to execute any of the methods described above.

[0042] The present invention provides a method for imaging and detecting marine ship targets, an electronic device and a storage medium. Aiming at the characteristics of marine SAR images, the imaging detection process is improved. By screening the target area in the early stage of detection and excluding the non-target area, the calculation cost is saved and the ship target detection result is obtained more efficiently. The present invention screens the target area from both the distance and azimuth directions, and then performs ship target detection on the screened target area slice image, thereby avoiding a large amount of imaging and detection of the non-target area, speeding up the overall imaging detection process, and realizing efficient detection of ship targets in marine SAR images. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 This is a schematic diagram of the steps of a method for imaging and detecting a marine ship target provided by an embodiment of the present invention;

[0045] Figure 2is a flow chart of another marine ship target imaging detection method provided by an embodiment of the present invention;

[0046] FIG3( a ) shows preliminary imaging in the range compression domain in one embodiment of the present invention;

[0047] FIG3( b ) shows a one-dimensional range signal after the preliminary imaging in the range compression domain shown in FIG3( a ) is compressed one-dimensionally in the azimuth direction;

[0048] Fig. 4(a) shows a preliminary image of the target response area without the target area;

[0049] FIG4( b ) shows the grayscale distribution diagram corresponding to FIG4( a );

[0050] Fig. 4(c) shows a preliminary image of the target response area with the target area;

[0051] FIG4( d ) shows the grayscale distribution diagram corresponding to FIG4( c );

[0052] Figure 5 A diagram showing the training effect of the yolov7 model in one embodiment of the present invention is shown;

[0053] Figure 6 The imaging detection results of large-scene ocean ship targets using GF-3 data in one embodiment of the present invention are shown. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments in the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0055] As mentioned above, in view of the characteristics of ocean SAR images, it is necessary to propose an imaging detection method with high detection efficiency in order to accelerate remote sensing SAR signal processing, improve the efficiency of effective information acquisition, and thus improve ocean observation capabilities and enhance the ability to grasp and predict ocean situation information.

[0056] The traditional method of marine ship target imaging detection is to first perform complete imaging processing on the SAR echo signal to obtain the entire ocean image, and then perform ship target detection through the target detection model to obtain the position information of the marine ship target. This method has a huge amount of calculation, and because the distribution of ships on the ocean image is not dense, the imaging results of the target-free area are not required for subsequent detection, which reduces the efficiency of the overall imaging detection.

[0057] In view of this, the present invention provides a method for imaging and detecting marine ship targets, which screens target areas in range and azimuth, and then uses the screened target area slice images for ship target detection, which can significantly improve the efficiency of SAR image imaging detection.

[0058] The specific implementation of the above concept is described below.

[0059] Please refer to Figure 1 The embodiment of the present invention provides a method for detecting marine ship target imaging, the method comprising:

[0060] Step 100, obtaining demodulated and processed spaceborne SAR echo data;

[0061] Step 102, based on the spaceborne SAR echo data, by performing matched filtering in the range direction, obtain a corresponding preliminary imaging in the range compression domain;

[0062] Step 104, preliminarily imaging the obtained range compression domain, and performing one-dimensional compression in the azimuth direction to obtain a corresponding one-dimensional range signal, that is, one-dimensional range compression domain data;

[0063] Step 106, using the trained one-dimensional range signal screening model, preliminarily screening the obtained one-dimensional range signals, screening in batches according to a fixed sliding window length, and selecting the one-dimensional range signals corresponding to the target area;

[0064] Among them, screening is performed in batches according to a fixed sliding window length, that is, one-dimensional distance signals are intercepted in batches according to a fixed sliding window length, and input into a trained one-dimensional distance signal screening model for screening;

[0065] The one-dimensional range signal screening model is a neural network model trained with the sample one-dimensional range signal and the marked corresponding range target area as input;

[0066] According to the output of the one-dimensional range signal screening model, it can be determined whether the input one-dimensional range signal corresponds to the target area W;

[0067] Step 108, obtaining a preliminary image of the corresponding target response area through indexing and imaging according to the range of the selected one-dimensional range signal;

[0068] The target response area preliminary image shows grayscale difference; imaging obtains the corresponding target response area preliminary image, and the BP imaging method can be used;

[0069] Step 110, based on the obtained preliminary image of the target response area, performing block screening through regional dynamic quantization to obtain a target area image;

[0070] Step 112, using the trained target detection model to detect each target area image, and obtain corresponding detection results;

[0071] The target detection model is trained with sample target area images as input and corresponding detection results as output, and the detection results include target positions.

[0072] The marine ship target imaging detection method provided by the embodiment of the present invention improves the imaging detection process by screening the target area, excluding the non-target area, imaging the target area and using the intelligent model (i.e., the target detection model) to perform ship target detection, which can effectively save computing costs, improve computing efficiency, and obtain ship target detection results more efficiently.

[0073] Described below Figure 1 How the various steps are performed.

[0074] Optionally, step 102 includes:

[0075] Step 102-0, the satellite-borne SAR echo data s0(τ,t m ) performs Fourier transform in the distance direction to obtain the Fourier transform result; where τ represents the fast time, that is, the time of a pulse in the azimuth direction, t m represents the slow time, i.e., the distance-up pulse repetition time;

[0076] Step 102-2, performing matched filtering in the frequency domain according to the Fourier transform result obtained in step 102-0;

[0077] Step 102-4, based on the matched filtering result obtained in step 102-2, a preliminary imaging in the range compression domain is obtained by inverse Fourier transform of the range. in, represents a distance matched filter.

[0078] Optionally, step 104 includes:

[0079] Based on the obtained range compression domain preliminary imaging s1(τ,t m ), by integrating and superimposing the fast time along the azimuth direction, the corresponding one-dimensional range signal s2(t m )=∫s1(τ,t m )dτ.

[0080] By adopting the above embodiment, the demodulated satellite SAR echo data can be compressed to obtain the corresponding one-dimensional range signal. The one-dimensional range signal area corresponding to the target area has a larger signal amplitude, while the one-dimensional range signal area corresponding to the non-target area generally has a smaller signal amplitude.

[0081] For step 106, the one-dimensional range signal screening model is used to determine whether the input one-dimensional range signal corresponds to a target area in the range. The one-dimensional range signal screening model is implemented using a neural network designed specifically for one-dimensional range image data target area screening, preferably using an RC-textCNN model.

[0082] The RC-textCNN model has the advantage of peak sensitivity and is suitable for the sparse characteristics of ship targets in ocean SAR images. The trained RC-textCNN model can efficiently and accurately identify target areas in ocean SAR images. First, the constructed RC-textCNN model is trained, and the data in the training set is input into the RC-textCNN model. The updated network parameters are obtained through the training cycle process; then, the verified data is input into the RC-textCNN model, and the high-dimensional feature sequence is extracted through the convolution block, and then converted into a one-dimensional sequence. After that, the dimensionality reduction operation is performed through three fully connected layers, and finally the probability of each item is output through the Softmax function. The output probability exceeds the set threshold threshold_tar, which corresponds to the target area W.

[0083] Optionally, step 108 includes:

[0084] Step 108-0, according to the selected one-dimensional distance signal s2(t m ) in the range direction, and extract the corresponding range compression domain data s from the preliminary imaging of the range compression domain through indexing. tar (τ,t m );

[0085] Step 108-2, based on the distance compression domain data s obtained by extraction tar (τ,t m ), and the corresponding preliminary image I of the target response area is obtained by imaging.

[0086] The above embodiment extracts the distance compression domain data corresponding to the target area from the preliminary imaging of the distance compression domain through indexing, and only images the target area obtained by the preliminary screening, thereby effectively reducing the imaging calculation amount.

[0087] Optionally, step 110 includes:

[0088] The obtained preliminary image of the target response area is divided into blocks without duplication in azimuth to obtain multiple slice images, and the following operation is performed on each slice image:

[0089] Step A, log stretching the grayscale matrix of the slice image, the expression is:

[0090] I′(x,y)=log(I(x,y)+1)

[0091] Where I represents the original grayscale matrix, I(x, y) represents the element with coordinates (x, y) in the grayscale matrix I, that is, the grayscale value of the pixel, I′ represents the grayscale matrix after log stretching, and I′(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the grayscale matrix I′;

[0092] Step B, performing denoising filtering on the stretched grayscale matrix I′ to obtain a filtered grayscale matrix I″;

[0093] Step C, mapping the filtered grayscale matrix I″ to a uniform grayscale range of 0 to 255 to obtain the final grayscale matrix I tar ;

[0094] Step D, based on the final grayscale matrix I tar , determine the grayscale variance of the slice image;

[0095] Step E: Screening is performed according to a preset variance threshold, and only the slice images whose grayscale variance exceeds the threshold are retained as the target area images.

[0096] By processing each slice image, several slice images can be obtained as target area images, which are subsequently input into the target detection model for target detection.

[0097] Furthermore, in step B, the stretched grayscale matrix I′ is subjected to denoising filtering using Lee filtering, and the expression is:

[0098]

[0099] Among them, s represents the size of the filter, k and l are the horizontal and vertical coordinates of each pixel in the filter, d(k-k0,l-l0) represents the distance between each pixel in the filter gray matrix and the center point of the filter, k0 and l0 represent the horizontal and vertical coordinates of the center point of the filter, respectively. For the convenience of representation, represents the sum of the squares of the distances d(k-k0,l-l0), represents the estimated noise variance, and I″ represents the grayscale matrix after filtering.

[0100] Lee filtering has good denoising capability, is relatively effective for speckle noise, and is suitable for ocean SAR images. In other embodiments, other denoising filtering schemes may also be used.

[0101] Furthermore, in step C, the filtered grayscale matrix I″ is mapped to a uniform grayscale range of 0 to 255 using the following formula:

[0102]

[0103] Among them, I″(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the filtered grayscale matrix I″, I″ min and I″ max Respectively represent the minimum and maximum grayscale values ​​of the pixel points in the filtered grayscale matrix I″, I tar (x,y) represents the final grayscale matrix I tar The gray value of the pixel with coordinates (x, y), n is a parameter used to control the stretch mapping, and takes a positive number greater than 1. The power function is used to increase the contrast between the target and the background in the image and compensate for the compression loss of the part with higher gray values ​​in the gray mapping. In order to control the degree of stretch mapping, n is usually taken between 2 and 5.

[0104] Through mapping, the value of each element in the stretched and filtered grayscale matrix can be mapped to a real and feasible grayscale interval. The grayscale matrix I tar Reflected in a limited grayscale range, although the brightness of the slices in the target area is reduced, it can still ensure a large contrast between the target and the background, so as to perform variance calculation and image screening.

[0105] By adopting the above embodiment, the image is sliced ​​and divided in the azimuth direction, and the grayscale matrix of each slice image is subjected to log stretching operation, which can compress the original highlight area in the same slice and stretch the grayscale distribution of the background image to enhance the ability to distinguish the target area. Then, with filtering and denoising, the noise and blur in the image can be reduced. Finally, through the operation of sin plus power function, the image grayscale range is mapped to a unified grayscale range of 0 to 255, and the local image grayscale histogram is dynamically stretched to obtain the final slice grayscale matrix. After the above-mentioned regional dynamic quantization, for the target-free area, the corresponding slice image grayscale variance is small, and the overall grayscale is relatively average. For the target area, due to the large difference between the target and the background, the corresponding slice image grayscale variance is more obvious. Although the target grayscale value is high, the grayscale averaged by the overall slice is less. According to the preset grayscale mean threshold, only the slice images with grayscale mean exceeding the threshold are retained, and the target area image can be selected from each slice image.

[0106] For step 112, the target detection model can be implemented using the yolov7 model. The yolov7 model has the advantages of high accuracy and high speed.

[0107] Furthermore, the target detection model is obtained by training, pruning and retraining the yolov7 model.

[0108] Specifically, the target detection model can be obtained in the following way:

[0109] Construct yolov7 model;

[0110] The SSDD open source ship target dataset and the self-made GF3 nearshore target dataset were used as training datasets to perform initial training of the yolov7 model and obtain the initial weight file best.pt;

[0111] Combined with the sparsity characteristics of marine ship targets, the yolov7 model is pruned by weight and neurons, and the model is fine-tuned to ensure detection accuracy. The specific operations include:

[0112] Based on the initial training, sensitivity analysis is used to obtain the importance value of each weight parameter;

[0113] According to a certain pruning ratio ch_sparsity, the least important weight values ​​and neurons are set to zero;

[0114] Fine-tune the pruned model and increase the number of epochs of the training dataset to restore its detection performance;

[0115] Repeat the above steps until the desired model size and detection performance are achieved, and the final detection weight file train_best.pt is obtained.

[0116] In the above embodiment, the present invention adopts the YOLOv7 target detection lightweight compression model obtained by training, pruning and retraining the YOLOv7 model as the target detection model, and considers the influence of land interference factors on the target area, constructs a GF-3 nearshore target data set, and pre-trains the detection model for complex sea and land scenes on the shore through the combination of positive and negative samples, and jointly trains with the SSDD open source ship target data set to improve the accuracy and universality of SAR ship target detection.

[0117] The present invention also verifies the proposed marine ship target imaging detection method, and the parameters required for the detection process are shown in Table 1 below:

[0118] Table 1 Example parameters

[0119]

[0120] like Figures 2 to 6 As shown, the present invention uses the ocean SAR images in the existing database for verification, and the specific steps are as follows Figure 2As shown, reading in the Star Wars SAR echo data corresponds to step 100, the range matched filtering corresponds to step 102, the azimuth one-dimensional compression corresponds to step 104, the trained one-dimensional range signal screening model adopts the RC-TextCNN network, the RC-TextCNN network one-time screening corresponds to step 106, the index distance compression domain data and fast imaging correspond to step 108, the image area dynamic quantization corresponds to step 110, the log function stretching corresponding to step A, the Lee filtering corresponding to step B and the sin+power function mapping corresponding to step C are used to realize the positioning of the target area image, the trained target detection model adopts the yolov7 target detection lightweight compression model, and the intelligent detection of ship targets corresponds to step 112.

[0121] Fig. 3(a) shows the preliminary imaging in the range compression domain in one embodiment of the present invention, which is the range compression domain imaging data that needs to be screened for the first time by the RC-TextCNN model, and is obtained from the spaceborne SAR data after range matching filtering. Fig. 3(b) shows the one-dimensional range signal after the preliminary imaging in the range compression domain shown in Fig. 3(a) is compressed in one dimension in azimuth. Fig. 3(b) is the data further compressed in azimuth based on Fig. 3(a). The RC-TextCNN model analyzes and screens the part with concentrated signal amplitude for the data shown in Fig. 3(a).

[0122] Figure 4(a) shows a preliminary image of the target response area without a target area; Figure 4(b) shows the grayscale distribution map corresponding to Figure 4(a); Figure 4(c) shows a preliminary image of the target response area with a target area; Figure 4(d) shows the grayscale distribution map corresponding to Figure 4(c); In Figures 4(b) and 4(d), the horizontal axis represents the grayscale value, ranging from 0 to 255, and the vertical axis represents the proportion of the number of pixels corresponding to the grayscale value relative to the total number of pixels, and the sum is 1; the secondary screening in the azimuth direction is to exclude the part with a relatively uniform grayscale distribution in the area, so as to locate the target area.

[0123] Figure 5 This is the training effect diagram of the yolov7 target detection lightweight compression model. The accuracy of this target detection model can reach 94.5%, and the detection effect is excellent.

[0124] Figure 6 The results of large-scale ocean ship target imaging detection using GF-3 data are shown. The method provided by the present invention can quickly and efficiently detect ship targets, with a total time of 6.5 seconds, which greatly improves the efficiency of imaging detection and verifies the practicability and reliability of the proposed method.

[0125] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, a method for imaging and detecting marine ship targets in any embodiment of the present invention is implemented.

[0126] An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the processor executes a method for imaging and detecting marine ship targets in any embodiment of the present invention.

[0127] Specifically, a system or device equipped with a storage medium can be provided, on which software program code that implements the functions of any of the above-mentioned embodiments is stored, and a computer (or CPU or MPU) of the system or device can be enabled to read and execute the program code stored in the storage medium.

[0128] In this case, the program code itself read from the storage medium can realize the function of any one of the above-mentioned embodiments, and thus the program code and the storage medium storing the program code constitute a part of the present invention.

[0129] The storage medium embodiments for providing the program code include a floppy disk, a hard disk, a magneto-optical disk, an optical disk (such as CD-ROM, CD-R, CD-RW, DVD-ROM, DVD-RAM, DVD-RW, DVD+RW), a magnetic tape, a non-volatile memory card, and a ROM. Alternatively, the program code can be downloaded from a server computer by a communication network.

[0130] In addition, it should be clear that the functions of any of the above embodiments can be implemented not only by executing the program code read by the computer, but also by enabling an operating system operating on the computer to complete part or all of the actual operations based on instructions from the program code.

[0131] In addition, it can be understood that the program code read from the storage medium is written to a memory provided in an expansion board inserted into the computer or to a memory provided in an expansion module connected to the computer, and then based on the instructions of the program code, a CPU installed on the expansion board or expansion module is enabled to perform part or all of the actual operations, thereby realizing the functions of any of the above-mentioned embodiments.

[0132] In summary, the present invention provides a method, electronic device and storage medium for imaging and detecting marine ship targets. Different from the traditional "imaging + detection" mode, the present invention takes into account the needs of ship target detection. Based on the sparsity characteristics of marine ship targets, during the imaging process, the distance domain image data is compressed in azimuth to perform the first screening, which reduces the amount of data to be processed from the distance upward and reduces the computational cost of the imaging process; before inputting the target detection system, the image is dynamically quantized in grayscale, threshold filtered with the grayscale mean, and a second screening is performed from the azimuth upward, and then the screened slice image is used for ship target detection, which reduces the computational cost of the detection process, speeds up the overall imaging detection process, and achieves efficient detection of the target.

[0133] It should be noted that, in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the statement "comprise a ..." do not exclude the presence of other identical factors in the process, method, article or device including the elements.

[0134] A person of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions, and the aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it executes the steps of the above method embodiments; and the aforementioned storage medium includes: ROM, RAM, magnetic disk or optical disk, etc., various media that can store program codes.

[0135] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting marine ship target imaging, characterized in that: include: Obtain demodulated and processed spaceborne SAR echo data; Based on the spaceborne SAR echo data, a corresponding preliminary imaging in the range compression domain is obtained by performing matched filtering upward in the range; The obtained range compression domain is initially imaged, and the corresponding one-dimensional range signal is obtained by performing one-dimensional compression in the azimuth direction; Using the trained one-dimensional range signal screening model, the obtained one-dimensional range signals are preliminarily screened, and the one-dimensional range signals corresponding to the target areas are screened in batches according to a fixed sliding window length, and the one-dimensional range signal screening model is a neural network model trained with the sample one-dimensional range signal and the marked corresponding range target area as input; According to the range of the selected one-dimensional range signal, a preliminary image of the corresponding target response area is obtained through indexing and imaging; Based on the obtained preliminary image of the target response area, performing block screening through regional dynamic quantization to obtain a target area image; The obtained target area images are detected using a trained target detection model to obtain corresponding detection results; wherein the target detection model is trained with sample target area images as input and corresponding detection results as output, and the detection results include target positions.

2. The method according to claim 1, characterized in that The method of obtaining the corresponding preliminary imaging in the range compression domain by performing matched filtering upward in the range includes: Performing a range-direction Fourier transform on the satellite-borne SAR echo data to obtain a Fourier transform result; According to the obtained Fourier transform result, matched filtering is performed in the frequency domain; Based on the matched filtering results, preliminary imaging in the range compression domain is obtained through inverse Fourier transform.

3. The method according to claim 2, characterized in that The one-dimensional compression in the azimuth direction is performed to obtain a corresponding one-dimensional range signal, including: Based on the obtained preliminary imaging in the range compression domain, the corresponding one-dimensional range signal is obtained by integrating and superimposing the fast time along the azimuth direction.

4. The method according to claim 1, characterized in that: The method of obtaining a preliminary image of a corresponding target response area by indexing and imaging according to the range of the selected one-dimensional range signal includes: According to the range range of the selected one-dimensional range signal, corresponding range compression domain data is extracted from the range compression domain preliminary imaging through indexing; Based on the extracted distance compression domain data, a preliminary image of the corresponding target response area is obtained by imaging.

5. The method according to claim 1, characterized in that The block screening is performed through regional dynamic quantization to obtain a target region image, including: The obtained preliminary image of the target response area is divided into blocks without duplication in azimuth to obtain multiple slice images, and the following operation is performed on each slice image: Log stretching is performed on the grayscale matrix of the slice image. The expression is: I′(x,y)=log(I(x,y)+1) Where I represents the original grayscale matrix, I(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the grayscale matrix I, I′ represents the grayscale matrix after log stretching, and I′(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the grayscale matrix I′; Perform denoising filtering on the stretched grayscale matrix I′ to obtain a filtered grayscale matrix; Map the filtered grayscale matrix to a uniform grayscale range of 0 to 255 to obtain the final grayscale matrix; Based on the final grayscale matrix obtained, the grayscale variance of the slice image is determined; Screening is performed according to a preset variance threshold, and only slice images whose grayscale variance exceeds the threshold are retained as target area images.

6. The method according to claim 5, characterized in that The denoising filter is performed on the stretched grayscale matrix using Lee filtering, and the expression is: Where s represents the size of the filter, k and l are the horizontal and vertical coordinates of each pixel in the filter, d(k-k0,l-l0) represents the distance between each pixel in the filter grayscale matrix and the center of the filter, k0 and l0 represent the horizontal and vertical coordinates of the center of the filter, respectively. represents the sum of the squares of the distances d(k-k0,l-l0), represents the estimated noise variance, I″ represents the grayscale matrix after filtering; The filtered grayscale matrix I″ is mapped to a uniform grayscale range of 0 to 255 using the following formula: Among them, I″(x, y) represents the grayscale value of the pixel with coordinates (x, y) in the filtered grayscale matrix I″, I″ min and I″ max They represent the minimum and maximum grayscale values ​​of the pixel points in the filtered grayscale matrix I″, respectively. n is a parameter used to control the stretch mapping and is a positive number greater than 1. tar (x,y) represents the final grayscale matrix I tar The grayscale value of the pixel with coordinates (x, y).

7. The method according to claim 1, characterized in that The one-dimensional distance signal screening model is implemented using the RC-textCNN model.

8. The method according to claim 1, characterized in that The target detection model is implemented using the yolov7 model.

9. An electronic device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed in a computer, the computer is caused to execute the method according to any one of claims 1 to 8.