High-frequency radar time-frequency domain ship target detection method and system based on improved YOLOv5s

Abstract

The application discloses a high-frequency radar time-frequency domain ship target detection method and system based on an improved YOLOv5s, and the method comprises the following steps: according to the obtained high-frequency radar receiving data, an initial radar detection matrix and a pretreated time-frequency spectrum are calculated, and the distance element information of a target is obtained according to the pretreated time-frequency spectrum; the pretreated time-frequency spectrum is input into the trained improved YOLOv5s model for target detection, so that the boundary box of the target is obtained; a greedy algorithm is used to extract the time-frequency ridge of the boundary box of the target in combination with the corresponding pretreated time-frequency spectrum, and the central Doppler frequency of the target is obtained according to the extracted time-frequency ridge; and finally, the target radar detection matrix is obtained according to the distance element and the central Doppler frequency of the target in combination with the initial radar detection matrix. The application can guarantee the ease of use of the method while improving the detection performance of the high-frequency radar on the ship target.

Description

Technical Field

[0001] This invention relates to the field of ship target detection, specifically to a high-frequency radar time-frequency domain ship target detection method and system based on an improved YOLOv5s. Background Technology

[0002] High-frequency radar systems can achieve continuous, all-weather, 24 / 7 monitoring of large sea areas, making them an important technical means for dynamic monitoring of the marine environment. However, when using such systems for ship target detection, the echoes from ships often exhibit non-stationary Doppler and energy characteristics due to fluctuations in the ship's cross-section and speed. Under these circumstances, traditional constant false alarm rate (CFAR) detection algorithms are insufficient to achieve ideal detection results.

[0003] Due to the advantage of time-frequency analysis methods in capturing the time-varying characteristics of non-stationary signals, many scholars have devoted themselves to researching time-frequency domain ship target detection methods to improve the target detection performance of high-frequency radar systems. The core idea of ​​high-frequency radar time-frequency domain detection methods is to transform the traditional range-Doppler spectrum midpoint target detection task into a ridge line extraction task in the time spectrum, generally designed using edge tracking algorithms combined with energy determination. However, such methods require the integration of human experience to set a large number of parameters to control the detection process, and their adaptive detection performance is poor in situations with variable noise floors at high frequencies. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, this invention provides a high-frequency radar time-frequency domain ship target detection method and system based on improved YOLOv5s. By improving the YOLOv5s model, the high-frequency radar ship target energy detection task is transformed into a target detection task in image processing, achieving the technical effects of good detection effect, strong feature extraction capability and convenient operation.

[0005] According to one aspect of the present invention, a high-frequency radar time-frequency domain ship target detection method based on an improved YOLOv5s is provided, comprising: Based on the acquired high-frequency radar received data, the initial radar detection matrix and preprocessed time spectrum are calculated, and the range information of the target is obtained based on the preprocessed time spectrum. The preprocessed temporal spectrum is input into the trained improved YOLOv5s model for target detection to obtain the bounding box of the target; A greedy algorithm is used to extract the time-frequency ridge of the target by combining its bounding box with its corresponding preprocessed time-frequency spectrum, and the center Doppler frequency of the target is obtained based on the extracted time-frequency ridge. Based on the target's range element and center Doppler frequency, combined with the initial radar detection matrix, the final target radar detection matrix is ​​obtained. The training of the improved YOLOv5s model includes: Construct a training set for the preprocessed spectrum; The bottleneck structure of YOLOv5s is optimized by using dynamic snake convolution module and triple attention mechanism module to build an improved YOLOv5s model; The improved YOLOv5s model is trained using the training set of the preprocessed time spectrum to obtain the trained improved YOLOv5s model.

[0006] As a further technical solution, after acquiring the high-frequency radar received data, it also includes: Based on the acquired high-frequency radar received data, obtain the high-frequency radar range-slow time spectrum; Based on the high-frequency radar range-slow time spectrum, the range-Doppler spectrum and the time spectrum of each range element are calculated respectively. Based on the dimension of the range-Doppler spectrum, an initial radar detection matrix of the same dimension is generated.

[0007] As a further technical solution, a training set for the preprocessed temporal spectrum is constructed, including: Based on the distance-Doppler spectrum, the first-order spectral region Doppler range and the ground clutter Doppler range are calculated respectively. Set the first-order spectral region and the region of ground clutter Doppler within the time spectrum to 0; The time spectrum after the zeroing operation is normalized to obtain the preprocessed time spectrum.

[0008] As a further technical solution, the method also includes: dividing the preprocessed time spectrum into a training set and a test set of the preprocessed time spectrum according to a preset ratio.

[0009] As a further technical solution, the method also includes: The initial radar detection matrix is ​​determined based on the target's received data field information; Based on the target's range information and center Doppler frequency, the corresponding element in the initial radar detection matrix is ​​set to 1; The detection results in the initial radar detection matrix are merged using an aggregation method to obtain the final radar detection matrix.

[0010] As a further technical solution, the method further includes: after a preset time or a preset number of reception sessions, retraining the constructed improved YOLOv5s model using a new preprocessed time-spectrum training set to obtain new pre-trained weights.

[0011] According to one aspect of the present invention, a high-frequency radar time-frequency domain ship target detection system based on an improved YOLOv5s is provided, comprising: The first main module is used to calculate the initial radar detection matrix and the preprocessed time spectrum based on the acquired high-frequency radar received data, and to obtain the target's range information based on the preprocessed time spectrum. The second main module is used to input the preprocessed temporal spectrum into the trained improved YOLOv5s model for target detection and to obtain the bounding box of the target. The third main module is used to extract the time-frequency ridge of the target by combining the bounding box of the target with its corresponding preprocessed time-frequency spectrum using a greedy algorithm, and to obtain the center Doppler frequency of the target based on the extracted time-frequency ridge. The fourth main module is used to obtain the final target radar detection matrix based on the target's range element and center Doppler frequency, combined with the initial radar detection matrix. The training of the improved YOLOv5s model includes: Construct a training set for the preprocessed spectrum; The bottleneck structure of YOLOv5s is optimized by using dynamic snake convolution module and triple attention mechanism module to build an improved YOLOv5s model; The improved YOLOv5s model is trained using the training set of the preprocessed time spectrum to obtain pre-trained weights, thus completing the training of the improved YOLOv5s model.

[0012] As a further technical solution, it also includes: a fifth main module, which is used to retrain the constructed improved YOLOv5s model using a new preprocessed time-spectrum training set after a preset time or a preset number of reception sessions to obtain new pre-trained weights.

[0013] According to one aspect of the present invention, an electronic device is provided, comprising: at least one processor, at least one memory, and a communication interface; wherein the processor, memory, and communication interface communicate with each other; the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to perform the method described thereon.

[0014] According to one aspect of the present invention, a non-transitory computer read storage medium is provided, the non-transitory computer read storage medium storing computer instructions that cause the computer to perform the method described herein.

[0015] This invention addresses the non-stationary characteristics of ship target echoes in high-frequency radar. Leveraging the analytical advantage of capturing time-varying signals in the time-frequency domain, it transforms the high-frequency radar ship target energy detection task into a target detection task within image processing. Compared to existing technologies, this invention offers the following advantages: The detection performance is good. By annotating time-frequency ridge samples of ship targets with cross-Doppler elements and time-varying amplitude in the training set, adaptive detection of strong targets, weak targets, and non-stationary targets can be achieved. With strong feature extraction capabilities, the improved YOLOv5s network significantly enhances its ability to extract time-frequency ridges by introducing dynamic snake convolution and triple attention mechanism modules. It is easy to operate. Since deep learning networks are end-to-end target detection models, only a reasonable confidence threshold needs to be set to control the target detection task, which is beneficial to improving the performance of high-frequency radar for ship target detection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a high-frequency radar time-frequency domain ship target detection method based on an improved YOLOv5s, provided as an embodiment of the present invention.

[0018] Figure 2 The network architecture diagram of the improved YOLOv5s model provided in the embodiments of the present invention is shown.

[0019] Figure 3 A schematic diagram of the CBL module structure in the improved YOLOv5s model provided in this embodiment of the invention.

[0020] Figure 4 This is a schematic diagram of the DSCBL module structure in the improved YOLOv5s model provided in this embodiment of the invention.

[0021] Figure 5 A schematic diagram of the C3 module structure in the improved YOLOv5s model provided in this embodiment of the invention.

[0022] Figure 6 This is a schematic diagram of the C3-DSCT module structure in the improved YOLOv5s model provided in this embodiment of the invention.

[0023] Figure 7 This is a schematic diagram illustrating the detection results of the improved YOLOv5 model for the input time spectrum provided in an embodiment of the present invention.

[0024] Figure 8 This is a schematic diagram illustrating the result of using a greedy algorithm to extract the time-frequency ridge of the predicted target's bounding box in conjunction with its corresponding preprocessed time-frequency spectrum, as provided in an embodiment of the present invention.

[0025] Figure 9 This is a schematic diagram of the target parameter conversion process provided in an embodiment of the present invention.

[0026] Figure 10 This is a schematic diagram of a radar detection matrix without merging operations, provided in an embodiment of the present invention.

[0027] Figure 11 This is a schematic diagram of a radar detection matrix performing a merging operation, provided in an embodiment of the present invention.

[0028] Figure 12 This is a schematic diagram of a high-frequency radar time-frequency domain ship target detection system based on an improved YOLOv5s, provided as an embodiment of the present invention.

[0029] Figure 13 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0030] It should be noted that: In recent years, data-driven deep learning models have demonstrated strong potential in tasks such as target detection. Multiple studies have shown that using a sufficiently large training set can enable deep learning models to achieve superior detection performance. The continuous operation capability of high-frequency radar allows this system to obtain abundant samples for model training. Therefore, the inventors, using the time-spectrum data of high-frequency radar as the data source and based on the YOLOv5s deep learning target detection network, combined with a greedy algorithm and image morphology processing techniques, designed a method suitable for detecting ship targets using high-frequency radar, aiming to improve the detection performance of ship targets using high-frequency radar.

[0031] The terms “comprising” and “having”, and any variations thereof, in the specification, claims, and accompanying drawings of this invention are intended to cover a non-exclusive inclusion, such as a process, method, system, product, or apparatus that includes a series of steps or units, not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0032] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices. The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be decomposed, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0033] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0034] This invention provides a high-frequency radar time-frequency domain ship target detection method based on an improved YOLOv5s, such as... Figure 1 As shown, it includes: Step 1: Calculate the initial radar detection matrix and preprocessed time spectrum based on the acquired high-frequency radar received data, and obtain the target range information based on the preprocessed time spectrum. Step 2: Input the preprocessed temporal spectrum into the trained improved YOLOv5s model for target detection to obtain the bounding box of the target; Step 3: Use a greedy algorithm to extract the time-frequency ridge of the target's bounding box by combining it with its corresponding preprocessed time-frequency spectrum, and obtain the target's center Doppler frequency based on the extracted time-frequency ridge; Step 4: Based on the target's range element and center Doppler frequency, and combined with the initial radar detection matrix, obtain the final target radar detection matrix.

[0035] The training of the improved YOLOv5s model includes: Construct a training set for the preprocessed spectrum; The bottleneck structure of YOLOv5s is optimized by using dynamic snake convolution module and triple attention mechanism module to build an improved YOLOv5s model; The improved YOLOv5s model is trained using the training set of the preprocessed time spectrum to obtain the trained improved YOLOv5s model.

[0036] Furthermore, such as Figures 2 to 6 As shown, the bottleneck structure of YOLOv5s is optimized using the Dynamic Snake Convolution module and the Triplet Attention module to construct an improved YOLOv5s network model. Then, the improved YOLOv5s model is trained using the constructed time-spectrum training set to obtain the optimal pre-trained weights.

[0037] In the high-frequency radar time-frequency domain ship target detection method based on the improved YOLOv5s provided in this embodiment of the invention, such as... Figure 6 As shown, dynamic serpentine convolution is designed for tubular topological structures. By limiting the connectivity of the region of interest, it controls the dynamic transformation of the convolution kernel, exhibiting excellent detection performance for continuous tubular targets. Given that target echoes from high-frequency radar often present as long and continuous curves in the time-frequency domain, replacing the original two-dimensional convolution module with dynamic serpentine convolution can improve the network's target detection performance in the application scenario of this invention.

[0038] In the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s provided in this embodiment of the invention, the features generated by the network of this invention are... Figure 3 The dimensions correspond to the time, Doppler frequency, and semantic information dimensions of the target echo, respectively. The triple attention mechanism module is a low-computation attention mechanism module; introducing this module helps the network learn features. Figure 3 The module helps improve the network's ability to analyze the interrelationships between time and Doppler frequency, time and semantics, and semantics and Doppler frequency in the target echo, thereby optimizing the network's performance in detecting target time-frequency ridges.

[0039] As a preferred embodiment, the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s provided in this embodiment of the invention further includes: collecting multiple high-frequency radar received data to obtain the high-frequency radar range-slow time spectrum and received data field information.

[0040] Furthermore, the present invention provides a high-frequency radar time-frequency domain ship target detection method based on an improved YOLOv5s, such as... Figure 9As shown, it also includes: First, using the Fast Fourier Transform method to obtain the range-Doppler spectrum based on the high-frequency radar range-slow time spectrum; Second, using the multiple synchronization squeezing algorithm to obtain the time spectrum of each range element based on the high-frequency radar range-slow time spectrum; Finally, generating a radar detection matrix of the same dimension based on the dimension of the range-Doppler spectrum, with all elements of the matrix set to 0.

[0041] Preferably, the distance-slow time spectrum is defined as follows: Its dimension size is Dimension, where N is the maximum distance element to be observed, and M is the number of sweep cycles. In order to receive the data field sequence number, in the embodiment... .

[0042] Preferably, the distance-Doppler spectrum is defined as follows: Its dimension size is Dimension, where N is the maximum distance element to be observed, and M is the number of sampling points in the Doppler frequency dimension. In order to receive the data field sequence number, in the embodiment... .

[0043] Preferably, the time spectrum is defined as follows: Its dimension size is Dimension, among which It is the number of sampling points in the time dimension of the time spectrum. The number of Doppler frequency dimension sampling points in the time spectrum. For the data field sequence number to be received, The distance element corresponding to the time spectrum, in the embodiment .

[0044] Preferably, the radar detection matrix is ​​defined as follows: Its dimension size is Dimension, where N is the maximum distance element to be observed, and M is the number of sweep cycles. To receive the data field sequence number, in the embodiment... .

[0045] As a preferred embodiment, the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s provided by this invention further includes: first, obtaining the first-order spectral region Doppler range based on the ocean echo Bragg scattering theorem according to the range-Doppler spectrum obtained in the aforementioned steps, and obtaining the ground clutter Doppler range based on the stationary characteristics of ground clutter; second, setting the first-order spectral region and the ground clutter Doppler region within the obtained time spectrum to 0; and third, normalizing the time spectrum after the 0-setting operation to obtain the preprocessed time spectrum.

[0046] Preferably, the preprocessing time spectrum is defined as follows: Its dimension size is Dimension, among which This refers to the number of sampling points in the time dimension of the frequency spectrum. The number of Doppler frequency dimension sampling points in the time spectrum. The received data field sequence number is n, where n is the distance element corresponding to the time spectrum. In the embodiment, .

[0047] Furthermore, during model training, the preprocessed time-frequency spectrum is divided into a training set and a detection set according to a preset ratio. The improved YOLOv5s model is trained using the training set, and the trained improved YOLOv5s model is used to detect targets on the detection set. It should be noted that the preset ratio can be adaptively adjusted according to actual conditions. After obtaining the training set of the preprocessed time-frequency spectrum, time-frequency ridge samples of ship targets annotated across Doppler elements and with time-varying amplitude can be trained to achieve adaptive detection of strong targets, weak targets, and non-stationary targets. Figure 7 As shown, the lines are not straight and sway left and right, indicating that the target time-frequency ridge line crosses Doppler elements, and the brightness of the lines is not uniform, indicating that the amplitude of the target time-frequency ridge line varies over time.

[0048] It should be noted that after the trained improved YOLOv5s model has been used for a period of time, a new preprocessed time spectrum can be obtained based on new high-frequency radar received data. The model can then be retrained using the new preprocessed time spectrum to obtain new optimal training weights.

[0049] As a preferred embodiment, the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s provided by this invention further includes: first, recording the target's range metadata and received data field information according to the preprocessed time spectrum to be detected; second, using the optimal pre-trained weights obtained previously, setting a confidence threshold to perform target detection on the preprocessed time spectrum, and obtaining the predicted target's bounding box, such as... Figure 7 As shown; next, a greedy algorithm is used to extract the time-frequency ridge from the predicted target's bounding box and its corresponding preprocessed time-frequency spectrum, thereby calculating the target's center Doppler frequency, as shown. Figure 8 As shown.

[0050] Preferably, the confidence threshold is defined as follows: .

[0051] As shown in the figure, the improved YOLOv5 model detects the input time-frequency spectrum, with the red rectangle representing the bounding box of the predicted target, generated by YOLOv5 network inference. The figure also shows the result of using a greedy algorithm to extract the time-frequency ridge from the predicted target's bounding box in conjunction with its corresponding preprocessed time-frequency spectrum; the yellow line represents the target's time-frequency ridge extracted by the greedy algorithm. This embodiment of the invention utilizes the improved YOLOv5 model to predict the target's bounding box, and then combines it with a greedy algorithm for high-frequency radar time-frequency domain ship target detection. The single time-frequency ridge highlighted in yellow in the figure oscillates within a certain Doppler range; by calculating the Doppler mean of this time-frequency ridge, the target's center Doppler frequency can be obtained.

[0052] As a preferred embodiment, the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s provided in this embodiment of the invention further includes: first, finding the corresponding radar detection matrix based on the obtained received data field information; then, setting the corresponding element in the radar detection matrix to 1 based on the obtained target range element information and the target's center Doppler frequency; finally, merging the detection results in the radar detection matrix using an aggregation method to obtain the final radar detection matrix.

[0053] Preferably, the final radar detection matrix is ​​defined as follows: Its dimension size is The dimension is defined as follows: N is the maximum distance element to be observed, M is the number of sampling points in the Doppler frequency dimension, and k is the sequence number of the received data field. In the embodiment... .

[0054] It should be noted that because targets may spread in high-frequency radar, the same target may have multiple detection results in an adjacent area within the radar detection matrix, such as... Figure 10 As shown, a merging operation is therefore required. Specifically, graphical image processing techniques can be used to obtain the distribution of each 8-connected neighborhood region in the radar detection matrix, and then the center of each 8-connected neighborhood can be calculated as the target detection result, such as... Figure 11 As shown.

[0055] The implementation of the various embodiments of the present invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s. This device is used to execute the high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s in the above method embodiments.

[0056] See Figure 12The system includes: a first main module, used to calculate an initial radar detection matrix and a preprocessed time spectrum based on acquired high-frequency radar received data, and to obtain the target's range metadata based on the preprocessed time spectrum; a second main module, used to input the preprocessed time spectrum into a trained improved YOLOv5s model for target detection, and to obtain the target's bounding box; a third main module, used to use a greedy algorithm to extract the target's bounding box and its corresponding preprocessed time spectrum for time-frequency ridge extraction, and to obtain the target's center Doppler frequency based on the extracted time-frequency ridge; a fourth main module... The module is used to obtain the final target radar detection matrix based on the target's range element and center Doppler frequency, combined with the initial radar detection matrix. The training of the improved YOLOv5s model includes: constructing a training set of preprocessed temporal spectra; optimizing the bottleneck structure of YOLOv5s using a dynamic serpentine convolution module and a triple attention mechanism module to construct an improved YOLOv5s model; and training the improved YOLOv5s model using the preprocessed temporal spectra training set to obtain pre-training weights, thus completing the training of the improved YOLOv5s model.

[0057] The high-frequency radar time-frequency domain ship target detection system based on the improved YOLOv5s provided in this invention addresses the problem that ship target echoes in high-frequency radar systems often exhibit non-stationary characteristics such as Doppler migration and energy fluctuations. Figure 12 Several modules in the system, based on the analytical advantage of capturing time-varying signals in the time-frequency domain, transform the high-frequency radar ship target energy detection task into a target detection task in image processing. This enables adaptive detection of strong targets, weak targets, and non-stationary targets. Furthermore, by improving the YOLOv5s network, the ability to extract time-frequency ridges has been significantly enhanced, thereby improving the performance of high-frequency radar ship target detection.

[0058] It should be noted that the system embodiments provided by this invention, in addition to implementing the methods in the above method embodiments, are also used to implement the methods in other method embodiments provided by this invention. The difference lies only in setting corresponding functional modules, and their principles are basically the same as those of the above system embodiments provided by this invention. As long as those skilled in the art, based on the above system embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and improve the equipment in the above system embodiments while ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example: Based on the above system embodiments, as a preferred embodiment, the high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s provided in this embodiment of the invention further includes: The first calculation submodule is used to obtain the high-frequency radar range-slow time spectrum based on the acquired high-frequency radar received data; The second calculation submodule is used to calculate the range-Doppler spectrum and the time spectrum of each range element based on the high-frequency radar range-slow time spectrum. The third calculation submodule is used to generate an initial radar detection matrix of the same dimension as the range-Doppler spectrum.

[0059] Based on the above system embodiments, as a preferred embodiment, the high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s provided in this embodiment of the invention further includes: The fourth calculation submodule is used to calculate the first-order spectral region Doppler range and the ground clutter Doppler range based on the range-Doppler spectrum. The fifth calculation submodule is used to set the first-order spectral region and the region of ground clutter Doppler within the time spectrum to 0; The sixth calculation submodule is used to normalize the time spectrum after the zeroing operation to obtain the preprocessed time spectrum.

[0060] Based on the above system embodiments, as a preferred embodiment, the high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s provided in this embodiment of the invention further includes: dividing the preprocessed time spectrum into a training set and a test set of the preprocessed time spectrum according to a preset ratio.

[0061] Based on the above system embodiments, as a preferred embodiment, the high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s provided in this embodiment of the invention further includes: The seventh calculation submodule is used to determine the initial radar detection matrix based on the target's received data field information; The eighth calculation submodule is used to set the corresponding element in the initial radar detection matrix to 1 based on the target's range information and center Doppler frequency; The ninth calculation submodule is used to merge the detection results in the initial radar detection matrix in an aggregation manner to obtain the final radar detection matrix.

[0062] Based on the above system embodiments, as a preferred embodiment, the high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s provided in this embodiment of the invention further includes: after a preset time or a preset number of reception sessions, retraining the constructed improved YOLOv5s model using a new preprocessed time-frequency training set to obtain new pre-trained weights.

[0063] The method in this embodiment of the invention is implemented using an electronic device; therefore, it is necessary to introduce the relevant electronic device. For this purpose, embodiments of the present invention provide an electronic device, such as... Figure 13 As shown, the electronic device includes: at least one processor, a communication interface, at least one memory, and a communication bus, wherein the at least one processor, the communication interface, and the at least one memory communicate with each other via the communication bus. The at least one processor invokes logical instructions stored in the at least one memory to execute all or part of the steps of the methods provided in the foregoing method embodiments.

[0064] Furthermore, when the logical instructions in at least one of the aforementioned memories are implemented as software functional units and sold or used as independent products, they are stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, is embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (a personal computer, server, or network device) to execute all or part of the steps of the methods described in the various method embodiments of the present invention. The aforementioned storage medium includes: USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks—various media for storing program code.

[0065] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, located in one place, or distributed across multiple network units. The purpose of this embodiment is achieved by selecting some or all of the modules according to actual needs. Those skilled in the art will understand and implement this without any inventive effort.

[0066] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0067] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0068] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0069] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0070] In summary, this invention obtains the high-frequency radar range-slow time spectrum based on received high-frequency radar data; calculates the range-Doppler spectrum and the time spectrum of each range element based on the range-slow time spectrum; calculates the first-order spectrum and the Doppler range of ground clutter based on the range-Doppler spectrum; obtains a preprocessed time spectrum based on the first-order spectrum and the ground clutter range, which is used to create a time-frequency target detection dataset; trains an improved YOLOv5s using the time-frequency target detection dataset to obtain pre-trained weights; detects ship target ridges in the time spectrum using the pre-trained weights; applies a greedy algorithm to obtain the target center Doppler; and finally, based on image processing technology, maps the time-frequency domain target detection results to the radar detection matrix corresponding to the range-Doppler spectrum. This invention can improve the detection performance of high-frequency radar for ship targets while ensuring ease of use.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s, characterized in that, include: Based on the acquired high-frequency radar received data, the initial radar detection matrix and preprocessed time spectrum are calculated, and the range information of the target is obtained based on the preprocessed time spectrum. The preprocessed temporal spectrum is input into the trained improved YOLOv5s model for target detection to obtain the bounding box of the target; A greedy algorithm is used to extract the time-frequency ridge of the target by combining its bounding box with its corresponding preprocessed time-frequency spectrum, and the center Doppler frequency of the target is obtained based on the extracted time-frequency ridge. Based on the target's range element and center Doppler frequency, combined with the initial radar detection matrix, the final target radar detection matrix is ​​obtained; The training of the improved YOLOv5s model includes: Construct a training set for the preprocessed spectrum; The bottleneck structure of the C3 module in the top-down path of the bottleneck layer of YOLOv5s is optimized by using a dynamic snake convolution module and a triple attention mechanism module to build an improved YOLOv5s model. Specifically, the optimization is to replace the first CBL layer of the bottleneck structure in the C3 module with a dynamic snake convolution module and add a triple attention mechanism after the second CBL layer. The improved YOLOv5s model is trained using the training set of the preprocessed time spectrum to obtain the trained improved YOLOv5s model.

2. The high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s according to claim 1, characterized in that, After acquiring the high-frequency radar received data, the process also includes: Based on the acquired high-frequency radar received data, obtain the high-frequency radar range-slow time spectrum; Based on the high-frequency radar range-slow time spectrum, the range-Doppler spectrum and the time spectrum of each range element are calculated respectively. Based on the dimension of the range-Doppler spectrum, an initial radar detection matrix of the same dimension is generated.

3. The high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s according to claim 2, characterized in that, Construct a training set for the preprocessed spectra, including: Based on the distance-Doppler spectrum, the first-order spectral region Doppler range and the ground clutter Doppler range are calculated respectively. Set the first-order spectral region and the region of ground clutter Doppler within the time spectrum to 0; The time spectrum after the zeroing operation is normalized to obtain the preprocessed time spectrum.

4. The high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s according to claim 3, characterized in that, The method further includes: dividing the preprocessed time spectrum into a training set and a test set of the preprocessed time spectrum according to a preset ratio.

5. The high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s according to claim 1, characterized in that, The method further includes: The initial radar detection matrix is ​​determined based on the target's received data field information; Based on the target's range information and center Doppler frequency, the corresponding element in the initial radar detection matrix is ​​set to 1; The detection results in the initial radar detection matrix are merged using an aggregation method to obtain the final radar detection matrix.

6. The high-frequency radar time-frequency domain ship target detection method based on improved YOLOv5s according to claim 1, characterized in that, The method further includes: after a preset time or a preset number of reception sessions, retraining the constructed improved YOLOv5s model using a new preprocessed time-spectrum training set to obtain new pre-trained weights.

7. A high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s, characterized in that, include: The first main module is used to calculate the initial radar detection matrix and the preprocessed time spectrum based on the acquired high-frequency radar received data, and to obtain the target's range information based on the preprocessed time spectrum. The second main module is used to input the preprocessed temporal spectrum into the trained improved YOLOv5s model for target detection and to obtain the bounding box of the target. The third main module is used to extract the time-frequency ridge of the target by combining the bounding box of the target with its corresponding preprocessed time-frequency spectrum using a greedy algorithm, and to obtain the center Doppler frequency of the target based on the extracted time-frequency ridge. The fourth main module is used to obtain the final target radar detection matrix based on the target's range element and center Doppler frequency, combined with the initial radar detection matrix. The training of the improved YOLOv5s model includes: Construct a training set for the preprocessed spectrum; The bottleneck structure of the C3 module in the top-down path of the bottleneck layer of YOLOv5s is optimized by using a dynamic snake convolution module and a triple attention mechanism module to build an improved YOLOv5s model. Specifically, the optimization is to replace the first CBL layer of the bottleneck structure in the C3 module with a dynamic snake convolution module and add a triple attention mechanism after the second CBL layer. The improved YOLOv5s model is trained using the training set of the preprocessed time spectrum to obtain pre-trained weights, thus completing the training of the improved YOLOv5s model.

8. The high-frequency radar time-frequency domain ship target detection system based on improved YOLOv5s according to claim 7, characterized in that, Also includes: The fifth main module is used to retrain the improved YOLOv5s model using a new pre-processed time-spectrum training set after a preset time or a preset number of reception sessions, so as to obtain new pre-trained weights.

9. An electronic device, characterized in that, include: The system includes at least one processor, at least one memory, and a communication interface; wherein the processor, memory, and communication interface communicate with each other. The memory stores program instructions that are executed by the processor, which invokes the program instructions to perform the method described in any one of claims 1 to 6.

10. A non-transitory computer read storage medium, characterized in that, The non-transitory computer read storage medium stores computer instructions, which cause the computer to perform the method described in any one of claims 1 to 6.

Images