Radar Target Detection Method and System Incorporating the Characteristics of the Difference in the Spatial Distribution of Fusion Amplitudes

Abstract

The present application discloses a radar target detection method and system that integrates the amplitude spatial distribution difference feature, belonging to the field of computer technology. The method includes: obtaining a feature tensor obtained by performing feature extraction on a radar image to be detected; performing maximum pooling operation and minimum pooling operation on the input feature tensor through a convolution kernel to obtain a maximum pooling tensor and a minimum pooling tensor; constructing an encoding rule according to the maximum pooling tensor and the minimum pooling tensor; encoding the feature tensor according to the encoding rule to obtain an encoded tensor; multiplying the maximum pooling tensor by the encoded tensor and taking the absolute value to obtain a first tensor; the first tensor represents the amplitude spatial distribution difference of the feature tensor; using the first tensor as a magnification factor to perform signal enhancement on the feature tensor and output a target feature map; performing target detection based on the target feature map to obtain a radar target detection result. The present application can improve the ability to extract target feature information and improve the accuracy of radar target detection.

Description

Technical Field

[0001] This application relates to the technical field of radar target detection, and particularly to a radar target detection method and system that fuse the amplitude spatial distribution difference features. Background Art

[0002] When performing radar target detection, related technologies often select a suitable amplitude threshold from a global perspective to distinguish targets and clutter. However, this method may mis-filter targets with small amplitudes or areas in the target region where the amplitudes are close to those of sea clutter.

[0003] According to the characteristic that the amplitude difference between the target region and the pure sea clutter region is obvious, simply using a method such as window sampling to capture the amplitude features of the target in the peripheral region is very likely to regard the peripheral region of the target as sea clutter and filter it out. Once this happens, the original target may be detected as multiple targets, or after the peripheral region is filtered out, the remaining core region of the target is too small and is regarded as noise with abnormal amplitude and filtered out. Summary of the Invention

[0004] The main purpose of the embodiments of this application is to propose a radar target detection method and system that fuse the amplitude spatial distribution difference features, aiming to accurately distinguish targets and sea clutter, and further achieve precise detection of targets.

[0005] To achieve the above object, on the one hand, an embodiment of this application proposes a radar target detection method that fuses the amplitude spatial distribution difference features. The method includes:

[0006] Obtain a feature tensor obtained by performing feature extraction on the radar image to be detected;

[0007] Perform max-pooling operation and min-pooling operation on the input feature tensor through a convolutional kernel to obtain a max-pooling tensor and a min-pooling tensor;

[0008] Construct an encoding rule according to the max-pooling tensor and the min-pooling tensor;

[0009] Encode the feature tensor according to the encoding rule to obtain an encoded tensor;

[0010] Multiply the max-pooling tensor by the encoded tensor and take the absolute value to obtain a first tensor; wherein, the first tensor represents the amplitude spatial distribution difference of the feature tensor;

[0011] Use the first tensor as a magnification factor to perform signal enhancement on the feature tensor and output a target feature map;

[0012] Perform target detection based on the target feature map to obtain a radar target detection result.

[0013] In some embodiments, according to the maximum pooling tensor and the minimum pooling tensor, an encoding rule is constructed, including the following steps:

[0014] Calculate the difference between the maximum pooling tensor and the minimum pooling tensor to obtain a difference tensor;

[0015] Obtain the global maximum difference and the global minimum difference from the difference tensor to obtain a global amplitude distribution interval;

[0016] Based on the global amplitude distribution interval, determine the encoding rule.

[0017] In some embodiments, encoding the feature tensor according to the encoding rule to obtain an encoded tensor includes the following steps:

[0018] Calculate the interval width according to the global amplitude distribution interval and the preset number of intervals;

[0019] Divide the global amplitude distribution interval into a plurality of encoding interval ranges according to the interval width;

[0020] Multiply each difference element in the difference tensor by the interval width to obtain an intermediate value;

[0021] Use the serial number of the encoding interval range to which the intermediate value belongs as the element corresponding encoding of the difference element;

[0022] Form an encoded tensor with all the element corresponding encodings.

[0023] In some embodiments, the expression of the encoding interval range is:

[0024] ;

[0025] ;

[0026] Wherein, represents the encoding interval range; represents the global minimum difference; k represents the serial number of the encoding interval range; represents the interval width; n represents the number of encoding intervals.

[0027] In some embodiments, the expression of the encoded tensor is:

[0028] ;

[0029] Wherein, is an element of the encoded tensor; b represents the number of samples; c represents the number of channels; w represents the width; h represents the height; represents an element of the difference tensor.

[0030] In some embodiments, the method further includes the following steps:

[0031] Obtain a radar image to be detected;

[0032] Preprocess the radar image to be detected to obtain a first radar image;

[0033] Extract features from the radar image to be detected through a backbone network combined with channel attention to obtain a first feature map;

[0034] Perform feature fusion on the first feature map through a feature pyramid structure to obtain a feature tensor.

[0035] In some embodiments, in the step of preprocessing the radar image to be detected to obtain a first radar image, the preprocessing steps include at least one of the following steps:

[0036] Perform data augmentation processing on the radar image to be detected;

[0037] Perform adaptive anchor box calculation processing on the radar image to be detected;

[0038] Perform adaptive image scaling processing on the radar image to be detected.

[0039] To achieve the above object, on the other hand, an embodiment of the present application proposes a radar target detection system that fuses the amplitude spatial distribution difference features. The system includes:

[0040] A first module for obtaining a feature tensor obtained by extracting features from a radar image to be detected;

[0041] A second module for performing a max-pooling operation and a min-pooling operation on the input feature tensor through a convolutional kernel to obtain a max-pooling tensor and a min-pooling tensor;

[0042] A third module for constructing an encoding rule according to the max-pooling tensor and the min-pooling tensor;

[0043] A fourth module for encoding the feature tensor according to the encoding rule to obtain an encoded tensor;

[0044] A fifth module for multiplying the max-pooling tensor by the encoded tensor and taking the absolute value to obtain a first tensor; wherein, the first tensor represents the amplitude spatial distribution difference of the feature tensor;

[0045] A sixth module for using the first tensor as a magnification factor to perform signal enhancement on the feature tensor and output a target feature map;

[0046] The seventh module is used to perform target detection based on the target feature map to obtain a radar target detection result.

[0047] To achieve the above object, on the other hand, an embodiment of the present application provides an electronic device, which includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the foregoing method is implemented.

[0048] To achieve the above object, on the other hand, an embodiment of the present application provides a computer-readable storage medium storing a computer program, and when the computer program is executed by a processor, the foregoing method is implemented.

[0049] The embodiments of the present application at least include the following beneficial effects: The present application provides a radar target detection method and system that fuse amplitude spatial distribution difference features. The solution obtains a feature tensor obtained by performing feature extraction on a radar image to be detected; performs a max-pooling operation and a min-pooling operation on the input feature tensor through a convolution kernel to obtain a max-pooling tensor and a min-pooling tensor; constructs an encoding rule according to the max-pooling tensor and the min-pooling tensor; encodes the feature tensor according to the encoding rule to obtain an encoded tensor; multiplies the max-pooling tensor by the encoded tensor and takes the absolute value to obtain a first tensor; wherein, the first tensor represents the amplitude spatial distribution difference of the feature tensor; uses the first tensor as a magnification factor to perform signal enhancement on the feature tensor and outputs a target feature map; performs target detection based on the target feature map to obtain a radar target detection result. The overall steps extract the amplitude spatial distribution difference features from the radar image, perform signal enhancement on the feature vector based on the amplitude spatial distribution difference features, can strengthen the ability to extract target feature information, better distinguish targets from sea clutter, and improve the accuracy of radar target detection. Description of the Drawings

[0050] The drawings are used to provide a further understanding of the technical solutions of the present application, and constitute a part of the specification. They are used to explain the technical solutions of the present application together with the embodiments of the present application, and do not constitute a limitation on the technical solutions of the present application.

[0051] Figure 1 is a flowchart of a radar target detection method that fuses amplitude spatial distribution difference features provided by an embodiment of the present application;

[0052] Figure 2 is a visualization schematic diagram of amplitude spatial distribution difference features provided by an embodiment of the present application;

[0053] Figure 3 is a schematic structural diagram of a target detection network provided by an embodiment of the present application;

[0054] Figure 4It is the overall network structure diagram provided by the embodiment of the present application after adding the ST layer;

[0055] Figure 5 It is the detailed structure diagram of the ST layer provided by the embodiment of the present application;

[0056] Figure 6 The structure diagram of the C3SE module provided by the embodiment of the present application;

[0057] Figure 7 For the feature map of Channel 1 of the Acode module in the embodiment of the present application, it shows the case where Channel 1 does not use the Acode module and the case where Channel 1 uses the Acode module;

[0058] Figure 8 For the feature map of the Acode module in the embodiment of the present application, it shows the case where Channel 2 does not use the Acode module;

[0059] Figure 9 For the feature map of the Acode module in the embodiment of the present application, it shows the case where Channel 1 uses the Acode module;

[0060] Figure 10 For the feature map of the Acode module in the embodiment of the present application, it shows the case where Channel 2 uses the Acode module;

[0061] Figure 11 The schematic diagram of the module of the radar target detection system for fusing the amplitude spatial distribution difference features provided by the embodiment of the present application;

[0062] Figure 12 The schematic diagram of the hardware structure of the electronic device provided by the embodiment of the present application. Detailed implementation manners

[0063] In order to make the purpose, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and are not used to limit the present application. When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the embodiments of the present application. They are only examples of devices and methods consistent with some aspects of the embodiments of the present application as detailed in the appended claims.

[0064] Although the functional modules are divided in the system schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order from the module division in the system or the order in the flowchart. The terms "first / S100", "second / S200", etc. in the specification, claims and the above-mentioned drawings are used to distinguish similar objects, and do not have to be used to describe a specific order or sequence.

[0065] It will be understood that the terms "first", "second", etc. used in this application may be used herein to describe various concepts, but unless otherwise specified, these concepts are not limited by these terms. These terms are only used to distinguish one concept from another. For example, without departing from the scope of the embodiments of this application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, as used herein, the words "if", "when" may be interpreted as "when...", "when...", or "in response to determining".

[0066] The terms "at least one", "a plurality of", "each", "any one", etc. used in this application, at least one includes one, two or more than two, a plurality includes two or more than two, each refers to each of the corresponding plurality, and any one refers to any one of the plurality.

[0067] Reference to "embodiments" in this text means that the specific features, structures, or characteristics described in connection with the embodiments may be included in at least one embodiment of this application. The phrase appears in various places in the specification and does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art will explicitly and implicitly understand that the embodiments described herein may be combined with other embodiments.

[0068] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0069] In the related art, a suitable amplitude threshold is often selected from a global perspective to distinguish targets from clutter. However, this method may mis-filter targets with small amplitudes, or areas in the target region that are close to sea clutter. And according to the characteristic that the amplitude difference between the target region and the pure sea clutter region is obvious, simply using the window sampling method to capture the amplitude characteristics of the target in the peripheral region is very likely to regard the peripheral region as sea clutter and be filtered out. Once this situation occurs, the original target may be detected as multiple targets, or after the peripheral region is filtered out, the remaining target core region is too small and is regarded as noise with abnormal amplitude and is filtered out, affecting the accuracy of the target detection result.

[0070] In view of this, in the embodiments of the present application, a radar target detection method and system that fuse the amplitude spatial distribution difference features are provided. This solution extracts the amplitude spatial distribution difference features of the echo through encoding, and combines a target detection network and a channel attention mechanism, which can better enhance the target signal and improve the detection accuracy of the target signal. By introducing a channel attention mechanism to improve the C3 module, the data information extraction ability between channels is optimized. Moreover, a shallow feature extraction layer is designed to integrate shallow feature information, which is beneficial to enhancing the detailed features of the target and improving the detection effect of small targets. More importantly, a feature fusion module (Acode module) based on the amplitude spatial distribution difference features is proposed to quantify and encode the amplitude differences between the targets and sea clutter within the grid, and multiply the result after max pooling to maximize the amplitude differences between the target domain and sea clutter, further enhancing the network's ability to extract target feature information, achieving significant discrimination between targets and sea clutter, and thus achieving precise detection of targets.

[0071] The radar target detection method that fuses the amplitude spatial distribution difference features provided by the embodiments of the present application relates to the technical field of radar target detection. The radar target detection method that fuses the amplitude spatial distribution difference features provided by the embodiments of the present application can be applied to a terminal, a server, or software running on a terminal or a server. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a vehicle-mounted terminal, etc., but is not limited thereto; the server side can be configured as an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The server can also be a node server in a blockchain network; the software can be an application that implements the radar target detection method that fuses the amplitude spatial distribution difference features, etc., but is not limited to the above forms.

[0072] This application can be used in numerous general or specific computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multi-processor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and so on. This application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. This application can also be practiced in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0073] Figure 1 is an optional flowchart of a radar target detection method that combines amplitude spatial distribution difference features provided by an embodiment of this application. Figure 1 The method in may include but is not limited to steps S101 to S107.

[0074] Step S101, obtain a feature tensor obtained by performing feature extraction on the radar image to be detected.

[0075] Step S102, perform a max pooling operation and a min pooling operation on the input feature tensor through a convolutional kernel to obtain a max pooling tensor and a min pooling tensor.

[0076] Step S103, construct an encoding rule according to the max pooling tensor and the min pooling tensor.

[0077] Step S104, encode the feature tensor according to the encoding rule to obtain an encoded tensor.

[0078] Step S105, multiply the max pooling tensor by the encoded tensor and take the absolute value to obtain a first tensor; wherein, the first tensor characterizes the amplitude spatial distribution difference of the feature tensor.

[0079] Step S106, use the first tensor as the magnification factor to perform signal enhancement on the feature tensor and output a target feature map.

[0080] Step S107, perform target detection based on the target feature map to obtain a radar target detection result.

[0081] Steps S101 to S107 shown in the embodiments of the present application can enhance the network's ability to extract target feature information, better distinguish targets from sea clutter, and improve the accuracy of radar target detection by extracting the amplitude spatial distribution difference features from the radar image and enhancing the feature vector based on the amplitude spatial distribution difference features.

[0082] In some embodiments, step S103 may include, but is not limited to, steps S201 to S203:

[0083] Step S201, calculate the difference between the maximum pooling tensor and the minimum pooling tensor to obtain a difference tensor;

[0084] Step S202, obtain the global maximum difference and the global minimum difference from the difference tensor to obtain the global amplitude distribution interval;

[0085] Step S203, determine the coding rule based on the global amplitude distribution interval.

[0086] In some embodiments, step S104 includes, but is not limited to, the following steps S301 to S305:

[0087] Step S301, calculate the interval width according to the global amplitude distribution interval and the preset number of intervals;

[0088] Step S302, divide the global amplitude distribution interval into several coding interval ranges according to the interval width;

[0089] Step S303, multiply each difference element in the difference tensor by the interval width to obtain an intermediate value;

[0090] Step S304, use the serial number of the coding interval range to which the intermediate value belongs as the element corresponding code of the difference element;

[0091] Step S305, form a coding tensor with all the element corresponding codes.

[0092] In step S302 of some embodiments, the expression of the coding interval range is:

[0093] ;

[0094] ;

[0095] Wherein, represents the coding interval range; represents the global minimum difference; represents the serial number of the coding interval range; represents the interval width; n represents the number of coding intervals.

[0096] In step S104 of some embodiments, the expression of the encoded tensor is:

[0097] ;

[0098] where is an element of the encoded tensor; b represents the number of samples; c represents the number of channels; w represents the width; h represents the height; represents an element of the difference tensor.

[0099] In some embodiments, the method further includes the following steps S108 to S110:

[0100] Step S108, obtain the radar image to be detected;

[0101] Step S109, preprocess the radar image to be detected to obtain the first radar image;

[0102] Step S100, perform feature extraction on the radar image to be detected through the backbone network combined with channel attention to obtain the first feature map;

[0103] Step S110, perform feature fusion on the first feature map through the feature pyramid structure to obtain the feature tensor.

[0104] In some embodiments, the preprocessing in step S109 includes at least one of the following steps S401 to S403:

[0105] Step S401, perform data augmentation processing on the radar image to be detected;

[0106] Step S402, perform adaptive anchor box calculation processing on the radar image to be detected;

[0107] Step S403, perform adaptive image scaling processing on the radar image to be detected

[0108] Next, in combination with a specific application example of the radar target detection scenario, the solution of the embodiment of the present application will be introduced and described in detail:

[0109] In the embodiment of the present application, a radar target detection method that fuses the amplitude spatial distribution difference feature is provided. This method can be applied to accurately detect radar targets. Specifically, the key to the method of the embodiment of the present application lies in the amplitude spatial distribution difference feature fusion module (Acode module), and the working principle of this module is based on the feature extraction that fuses the amplitude spatial distribution difference.

[0110] The input radar image is segmented using a small window. In the small window, the amplitude variation of sea clutter is not obvious. Once the window contains the target part, there will be obvious differences in the spatial distribution of sea clutter within the window.

[0111] Furthermore, the feature extraction process based on the fusion of amplitude spatial distribution differences may include the following steps: Segment the input radar image using a small window, and use coding to represent the amplitude spatial distribution difference features. Perform max pooling and min pooling on the input feature tensor with a kernel size of 3*3 and a stride of 3 respectively. Obtain a difference tensor through the difference between the two. Establish a global amplitude distribution interval based on the maximum and minimum values in the difference tensor. Divide the distribution interval into multiple sub-intervals for encoding the values in the difference tensor to obtain the final encoded tensor.

[0112] It should be noted that amplitude difference sampling can be first performed on the pure sea clutter area using small windows of various sizes (such as 3*3, 2*2, etc.), and the amplitude difference distribution of sea clutter in the small window is fitted and analyzed in combination with the amplitude statistical characteristics of sea clutter. According to the results of the fitting analysis, the most suitable small window size is selected for segmentation. In some embodiments, the fitting effect of the amplitude difference distribution in the 3*3 window is better. Therefore, 3*3 is used as the sampling window in the introduction, and a convolutional kernel with a stride of 3 is selected for pooling.

[0113] As Figure 2 shown, taking the example in Figure 2 After representing the amplitude spatial distribution difference features using coding, determine the global maximum difference and minimum difference, and then determine the coding interval. Based on the maximum and minimum values of the 3*3 window, determine the window difference, and multiply the window difference by the coding interval to obtain the final coding value corresponding to the window. Repeat this step. In a deeper feature map, it can be seen that a contour with relatively high coding values is formed around the target in the outer region, which can effectively distinguish the outer region and the sea clutter region compared with the relatively low coding values in the sea clutter region.

[0114] Based on this, the steps of the amplitude spatial distribution difference feature fusion module may include but are not limited to the following steps S1~S4:

[0115] Step S1, perform max pooling and min pooling operations on the input feature tensor to calculate the difference between the maximum and minimum values of all 3*3 grids in the feature tensor, forming a difference tensor.

[0116] Step S2, encode the difference tensor according to the coding rule to obtain the corresponding encoded tensor.

[0117] Step S3, multiply the result of max pooling by the encoded tensor, and then calculate the absolute value of the multiplied result.

[0118] Step S4, perform data normalization on the calculation result through the BatchNorm2d function, and finally input the data into the Relu activation function for output.

[0119] Specifically, the specific execution process of the amplitude spatial distribution difference feature fusion module is as follows:

[0120] Assume that the input of the module is a four-dimensional feature tensor , whose dimensions represent the number of samples, the number of channels, the height, and the width respectively. First, perform max-pooling and min-pooling on the feature tensor with a window size of 3*3 and a stride of 3 to obtain the max-pooling tensor and the min-pooling tensor . By taking the difference between the two, obtain the difference tensor . Then define 's global maximum difference and global minimum difference, which are and respectively. Then evenly divide the range from the global maximum difference to the global minimum difference into n intervals, and the interval width of each interval is , . Then the th interval's coding interval range is:

[0121] ;

[0122] ;

[0123] For each element of the difference tensor , encode it into a new integer , where b represents the number of samples, c represents the number of channels, w represents the width, and h represents the height, to obtain the coding rule of the coding tensor as shown in the following expression:

[0124] ;

[0125] Multiply the finally obtained coding tensor by the max-pooling tensor to obtain the first tensor.

[0126] Use the coding value of the first tensor as the magnification factor to enhance the target signal existing in the original input feature tensor , highlighting the difference in amplitude between the target and the sea clutter.

[0127] Furthermore, referring to Figure 3 and Figure 4As shown, a shallow information feature extraction layer (ST layer) is constructed based on the amplitude spatial distribution difference feature fusion module. This shallow information feature extraction layer is added to the feature fusion module of the object detection network, and the object detection network can use the YOLOv5 network. In some embodiments, other versions of the YOLO network or other object detection networks can also be used. Taking the YOLOv5 network as an example, as Figure 5 shown, Figure 4 Figure Figure 4 is a detailed structure diagram of the ST layer provided in the embodiment of the present application. Its structure can be designed to include 1 convolutional layer, 3 C3 layers, 1 splicing layer, 1 upsampling layer, and 1 Acode module. The feature extraction network uses the Backbone feature extraction network, which will perform multiple downsamplings on the input image, and the downsampling multiple each time is 2. P2 is a newly added high-resolution feature layer in the YOLOv5 network. By introducing the Acode module to fuse the amplitude and spatial distribution differences, the high-resolution feature extraction is strengthened, focusing on small object detection. The newly added P2 feature layer is set after the last feature layer of the original structure. Since the resolution of the last feature layer is 80*80, the added P2 has a resolution of 160*160. An additional detection head is introduced after P2, so that the object detection network can pay more attention to the relevant features of small objects.

[0128] These newly added feature extraction layers are designed to adjust the initial depth of feature extraction. By tensor splicing the feature information extracted by the second C3 layer in the backbone feature extraction network, it is further transmitted to the newly added ST feature layer. The optimized feature fusion network can combine high-level semantic features and low-level feature information, enabling the entire network to effectively utilize global context information, extract object features, and an innovative Acode module is introduced in the ST layer to further enhance the object feature extraction ability.

[0129] Based on this, the entire object detection network consists of a preprocessing module, a backbone network, an FPN structure (feature pyramid structure), a PAN structure (path aggregation structure), and a detection head. The functions of each module and the data flow are as follows: The input image undergoes adaptive scaling, anchor box calculation, and Mosaic data augmentation through the preprocessing module, and the enhanced and normalized image is output and input into the backbone network. The backbone network extracts shallow and deep features, and generates multi-scale feature maps through convolution and C3 modules, with the output including feature maps of different resolutions. Subsequently, these feature maps enter the FPN structure, and through upsampling and splicing operations, multi-scale fusion of shallow and deep features is achieved, generating feature maps with more detailed and semantic information. The fused feature maps are transmitted to the PAN structure, and through further feature splicing and convolution modules, the feature expression ability is strengthened, especially for the key features required for object localization and classification. Finally, the feature maps that have undergone feature fusion and enhancement are input into the detection head to complete object classification and bounding box regression.

[0130] Furthermore, in the path aggregation network where the object detection network fuses the extracted features, first, the above-mentioned Acode module is used to process the feature tensor obtained from the shallow information extraction layer, and then the processed feature map is convolved and spliced with the feature map obtained from the feature pyramid structure. The output target feature map is used for subsequent processing in the detection head. The path aggregation structure integrating the above Acode module can better distinguish objects from sea clutter, enhance the object signal, and is beneficial for the detection head to accurately detect and mark objects in the target feature map.

[0131] Furthermore, as Figure 3 and Figure 6 shown, the SE attention mechanism is integrated into the backbone network of the object detection network and applied in the C3 module to build a feature attention module (C3SE module).

[0132] As Figure 7 、 Figure 8 、 Figure 9 and Figure 10 shown, these figures are the comparison diagrams of the channel feature maps using the Acode module provided in the embodiments of the present application. By comparing the channel feature maps using the Acode module, it can be seen that after using the Acode module, the features in the target area in the feature layer are more concentrated, and the surrounding clutter is also significantly suppressed after passing through the Acode module, further verifying the enhancement effect of the Acode module on object features in the object detection network.

[0133] In summary, the beneficial effects of the embodiments of the present application are:

[0134] ​​(1) The method for extracting the difference feature of the fusion amplitude spatial distribution proposed in this application quantifies and encodes the amplitude difference between the target and sea clutter in the grid, effectively captures the amplitude feature of the target in the peripheral area, and can better distinguish the target from the sea clutter.

[0135] (2) Further, a target detection network with feature fusion is proposed for this method. Based on the YOLOv5 target detection network and with the amplitude spatial distribution difference feature fusion module as the core, a shallow feature information extraction layer and a fusion attention mechanism module are introduced to maximize the amplitude difference between the target and sea clutter. Compared with the existing YOLO series models and traditional detection methods, the method of this application embodiment has a significant improvement in detection performance.

[0136] (3) The channel attention mechanism is introduced to improve the C3 module, optimizing the data information extraction ability between channels and effectively improving the detection accuracy.

[0137] (4) The ST shallow feature extraction layer is designed to integrate shallow feature information through a dedicated feature extraction structure, enhance the detailed features of the target, and improve the detection effect on small targets.

[0138] (5) A feature fusion module (Acode module) for the dedicated amplitude spatial distribution difference feature is designed to quantify and encode the amplitude difference between the target and sea clutter in the grid, and multiply it with the maximum pooling result to maximize the amplitude difference between the target and sea clutter, further enhancing the network's ability to extract target feature information.

[0139] Please refer to Figure 11 , the embodiment of this application also provides a radar target detection system integrating the amplitude spatial distribution difference feature, which can implement the above-mentioned radar target detection method integrating the amplitude spatial distribution difference feature. The system includes:

[0140] The first module 1101 is used to obtain the feature tensor obtained by extracting features from the radar image to be detected;

[0141] The second module 1102 is used to perform maximum pooling operation and minimum pooling operation on the input feature tensor through a convolution kernel to obtain a maximum pooling tensor and a minimum pooling tensor;

[0142] The third module 1103 is used to construct an encoding rule according to the maximum pooling tensor and the minimum pooling tensor;

[0143] The fourth module 1104 is used to encode the feature tensor according to the encoding rule to obtain an encoded tensor;

[0144] The fifth module 1105 is configured to multiply the maximum pooling tensor by the encoded tensor and take the absolute value to obtain a first tensor, where the first tensor represents the amplitude spatial distribution difference of the feature tensor.

[0145] The sixth module 1106 is configured to use the first tensor as a magnification factor to perform signal enhancement on the feature tensor and output a target feature map.

[0146] The seventh module 1107 is configured to perform target detection based on the target feature map to obtain a radar target detection result.

[0147] In some embodiments, the system further includes the following modules:

[0148] The eighth module is configured to preprocess the radar image to be detected to obtain a first radar image.

[0149] The ninth module is configured to perform feature extraction on the radar image to be detected through a backbone network that combines channel attention to obtain a first feature map.

[0150] The tenth module is configured to perform feature fusion on the first feature map through a feature pyramid structure to obtain a feature tensor.

[0151] It can be understood that the content in the above method embodiments is applicable to the system embodiments. The functions specifically implemented by the system embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0152] An embodiment of the present application further provides an electronic device. The electronic device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the above-mentioned radar target detection method for fusing amplitude spatial distribution difference features. The electronic device can be any intelligent terminal including a tablet computer, an in-vehicle computer, etc.

[0153] It can be understood that the content in the above method embodiments is applicable to the device embodiments. The functions specifically implemented by the device embodiments are the same as those in the above method embodiments, and the beneficial effects achieved are also the same as those in the above method embodiments.

[0154] Please refer to Figure 12 , Figure 12 which illustrates the hardware structure of an electronic device in another embodiment. The electronic device includes:

[0155] The processor 1201 can be implemented in the form of a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, etc., and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0156] The memory 1202 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM), etc. The memory 1202 can store an operating system and other application programs. When implementing the technical solutions provided in the embodiments of this specification through software or firmware, the relevant program codes are stored in the memory 1202 and are called by the processor 1201 to execute the radar target detection method for fusing amplitude spatial distribution difference features in the embodiments of the present application;

[0157] The input / output interface 1203 is used to implement information input and output;

[0158] The communication interface 1204 is used to implement communication interaction between this device and other devices. Communication can be achieved through wired means (such as USB, network cable, etc.) or through wireless means (such as mobile network, WIFI, Bluetooth, etc.);

[0159] The bus 1205 transmits information between various components of the device (such as the processor 1201, the memory 1202, the input / output interface 1203, and the communication interface 1204);

[0160] Among them, the processor 1201, the memory 1202, the input / output interface 1203, and the communication interface 1204 achieve communication connections with each other inside the device through the bus 1205.

[0161] The embodiments of the present application also provide a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, it implements the above-mentioned radar target detection method for fusing amplitude spatial distribution difference features.

[0162] It can be understood that the content in the above method embodiments is applicable to the embodiments of this storage medium. The functions specifically implemented by the embodiments of this storage medium are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those of the above method embodiments.

[0163] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory may optionally include a memory remotely located relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0164] The embodiments described in the embodiments of the present application are for more clearly illustrating the technical solutions of the embodiments of the present application, and do not constitute a limitation on the technical solutions provided by the embodiments of the present application. Those skilled in the art will know that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided by the embodiments of the present application are equally applicable to similar technical problems.

[0165] Those skilled in the art can understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than those shown in the figures, or combine certain steps, or different steps.

[0166] The system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, that is, they may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0167] Those of ordinary skill in the art can understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices can be implemented as software, firmware, hardware, and appropriate combinations thereof.

[0168] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products, or devices.

[0169] It should be understood that in this application, "at least one (item)" means one or more, and "a plurality" means two or more. "And / or" is used to describe the association relationship of associated objects and indicates that three relationships can exist. For example, "A and / or B" can mean: only A exists, only B exists, and both A and B exist at the same time. Among them, A and B can be singular or plural. The character " / " generally indicates that the associated objects before and after are in an "or" relationship. "At least one (item) of the following" or its similar expressions refer to any combination of these items, including any combination of single items (items) or plural items (items). For example, at least one (item) of a, b, or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0170] In several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the system embodiments described above are only illustrative. For example, the above division of units is only a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed couplings or direct couplings or communication connections to each other can be through some interfaces. The indirect couplings or communication connections of devices or units can be in electrical, mechanical or other forms.

[0171] The units described above as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, in each embodiment of this application, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above integrated units can be implemented in the form of hardware or in the form of software functional units.

[0173] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes multiple instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods of various embodiments of this application. The foregoing storage medium includes: various media that can store programs such as USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), magnetic disks, or optical discs.

[0174] The preferred embodiments of the embodiments of this application have been described above with reference to the accompanying drawings, which does not limit the scope of rights of the embodiments of this application. Any modifications, equivalent replacements, and improvements made by those skilled in the art without departing from the scope and essence of the embodiments of this application shall be within the scope of rights of the embodiments of this application.

Claims

1. A radar target detection method that fuses the difference features of the amplitude spatial distribution, characterized in that It includes the following steps: Obtain the feature tensor obtained by performing feature extraction on the radar image to be detected; Perform max-pooling operation and min-pooling operation on the input feature tensor through a convolutional kernel to obtain a max-pooling tensor and a min-pooling tensor; Construct an encoding rule according to the max-pooling tensor and the min-pooling tensor; Encode the feature tensor according to the encoding rule to obtain an encoded tensor; Multiply the max-pooling tensor by the encoded tensor and take the absolute value to obtain a first tensor; wherein, the first tensor characterizes the amplitude spatial distribution difference of the feature tensor; Use the first tensor as the magnification factor to perform signal enhancement on the feature tensor and output a target feature map; Perform target detection based on the target feature map to obtain a radar target detection result.

2. The method according to claim 1, wherein The constructing an encoding rule according to the max-pooling tensor and the min-pooling tensor includes the following steps: Calculate the difference between the max-pooling tensor and the min-pooling tensor to obtain a difference tensor; Obtain the global maximum difference and the global minimum difference from the difference tensor to obtain a global amplitude distribution interval; Determine an encoding rule based on the global amplitude distribution interval.

3. The method according to claim 2, wherein The encoding the feature tensor according to the encoding rule to obtain an encoded tensor includes the following steps: Calculate the interval width according to the global amplitude distribution interval and the preset number of intervals; Divide the global amplitude distribution interval into several encoding interval ranges according to the interval width; Multiply each difference element in the difference tensor by the interval width to obtain an intermediate value; Use the serial number of the encoding interval range to which the intermediate value belongs as the element corresponding encoding of the difference element; Form an encoded tensor by all the element corresponding encodings.

4. The method according to claim 3, wherein The expression of the encoding interval range is: ； ； Among them, represents the coding interval range; represents the global minimum difference; k represents the serial number of the coding interval range; represents the interval width; n represents the number of coding intervals.

5. The method according to claim 1, wherein The expression of the encoded tensor is: ； Among them, is the element of the encoded tensor; b represents the number of samples; c represents the number of channels; w represents the width; h represents the height; represents the element of the difference tensor.

6. The method according to claim 1, wherein The method further includes the following steps: Obtain the radar image to be detected; Preprocess the radar image to be detected to obtain a first radar image; Perform feature extraction on the radar image to be detected through a backbone network combined with channel attention to obtain a first feature map; Perform feature fusion on the first feature map through a feature pyramid structure to obtain a feature tensor.

7. The method according to claim 6, wherein In the step of preprocessing the radar image to be detected to obtain a first radar image, the preprocessing steps include at least one of the following steps: Perform data enhancement processing on the radar image to be detected; Perform adaptive anchor box calculation processing on the radar image to be detected; Perform adaptive image scaling processing on the radar image to be detected.

8. A radar target detection system integrating the difference features of the amplitude spatial distribution, characterized in that It includes: A first module for obtaining the feature tensor obtained by performing feature extraction on the radar image to be detected; A second module for performing max-pooling operation and min-pooling operation on the input feature tensor through a convolutional kernel to obtain a max-pooling tensor and a min-pooling tensor; A third module for constructing an encoding rule according to the max-pooling tensor and the min-pooling tensor; A fourth module for encoding the feature tensor according to the encoding rule to obtain an encoded tensor; The fifth module is configured to multiply the max-pooling tensor and the encoded tensor and take the absolute value to obtain a first tensor; wherein, the first tensor characterizes the amplitude spatial distribution difference of the feature tensor. The sixth module is configured to use the first tensor as a magnification factor to perform signal enhancement on the feature tensor and output a target feature map.

9. An electronic device, characterized in that, It includes a processor and a memory. The memory is used to store programs. The processor executes the program to implement the method according to any one of claims 1 to 7.

10. A computer storage medium storing a program executable by a processor, characterized in that, The program executable by the processor, when executed by the processor, is used to implement the method according to any one of claims 1 to 7.

Images