Synthetic aperture radar target detection method and device, electronic equipment and medium

Abstract

The invention provides a synthetic aperture radar target detection method. The synthetic aperture radar target detection method comprises the following steps: a client sends a processing request to a server; the server side opens a preview of synthetic aperture radar data to the client side; the client selects data needing to be processed according to the preview; the server side preprocesses the data selected by the client side; the server side constructs a sample library of a deep learning model according to the preprocessed data; the server side carries out training on a target detection and recognition model based on the sample library; and the server side obtains a detection result based on the target detection and identification model according to a request of the client side, and returns the detection result to the client side.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology, and in particular to a synthetic aperture radar target detection method, device, electronic equipment and medium. Background Art

[0002] With the advent of the era of remote sensing big data, data resources are no longer the primary factor restricting the development of remote sensing applications. Currently, the remote sensing industry is undergoing a transformation from a traditional model to a data-driven one, closely integrating with artificial intelligence technology to promote the development of remote sensing applications in a more automated and intelligent direction.

[0003] Synthetic Aperture Radar (SAR), an advanced remote sensing technology, demonstrates unique advantages in disaster response, national defense and security monitoring, and other fields thanks to its all-day, all-weather observation capabilities. However, in practical applications, the implementation and promotion of SAR image target detection technology faces numerous challenges.

[0004] First, low data utilization is a major challenge in current SAR remote sensing applications. Due to data policy restrictions within domestic and international data centers, despite the abundance of high-resolution SAR data resources, user access is limited, and professional guidance documentation is lacking. This results in a severe shortage of samples for high-precision target extraction and recognition, hindering the training of big data-based artificial intelligence models and severely restricting the broader application of SAR data.

[0005] Secondly, SAR data processing is complex, involving extensive expertise and technical challenges. Firstly, the tedious preprocessing of SAR data requires converting single-view complex data into a form acceptable to intelligent models, which requires users to possess profound SAR expertise. However, professional SAR data processing software is currently scarce and difficult to use, increasing both user learning costs and time. Secondly, the construction, training, and inference of AI models also require users to possess solid theoretical knowledge and advanced programming skills, further limiting the widespread adoption and application of SAR intelligent remote sensing technology.

[0006] Another major issue is the lack of service-oriented SAR data application systems. Existing SAR-related application systems and software primarily focus on data distribution and processing, neglecting users' desire to directly purchase services and obtain analyzable results. Users prefer to simplify data processing by purchasing services, allowing them to focus on applying the data analysis results. Furthermore, service-oriented SAR data application systems can reduce unnecessary data distribution, improve data security, enhance user-friendliness, conserve computing resources, and reduce user learning and application costs.

[0007] Therefore, the current technical problems in the field of synthetic aperture radar image target detection, such as low data utilization, great difficulty in data processing, and lack of service-oriented SAR data application systems, need to be solved urgently. Summary of the Invention

[0008] In view of this, the embodiments of the present disclosure provide a synthetic aperture radar target detection method, device, electronic device and medium, which at least partially solve the problems existing in the prior art.

[0009] In a first aspect, an embodiment of the present disclosure provides a synthetic aperture radar target detection method, which includes:

[0010] The client sends a processing request to the server;

[0011] The server opens a preview image of the synthetic aperture radar data to the client;

[0012] The client selects data to be processed according to the preview image;

[0013] The server pre-processes the data selected by the client;

[0014] The server side constructs a sample library of the deep learning model based on the preprocessed data;

[0015] The server side trains the target detection and recognition model based on the sample library;

[0016] The server obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client.

[0017] According to a specific implementation of an embodiment of the present disclosure, the server side preprocesses the data selected by the client side, including: calculating the backscatter coefficient of the ground object on the single-view complex data obtained from the server side to obtain a backscatter coefficient image; converting the synthetic aperture radar system coordinates into geographic coordinates, and converting the slant range image into the ground range image to complete geocoding.

[0018] According to a specific implementation of the embodiment of the present disclosure, target detection is performed on the backscatter image using the target detection and recognition model, and the area where the target is detected is geocoded.

[0019] According to a specific implementation of the embodiment of the present disclosure, the server constructs a sample library of a deep learning model based on the preprocessed data, including:

[0020] Use geographic information system software to open the backscatter coefficient image and mark the target box, create a vector file corresponding to the target box, parse the backscatter coefficient image and the corresponding annotation file, construct a satellite dataset, and convert the satellite dataset format into a sample dataset format suitable for training a deep learning model.

[0021] According to a specific implementation of the embodiment of the present disclosure, the client sends a processing request to the server, including: the client sends a view acquisition request to the server or the client sends a data detection request to the server.

[0022] According to a specific implementation of the embodiment of the present disclosure, the server side obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client, including: dividing the backscatter coefficient image into slices that meet the input size of the neural network for detection, and configuring a preset overlap between the slices;

[0023] Group all prediction boxes according to the categories identified by the model;

[0024] Sort the prediction boxes in each group from high to low according to the confidence score of the model to generate a descending list;

[0025] Take the prediction box with the highest confidence from each list, and calculate the intersection-over-union ratio of the remaining prediction boxes in the list with the prediction box with the highest confidence. If the intersection-over-union ratio is greater than the threshold, delete the prediction box.

[0026] Repeat the calculation of intersection and union until the list is empty.

[0027] According to a specific implementation method of an embodiment of the present disclosure, the server side obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client, including: slicing the detection result, and geocoding the slice and returning it to the client.

[0028] In a second aspect, an embodiment of the present disclosure provides a synthetic aperture radar target detection device, comprising:

[0029] The client is used to send processing requests to the server;

[0030] The server side includes: a data storage module, a data preprocessing module, a sample library construction module and a target detection module;

[0031] The data storage module is used to store the single-view complex-level data of the synthetic aperture radar and open the preview image of the synthetic aperture radar data to the client;

[0032] The data preprocessing module is used to calculate the backscatter coefficient of the ground object for the data selected by the client to obtain a backscatter coefficient image, and complete geocoding for the synthetic aperture radar data;

[0033] The sample library construction module is used to construct a satellite dataset based on the backscatter coefficient image and convert the satellite dataset format into a sample dataset format suitable for training a deep learning model;

[0034] The target detection module is used to train the target detection and recognition model based on the sample data, perform target detection and recognition according to the trained model, and return the target detection result to the client.

[0035] In a third aspect, an embodiment of the present disclosure provides an electronic device, the electronic device comprising:

[0036] at least one processor; and,

[0037] a memory communicatively connected to the at least one processor; wherein,

[0038] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the above-mentioned synthetic aperture radar target detection method.

[0039] In a fourth aspect, an embodiment of the present disclosure provides a non-transitory computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the above-mentioned synthetic aperture radar target detection method.

[0040] In summary, compared with the prior art, this embodiment has the following advantages:

[0041] 1. By introducing a C / S (client / server) architecture, a new service model is provided for SAR remote sensing target detection applications. This model, with application services as its core concept, effectively lowers the barriers to entry for users in terms of data, software, and hardware, making it easier for more users to conduct SAR remote sensing application research. At the same time, this model also significantly improves user experience and increases user satisfaction.

[0042] 2. On the server side, the system integrates multiple links, including data resources, SAR data preprocessing, sample library construction, and intelligent model construction / training / inference, to form a complete software infrastructure system. This system not only improves data utilization in the data center without violating the current remote sensing data distribution policy, but also effectively reduces the computing pressure on the client, enabling the client to process data more efficiently.

[0043] 3. The traditional data preprocessing steps have been reformed. Guided by target detection results, they are integrated with data value-added steps, eliminating unnecessary computational steps. By extracting detection results directly from more primary data and further processing only the areas containing detection results, the time required for data preprocessing is significantly shortened, thereby accelerating the delivery of results. This optimization not only improves work efficiency but also provides users with more timely and accurate services.

[0044] In summary, this embodiment addresses the various difficulties and challenges users encounter when implementing SAR remote sensing target detection applications. By transferring complex data processing processes and heavy computing tasks to the server side of a professional unit, the professional processing on the server side not only ensures efficient data utilization, but also promotes the optimal allocation of data center resources, allowing data resources to maximize the service of SAR remote sensing target detection applications. Users can easily analyze target detection results without directly acquiring or processing raw data. This transformation greatly simplifies the user's workflow, lowers the technical threshold for users in data processing and computing, and enables more users to easily participate in SAR remote sensing target detection applications. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with specific embodiments and accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0046] As used herein, the term "including" and its variations represent open inclusion, i.e., "including but not limited to." Unless otherwise stated, the term "or" means "and / or." The term "based on" means "based at least in part on." The terms "an example embodiment" and "an embodiment" mean "at least one example embodiment." The term "another embodiment" means "at least one additional embodiment." The terms "first," "second," etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0047] Figure 1 A schematic flow chart of a synthetic aperture radar target detection method provided by the first embodiment of the present invention;

[0048] Figure 2 It is a schematic diagram of the preprocessing of SAR data;

[0049] Figure 3 Schematic diagram of the process of building object detection and recognition models;

[0050] Figure 4A schematic structural diagram of a synthetic aperture radar target detection device provided by a second embodiment of the present invention;

[0051] Figure 5 A schematic diagram of a C / S architecture of a client and a server;

[0052] Figures 6A-6B A schematic diagram showing the visualization of the model recognition results;

[0053] Figure 7 This is a schematic diagram of the target slice after geocoding;

[0054] Figure 8 An exemplary structural diagram of a device capable of implementing the method according to an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0055] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.

[0056] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present disclosure.

[0057] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, it should be understood by those skilled in the art that an aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described herein can be used to implement an apparatus and / or practice a method. In addition, other structures and / or functionalities other than one or more of the aspects described herein can be used to implement this apparatus and / or practice this method.

[0058] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The illustrations only show components related to the present disclosure and are not drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component can be changed at will, and the component layout type may also be more complicated.

[0059] Additionally, in the following description, specific details are provided to provide a thorough understanding of the examples. However, one skilled in the art will appreciate that the aspects described can be practiced without these specific details.

[0060] See also Figure 1 A first embodiment of the present invention provides a synthetic aperture radar target detection method, which can be executed by an electronic device, and in particular, by one or more processors within the electronic device, to implement the following steps:

[0061] S101: The client sends a processing request to the server.

[0062] In this embodiment, the electronic device may be a device with computing capabilities such as a laptop computer, a desktop computer, a workstation, a server, or a field programmable gate array (FPGA).

[0063] The interaction between the client and the server mainly relies on the TCP / IP protocol to realize the sending of requirements and the receiving of results, as follows:

[0064] The client provides a quick view request function, which allows users to send a request to the server for a quick view (preview) of L1A (single-view complex data) level data. This step is designed to help users quickly browse and select the required images, greatly simplifying the data screening process. After browsing the quick view, users can select the images for target detection based on their actual needs and send detection instructions to the server through the client. After receiving the user's quick view acquisition or target detection instructions, the server will perform the corresponding processing and send the results (including the quick view or detection result file) back to the client via the TCP / IP protocol. The client is responsible for receiving and displaying these files according to the user's actual needs, ensuring efficient and accurate data exchange.

[0065] S102: The server opens a preview image of synthetic aperture radar data to the client.

[0066] In this embodiment, the server needs to establish a data storage system for storing large amounts of SAR data. This data may include raw data, processed data, and preview images. The data management system should have rapid retrieval and access capabilities so that it can quickly locate the required data upon receiving a client request. Before providing a preview image to the client, the server needs to process the SAR data. This includes steps such as data decompression, format conversion, and image enhancement to ensure the quality and clarity of the preview image. The processing may also involve correction and filtering of the SAR data to reduce noise and interference and improve data usability. The server needs to generate a corresponding preview image based on the characteristics of the SAR data. The preview image should reflect the main features and structure of the SAR data while maintaining a small file size for fast transmission and display. The preview image generation algorithm may include steps such as image scaling, color mapping, and contrast adjustment to ensure that the preview image is visually appealing and easy to understand.

[0067] The generated preview image needs to be transmitted to the client via the network. The server should support multiple data transmission methods, such as HTTP and FTP, to adapt to different network environments and client requirements.

[0068] During data transmission, the server may need to consider the balance between data compression and transmission speed to ensure that the preview image can reach the client quickly and accurately.

[0069] The client needs to provide a user interface that allows the user to enter preview image request parameters, such as the SAR data ID and preview image size. Based on the user's input parameters, the client constructs a preview image request and sends it to the server over the network. The client needs to receive the preview image data sent by the server and decode it into a displayable image format. The client should also provide an image display area to display the received preview image. In this area, the user can browse and select the SAR data to be processed.

[0070] S103: The client selects data to be processed according to the preview image.

[0071] In this embodiment, the client intuitively selects the synthetic aperture radar data to be processed based on the received preview image.

[0072] S104: The server pre-processes the data selected by the client.

[0073] In this embodiment, the SAR data obtained from the data center is usually L1A level, that is, single-view complex data, which contains information from two signal channels and is not directly presented as a visual image. In order to perform intelligent analysis, these data need to undergo a specific preprocessing process.

[0074] Traditional L1A data preprocessing involves two steps: first, radiometric calibration is performed to calculate the backscatter coefficient of the ground object. This step converts the data into an image form recognizable by human vision. However, the image at this point still adheres to the SAR system's side-view imaging geometry (slant-range image), which differs significantly from the geometry of the real geographic environment. Therefore, a further geocoding step is required to convert the SAR system coordinates into geographic coordinates and simultaneously convert the slant-range image into a ground-range image, thereby obtaining a secondary image that conforms to the real geographic geometry. However, the geocoding process involves complex physical and geometric calculations (such as the slant-range-Doppler model), resulting in high algorithmic complexity.

[0075] To reduce the time required for geocoding, this embodiment of the present application applies an intelligent algorithm to target detection directly on the ungeocoded backscatter image. This method efficiently identifies target regions within the image and then geocodes only those regions, significantly reducing overall processing time. Ultimately, the processed results are delivered directly to the user.

[0076] Specifically, for the preprocessing of SAR data, see Figure 2 , Figure 2 Schematic diagram of SAR data preprocessing.

[0077] Take L1A data as an example. Figure 2 As shown in Figure 1, the preprocessing of SAR data includes the following steps:

[0078] S201: Perform radiometric calibration on the L1A-level SAR data to calculate the backscatter coefficient of the ground object, and convert the SAR data into a backscatter coefficient image that can be recognized by human vision.

[0079] In this embodiment, the ground object backscatter coefficient is calculated as follows:

[0080] The calibration constant is determined by using a corner reflector with a known radar cross section in the calibration field and an active calibrator, as shown in formula (1):

[0081] K=[P I -P N G img2 ] / σ c (1)

[0082] Where: P I =I 2 +Q 2 , corresponds to the power of the active calibrator or corner reflector in the SAR complex image, I and Q correspond to the real and imaginary parts of the L1A level complex image respectively. N is the echo noise power; G img2is the imaging processing gain of noise; σ c is the radar cross section (RCS) of the point target; K is the calibration constant. Therefore, ignoring the influence of noise, the above formula can be simplified to:

[0083] K=P I / σ c (2)

[0084] Convert each parameter into dB value, that is

[0085] K dB =10log 10 P I -10log 10 σ c (3)

[0086] This value is written into the L1A image metadata file, and the corresponding field is CalibrationConst.

[0087] For L1A products, the backscatter coefficient can be calculated according to formula (4):

[0088]

[0089] Among them, P I =I 2 +Q 2 , I and Q correspond to the real and imaginary parts of the L1A complex image respectively, QualifyValue is the maximum value of the scene image before quantization, which can be obtained through the QualifyValue field of the L1A image metadata file.

[0090] S202: Performing model analysis on the backscatter coefficient image through a target detection and recognition model to identify the target area portion in the image, and geocoding the target area portion detection result slices.

[0091] Specifically, assuming that the satellite velocity vector corresponding to the imaging moment of a pixel in the WGS84 coordinate system is Satellite position vector They are all functions of time t. Since spaceborne synthetic aperture radar systems are usually processed with zero Doppler, the velocity vector of the ground pixel is considered to be a zero vector in the WGS84 coordinate system. Therefore, the range-zero Doppler equation group can be used to geocode the SAR data. The specific equation group is shown in formula (5):

[0092]

[0093] Wherein, R is the distance (slant distance) between the satellite and the pixel, which can be directly calculated by the parameter file; the satellite velocity vector Vs and the position vector RS can be determined based on the GPS record fitting orbit in the parameter file and the imaging time of the pixel. a and b are the lengths of the major and minor axes of the earth ellipsoid under WGS84 coordinates, respectively. H is the height above the ellipsoid surface (which may need to be calculated or known in specific applications). According to the distance-zero Doppler equation, the WGS84 coordinates X, Y, and Z corresponding to each pixel of the L1A-level data can be calculated, and can be converted into longitude and latitude and altitude coordinates through the earth ellipsoid equation. In actual operation, the elevation information in the digital elevation model (DEM) data is usually used for numerical iterative calculation to complete the final longitude and latitude calculation. This step is usually time-consuming. However, the embodiment of the present application only calculates the target area detected on the backscatter coefficient image, thereby significantly reducing the time consuming of the overall process and improving the calculation efficiency.

[0094] S105, the server side constructs a sample library of the deep learning model based on the preprocessed data.

[0095] In this embodiment, based on the radiation calibration of L1A-level products, professional GIS software is used to open and process backscatter coefficient images, and corresponding vector files are created by marking target boxes on these images. These files follow the ESRI Shapefile format standard. Then, the backscatter coefficient images and their annotation files are parsed to construct a satellite dataset. In order to improve the versatility and compatibility of the dataset, this embodiment provides a set of interface conversion tools that can convert satellite datasets into the COCO dataset format widely used in the field of artificial intelligence. This format can seamlessly connect to mainstream deep learning frameworks such as MMDetection and PaddlePaddle for model training.

[0096] The target slices are obtained by cutting from the backscatter coefficient data based on the geometric center of the target. These slices are consistent with the original data in terms of image format, DN value (digital value representing the brightness or reflection intensity of the pixel, etc.), projection information and resolution. In order to match the resolution of different sensors and the training data size commonly used in deep learning networks, this embodiment follows the following cutting rules: For data with a sampling interval less than or equal to 0.5 meters, the slice size is set to 1024 pixels × 1024 pixels. For data with a sampling interval between 0.5 meters and 5 meters (inclusive), the slice size is adjusted to 512 pixels × 512 pixels. For data with other resolutions, the slice size is unified to 256 pixels × 256 pixels. Taking the sliding beam (SL) mode of the high-resolution satellite No. 3 as an example, its nominal resolution is 1 meter, so a slice format with a size of 512 × 512 pixels centered on the target is adopted. In addition, the parsing file adopts the XML format and must contain the key structure information shown in Table (1).

[0097]

[0098]

[0099] S106: The server trains a target detection and recognition model based on the sample library.

[0100] This embodiment provides a set of programs that can convert a dataset that follows the format specifications of the domestic remote sensing satellite data sample library into the COCO dataset format, which is widely compatible with major deep learning frameworks. Using these frameworks, the dataset can be used for model training and testing. Typically, 70% of the data in the sample library is used as a training set, and the remaining 30% is used as a test set. By adjusting hyperparameters such as the number of training rounds and learning rate, and repeatedly training and testing, until the model's detection rate on the test set reaches 90% and the false alarm rate is less than 10%, the training and testing process is completed, and the resulting model can be used for subsequent model detection and recognition.

[0101] In this embodiment, a rich sample library is constructed, which contains a large amount of SAR image data, specifically for the detection and identification of ship targets. On the server side, deep learning technology is used to train an efficient SAR ship target detection and recognition model based on these sample libraries. When the user makes a request to the server through the client to perform ship target detection and recognition on SAR image data, the server will respond quickly and call the neural network model that has been trained and optimized for intelligent reasoning. This reasoning process can analyze the input SAR backscatter coefficient image target slice, and through a multi-layer convolutional neural network (CNN) structure, the processed backscatter image is subjected to feature extraction, thereby accurately determining the position of the ship target and identifying its category. For details, see Figure 3 , Figure 3 Schematic diagram of the process of building models for object detection and recognition.

[0102] S107: The server obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client.

[0103] Specifically, deep learning models typically receive small slices of fixed size as neural network input during training, but in the actual target detection and recognition process, the input is a whole backscatter coefficient image. Therefore, during the model target detection and recognition stage, the backscatter coefficient image needs to be segmented into multiple slices that meet the neural network input size for detection. In order to avoid the target being cut off due to being located at the edge of the slice, which would cause the model to miss detection, a 50% overlap is used to cut the slices. Considering that the resolution of the SAR images mainly used is less than 1 meter, a slice size of 512×512 is selected, with a preset overlap, such as 50% overlap, to ensure that the sizes of most typical targets can be covered.

[0104] However, due to the overlap between adjacent slices, the same object may appear in multiple slices, causing the model to detect the object multiple times and form multiple detection boxes on the same object. To solve this problem, the non-maximum suppression algorithm (NMS) is used to optimize the detection results and remove redundant detection boxes. The specific steps of the algorithm are as follows:

[0105] First, all predicted boxes are grouped according to the categories identified by the model.

[0106] Then, the prediction boxes in each group are sorted from high to low according to the confidence score of the model to generate a descending list.

[0107] Next, the prediction box with the highest confidence score is taken from each list, and its intersection over union (IoU) with the remaining prediction boxes in the list is calculated. If the IoU between a prediction box and the current highest confidence prediction box is greater than a set threshold, the prediction box is removed from the list.

[0108] Repeat the above steps until the list is empty. Finally, the remaining predicted box is the optimal detection box.

[0109] After detecting a target in a SAR image, a series of slices containing the target's location and related information are generated. To ensure that these detection results correspond to the actual geographic location, these detection result slices are geocoded. The geocoding process primarily converts the pixel coordinates in the detection result slices into longitude and latitude coordinates in a geographic coordinate system. This typically involves a series of complex mathematical transformations, including but not limited to projection transformations and coordinate conversions. After geocoding, these detection result slices with geographic location information are packaged or serialized for transmission to the client over the network. After receiving this data, the client can further display, analyze, or process it.

[0110] In summary, compared with the prior art, this embodiment has the following advantages:

[0111] 1. This invention adopts a client-server architecture, achieving seamless connection between the user end and the server end. Users only need to use the client to enjoy one-stop application services such as request acceptance and target detection visualization results acquisition, without having to download and process large amounts of SAR data themselves, thus greatly simplifying the user operation process. Users can devote more time and energy to in-depth analysis and interpretation of detection results, rather than wasting time on tedious data processing, thus significantly improving research efficiency;

[0112] 2. The server-side model construction relies on a large amount of archived SAR data. This data covers target information in a variety of scenarios and conditions, providing a rich sample resource for model training. By efficiently utilizing this archived data, this embodiment can build a more accurate and generalizable SAR target detection model, thereby improving the overall utilization of SAR data in the field of target detection.

[0113] 3. The server is responsible for completing complex processes such as SAR data preprocessing, sample library construction, and model training. These tasks often have high technical barriers and computing requirements for ordinary users. By shifting these tasks to the server, this embodiment effectively reduces the difficulty for users to develop applications using SAR data, allowing users without specialized technical skills and computing resources to easily enjoy SAR target detection services. Furthermore, the server's powerful computing capabilities can also reduce the computing pressure on users, conserving their computing resources and further improving the user experience.

[0114] In summary, this embodiment achieves intelligent, efficient and convenient SAR target detection services by adopting a C / S architecture, making full use of archived data, and reducing the difficulty of user application, providing users with a higher quality and more efficient SAR target detection service experience.

[0115] The second embodiment of the present invention provides a synthetic aperture radar target detection device (such as Figure 4 shown), which includes:

[0116] The client 410 is used to send a processing request to the server. This embodiment provides a simple client for user use, whose core functions are concentrated on the sending and receiving of messages / instructions / files with the server. The client is developed using the Java programming language to ensure its cross-platform compatibility and good user interaction experience. In terms of communication protocol, the client uses the TCP / IP network transmission protocol, and establishes a data flow channel with the server by specifying the server's IP address and the corresponding service port. The client first attempts to establish a TCP / IP connection with the server, and makes a connection request through the specified server IP address and port number. When the user sends the "Please send quicklookimages" command through the client interface, the client will encapsulate the command into a data packet and send it to the server. Subsequently, the client enters a waiting state, ready to receive the quick view file sent by the server. Once the quick view file is received, the client will parse it and print out the quick view information, and then end the session.

[0117] If the server fails to execute the command normally, it will send an error message to the client. After receiving the error message, the client will immediately print it out for the user's reference.

[0118] Users can also request the server to perform target detection on the specified data by sending the "Please detect: [data name]" command through the client. After sending the command, the client enters a waiting state again, preparing to receive the server's response message. If it receives the "[Server Info] detection successful" message, it indicates that the detection was successful. At this point, the client will begin receiving the visual detection result file and complete the task after receiving it. If it receives other messages from the server (such as detection failure information), the client will print the corresponding error message to let the user know the reason for the detection failure.

[0119] The server side 420 includes: a data storage module 4201, a data pre-processing module 4202, a sample library construction module 4203 and a target detection module 4204;

[0120] The data storage module is used to store the single-view complex-level data of the synthetic aperture radar and open the preview image of the synthetic aperture radar data to the client;

[0121] The data preprocessing module is used to calculate the backscatter coefficient of the ground object for the data selected by the client to obtain a backscatter coefficient image, and complete geocoding for the synthetic aperture radar data;

[0122] The sample library construction module is used to construct a satellite dataset based on the backscatter coefficient image and convert the satellite dataset format into a sample dataset format suitable for training a deep learning model;

[0123] The target detection module is used to train the target detection and recognition model based on the sample data, perform target detection and recognition according to the trained model, and return the target detection result to the client.

[0124] When the server starts, the system first registers the server's workspace. This step includes setting the data path, application path, and other necessary system resources. These resources will serve as the basis for subsequent threads to handle the actual detection tasks. By pre-registering these resources, the server can ensure that when handling client requests, it can quickly and accurately locate and call the required resources, thereby improving task processing efficiency and accuracy.

[0125] The server will then create a socket object to implement network communication. At the same time, the server will register a service port (such as port 8888) open to the client so that the client can establish a connection with the server through this port. By establishing a socket object and registering the service port, the server provides the necessary network infrastructure for communication with the client. The server begins listening to port 8888 to wait for the client's connection request. Once the client's connection request is heard, the server will immediately assign a worker thread to the client in the thread pool. The worker thread is responsible for processing the client's request, including receiving instructions, executing detection tasks, returning results, etc. Through the management of the thread pool, the server can efficiently manage thread resources and prevent excessive consumption of server resources due to too many client connections.

[0126] After allocating a thread to a client and processing its request, the server will continue to listen on port 8888 to wait for the next client's connection request.

[0127] This process will repeat until the administrator actively shuts down the server. By continuously monitoring and looping, the server can ensure that all client requests are processed in a timely manner while maintaining system stability and reliability.

[0128] This example describes the correspondence between client and server worker threads and their functional implementation, while also highlighting deadlock avoidance mechanisms. A one-to-one correspondence is established between client and server worker threads, ensuring that each thread does not interfere with each other when processing client requests, improving the system's concurrent processing capabilities and stability.

[0129] When the main thread listens to the client's connection request, it will immediately start a corresponding worker thread.

[0130] After the worker thread is started, it immediately establishes a data communication channel with the client for the subsequent transmission of messages, instructions, and files. The worker thread continuously listens for instructions or messages from the client to ensure that the client's requests can be processed in a timely manner.

[0131] When a worker thread receives a command from a client, it immediately interprets the command and responds accordingly. This response includes, but is not limited to, providing server status information or error messages during command execution, allowing the client to stay informed of the server's operational status.

[0132] Search for the quick view path in the workspace registered on the server and send the quick view to the client according to the client's instructions.

[0133] The client searches for the raw data required by the client in the workspace registered on the server, and calls the intelligent detection model program in a blocking manner to perform SAR target detection and generate visual detection results.

[0134] The visualization results of successful detection are sent according to the client instructions; if the detection fails, an error message is returned to the client.

[0135] When the client disconnects, the resources occupied by the worker thread are released to ensure efficient use of system resources.

[0136] When a worker thread blocks calls to the intelligent detection model program, it simultaneously launches two auxiliary threads to prevent deadlocks caused by cache overflow. These auxiliary threads monitor the input and error streams between the worker thread and the main process, respectively. By promptly acquiring and processing inter-thread communication cache information, these two auxiliary threads ensure that the cache does not overflow, effectively avoiding deadlocks.

[0137] The C / S architecture of client and server can be found in Figure 5 , Figure 5 This is a schematic diagram of a C / S architecture of a client and a server. Figure 5 As shown in the figure, in a C / S (Client / Server) architecture, the TCP / IP communication protocol is used to build a data transmission channel. The server continuously listens to a preset service port, waiting for connection requests from clients. Whenever a client attempts to establish a connection with the server, the server immediately allocates a separate processing thread for the client.

[0138] This independent thread is specifically designed to handle all requests related to the client. It receives data and instructions from the client and then allocates server-side data and application resources based on these instructions. During this process, the thread performs a series of complex operations tailored to user needs, including but not limited to data preprocessing and intelligent SAR target detection.

[0139] During the data preprocessing phase, the thread processes the received raw data to ensure it meets the requirements of subsequent analysis. Next, using the SAR target intelligent detection solution, the thread conducts in-depth analysis of the processed data to identify and locate the target objects in the SAR image.

[0140] Finally, the thread encapsulates the analysis results into a data format that the client can understand and returns them to the client through the previously established TCP / IP channel. In this way, the client can obtain the SAR target detection results in real time and make corresponding decisions or actions based on these results.

[0141] In addition, the present invention also tests the present invention. The details are as follows:

[0142] The L1A level products of the Gaofen-3 satellite in spotlight mode collected on May 30, 2021 were processed and corresponding case tests were carried out based on them. The specific information of the image data used is as follows:

[0143] Name: GF3_KAS_SL_025289_E139.7_N35.3_20210530_L1A_HH_L10005673715

[0144] Data Specifications: The unpacked data file is 1.65GB in size and contains a single-view complex image stored in tiff format, along with accompanying metadata files. These files together constitute the basic data resources for this test.

[0145] To ensure the smooth progress of the test, a server with powerful computing capabilities was selected. Its key software and hardware parameters are as follows:

[0146] CPU Configuration: The server is equipped with two Hygon C86 3185 8-core processors, each with a clock speed of up to 3.00GHz. This configuration provides ample computing resources for complex data processing tasks.

[0147] Memory capacity: The server is equipped with up to 256GB of memory, ensuring that data exchange and caching requirements during data processing are fully met.

[0148] GPU acceleration: To further improve data processing speed, the server is also equipped with an NVIDIA Quadro RTX6000 graphics card. This graphics card has 24GB of video memory and can efficiently support parallel computing tasks, especially in application scenarios such as image processing and deep learning.

[0149] Operating system: The server runs Windows 10 Professional for Workstations 22H2 64-bit operating system.

[0150] The following will show in detail the specific performance of the embodiments of the present application and the traditional solution in the test, including the calculation time, the size of the data obtained by the user, and the usage of the user's computer hardware and software resources.

[0151] (1) Computation time comparison

[0152] To evaluate the efficiency of the embodiment of the present application, the full-process calculation time of the embodiment of the present application and the traditional solution were tested on the same server. The test results show that the traditional solution takes 28 minutes and 25.0934 seconds to complete the full-process calculation, while the embodiment of the present application only takes 7 minutes and 42.1988 seconds, saving about 73% of the calculation time. The specific time distribution is shown in Table (2):

[0153]

[0154] (2) Comparison of user acquisition data size

[0155] This embodiment adopts a client-server architecture, aiming to reduce the burden of data processing on users, allowing them to focus more on analyzing and using the results. By comparison, if users process data independently without using this embodiment, they would need to obtain all L1A-level data, which, in this test example, would be 1.65GB in size. However, with this solution, users only need to obtain the visualization results, namely 29 geocoded L2-level target slices, with a data size of only 572KB, significantly reducing the amount of data required.

[0156] (3) Comparison of user computer software and hardware resource usage

[0157] From the perspective of user usability, the use of computer software and hardware resources was compared when users used this solution and when they processed data themselves. The test results showed:

[0158] In this embodiment, users only need to install the client software (47.4KB) and configure the JAVA runtime environment (256MB) on their computers. The client software is simple in function and only needs to complete operations such as sending instructions and receiving results, without the need for GPU computing acceleration support.

[0159] Process data by yourself: users need to configure GPU acceleration, artificial intelligence Python computing environment (about 8.77GB, including Cuda, Paddle, Numpy, etc.), backscatter coefficient calculation, geocoding, training, intelligent analysis and other programs (about 450MB), artificial intelligence model (about 200MB) and the data set used for training the model (about 1.2GB). In addition, users also need to process the original data (the test sample is 1.65GB) and the intermediate data in the data preprocessing process (the test sample is about 423MB), and generate intermediate files for the model training process (about 249GB). At the same time, users also need to invest a lot of manpower and material resources in data labeling and training, including manually labeling 2191 positive and negative samples, and the total training time is about 34 hours. In addition, users also need to have rich professional knowledge for environment configuration, code debugging, data processing, training parameter adjustment and other tasks. Figures 6A-6BA visual diagram of the model recognition results is shown. Figure 7 A schematic diagram of the target tiles after geocoding is shown.

[0160] A third embodiment of the present invention further provides an electronic device, comprising:

[0161] at least one processor; and,

[0162] a memory communicatively connected to the at least one processor; wherein,

[0163] The memory stores instructions that can be executed by the at least one processor. The instructions are executed by the at least one processor to enable the at least one processor to perform the synthetic aperture radar target detection method of any of the aforementioned embodiments.

[0164] The fourth embodiment of the present invention further provides a non-transitory computer-readable storage medium, wherein the non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the synthetic aperture radar target detection method described in any of the aforementioned embodiments.

[0165] The fifth embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-transitory computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes the synthetic aperture radar target detection method of any of the aforementioned embodiments.

[0166] The sixth embodiment of the present invention further provides a computer program, which includes program instructions. When the program instructions are executed by a computer, the computer is caused to execute the synthetic aperture radar target detection method of any of the aforementioned embodiments.

[0167] Figure 8 A schematic diagram of a method or device 1000 that can implement an embodiment of the present invention is shown. In some embodiments, the method or device 1000 may include more or fewer devices than shown. In some embodiments, the method or device 1000 may be implemented using a single device or multiple devices. In some embodiments, the method or device 1000 may be implemented using cloud-based or distributed devices.

[0168] like Figure 8As shown, device 1000 includes a processor 1001, which can perform various appropriate operations and processes according to the programs and / or data stored in a read-only memory (ROM) 1002 or the programs and / or data loaded from a storage portion 1008 into a random access memory (RAM) 1003. Processor 1001 can be a multi-core processor or can include multiple processors. In some embodiments, processor 1001 can include a general-purpose main processor and one or more special coprocessors, such as a central processing unit (CPU), a graphics processing unit (GPU), a neural network processor (NPU), a digital signal processor (DSP), etc. Various programs and data required for the operation of device 1000 are also stored in RAM 1003. Processor 1001, ROM 1002, and RAM 1003 are connected to each other via a bus 1004. An input / output (I / O) interface 1005 is also connected to bus 1004.

[0169] The processor and memory are used together to execute the program stored in the memory. When the program is executed by the computer, the methods, steps or functions described in the above embodiments can be implemented.

[0170] The following components are connected to the I / O interface 1005: an input section 1006 including a keyboard, a mouse, a touch screen, and the like; an output section 1007 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and a speaker; a storage section 1008 including a hard disk and the like; and a communication section 1009 including a network interface card such as a LAN card or a modem. The communication section 1009 performs communication processing via a network such as the Internet. A drive 1010 is also connected to the I / O interface 1005 as needed. A removable medium 1011, such as a magnetic disk, an optical disk, a magneto-optical disk, or a semiconductor memory, is installed in the drive 1010 as needed, so that computer programs read therefrom can be installed into the storage section 1008 as needed. Figure 8 Only some components are shown schematically, which does not mean that the device 1000 only includes Figure 8 Components shown.

[0171] The systems, devices, modules, or units described in the above embodiments may be implemented by a computer or its associated components. The computer may be, for example, a mobile terminal, a smartphone, a personal computer, a laptop computer, an in-vehicle human-computer interaction device, a personal digital assistant, a media player, a navigation device, a game console, a tablet computer, a wearable device, a smart TV, an Internet of Things system, a smart home, an industrial computer, a server, or a combination thereof.

[0172] Although not shown, in an embodiment of the present invention, a computer-readable storage medium is provided, on which a computer program / instruction is stored, and when the computer program / instruction is executed by a processor, the text adversarial sample generation method described in Example 1 is implemented.

[0173] Storage media in embodiments of the present invention include permanent and non-permanent, removable and non-removable items that can be used to store information using any method or technology. Examples of storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information that can be accessed by a computing device.

[0174] Although not shown, an embodiment of the present invention further provides a computer program product, including: a computer program / instruction, which implements the synthetic aperture radar target detection method described in Example 1 when executed by a processor.

[0175] The methods, programs, systems, and apparatuses of the embodiments of the present invention may be executed or implemented in a single or multiple networked computers, or may be practiced in a distributed computing environment. In the embodiments of this specification, in these distributed computing environments, tasks may be performed by remote processing devices connected via a communication network.

[0176] Those skilled in the art will appreciate that the embodiments of this specification may be provided as methods, systems, or computer program products. Therefore, those skilled in the art will appreciate that the functional modules / units or controllers and related method steps described in the above embodiments may be implemented using software, hardware, or a combination of software / hardware.

[0177] Unless explicitly stated, the actions or steps of the methods, procedures, and methods described in accordance with the embodiments of the present invention do not have to be performed in a specific order and can still achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0178] In this document, multiple embodiments of the present invention are described, but for the sake of brevity, the description of each embodiment is not exhaustive, and the same or similar features or parts between the embodiments may be omitted. In this document, "one embodiment", "some embodiments", "example", "specific example", or "some examples" are intended to apply to at least one embodiment or example according to the present invention, but not all embodiments. The above terms do not necessarily mean to refer to the same embodiment or example. Those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are mutually contradictory.

[0179] While the exemplary systems and methods of the present invention have been specifically shown and described with reference to the foregoing embodiments, these are merely examples of the best modes for implementing the present systems and methods. Those skilled in the art will appreciate that various changes may be made to the embodiments of the systems and methods described herein when implementing the present systems and / or methods without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A synthetic aperture radar target detection method, characterized in that: include: The client sends a processing request to the server; The server opens a preview image of the synthetic aperture radar data to the client; The client selects data to be processed according to the preview image; The server pre-processes the data selected by the client; The server side constructs a sample library of the deep learning model based on the preprocessed data; The server side trains the target detection and recognition model based on the sample library; The server obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client.

2. The synthetic aperture radar target detection method according to claim 1, wherein: The server side preprocesses the data selected by the client side, including: calculating the backscatter coefficient of the ground object on the single-view complex data obtained from the server side to obtain a backscatter coefficient image; converting the synthetic aperture radar system coordinates into geographic coordinates, and converting the slant range image into the ground range image to complete geocoding.

3. The synthetic aperture radar target detection method according to claim 2, characterized in that: Target detection is performed on the backscatter image using the target detection and recognition model, and the region where the target is detected is geocoded.

4. The synthetic aperture radar target detection method according to claim 2, wherein: The server constructs a sample library of the deep learning model based on the preprocessed data, including: Use geographic information system software to open the backscatter coefficient image and mark the target box, create a vector file corresponding to the target box, parse the backscatter coefficient image and the corresponding annotation file, construct a satellite dataset, and convert the satellite dataset format into a sample dataset format suitable for training a deep learning model.

5. The synthetic aperture radar target detection method according to claim 1, wherein: The client sends a processing request to the server, including: the client sends a view acquisition request to the server or the client sends a data detection request to the server.

6. The synthetic aperture radar target detection method according to claim 1, characterized in that: The server side obtains a detection result based on the target detection and recognition model according to the request of the client and returns the detection result to the client, including: dividing the backscatter coefficient image into slices that meet the input size of the neural network for detection, and configuring a preset overlap between the slices; Group all prediction boxes according to the categories identified by the model; Sort the prediction boxes in each group from high to low according to the confidence score of the model to generate a descending list; Take the prediction box with the highest confidence from each list, and calculate the intersection-over-union ratio of the remaining prediction boxes in the list with the prediction box with the highest confidence. If the intersection-over-union ratio is greater than the threshold, delete the prediction box. Repeat the calculation of intersection and union until the list is empty.

7. The synthetic aperture radar target detection method according to claim 1, characterized in that: The server obtains a detection result based on the target detection and recognition model according to the request of the client, and returns the detection result to the client, including: slicing the detection result, geocoding the slices and returning them to the client.

8. A synthetic aperture radar target detection device, characterized in that: include: The client is used to send processing requests to the server; The server side includes: a data storage module, a data preprocessing module, a sample library construction module and a target detection module; The data storage module is used to store the single-view complex-level data of the synthetic aperture radar and open the preview image of the synthetic aperture radar data to the client; The data preprocessing module is used to calculate the backscatter coefficient of the ground object for the data selected by the client to obtain a backscatter coefficient image, and complete geocoding for the synthetic aperture radar data; The sample library construction module is used to construct a satellite dataset based on the backscatter coefficient image and convert the satellite dataset format into a sample dataset format suitable for training a deep learning model; The target detection module is used to train the target detection and recognition model based on the sample data, perform target detection and recognition according to the trained model, and return the target detection result to the client.

9. An electronic device, characterized in that: The electronic device includes: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the synthetic aperture radar target detection method described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium, characterized in that The non-transitory computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the synthetic aperture radar target detection method according to any one of claims 1 to 7.

Images