A method for detecting dynamic small targets on the sea surface using airborne radar based on airspace perception

Abstract

The invention discloses a method for detecting dynamic small targets on the sea surface by an airborne radar based on airspace perception, and relates to the technical field of signal processing. First, data acquisition and preprocessing are performed; original sea surface image data is acquired by an airborne radar; then the acquired original sea surface image data is subjected to denoising and image normalization processing; then an energy enhancement module is constructed to enhance the signal strength of small targets; finally, an airspace perception backbone network module is constructed, and based on the scattering difference between the target area and the background area, the target existence probability of each pixel point is obtained, and small targets are identified. The method for detecting dynamic small targets on the sea surface by an airborne radar based on airspace perception provided by the invention realizes the accurate detection of dynamic small targets on the sea surface, and especially shows higher detection accuracy under complex backgrounds.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a method for detecting dynamic small targets on the sea surface using an airborne radar based on airspace perception. Background Art

[0002] Small target detection on the sea surface refers to identifying and locating targets with small size, weak reflection signal and slow movement speed in radar echo signals. These targets usually have low radar cross-section (RCS), so their echo signals are easily confused with background noise and clutter, resulting in detection difficulties. Typical small targets on the sea surface include submarine periscopes, small ships, frogmen, floating ice, etc. These targets are often not easy to be accurately identified in traditional radar systems. Compared with large targets such as cargo ships and aircraft carriers, small targets are much more difficult to detect, especially in the dynamic environment of the sea surface, where the difference between targets and clutter is small, resulting in false detection or missed detection by the radar system during detection. The detection of small targets on the sea surface has important application significance. In terms of marine ecological protection, early identification of illegal fishing activities or ecological destruction can effectively maintain marine biodiversity and promote sustainable development. At the same time, in the field of shipping, timely detection of small ships can help ensure navigation safety, reduce collision risks, and improve the efficiency of maritime transportation. In addition, this detection technology can also be applied to fields such as fishery management, marine tourism and environmental monitoring, helping relevant departments to better manage resources and protect the environment. Therefore, developing relevant technologies to improve the detection capability of small targets on the sea surface is an important step to promote the sustainable development and protection of the ocean. However, the detection of small targets on the sea surface faces a series of challenges in practical applications. First, the sea surface environment itself is extremely complex, and the background clutter and sea surface reflection signals are very strong. Especially under conditions such as waves and weather changes, the echo of small targets is easily masked by strong sea surface clutter. Secondly, small targets on the sea surface usually have a slower movement speed and lower reflection intensity, which makes their echo signals easily overlap with sea surface clutter in the Doppler frequency domain, causing signal aliasing. Traditional radar systems mostly rely on signal intensity differences and Doppler filtering technology to detect targets, but for small targets, their signals are too weak, and the frequency characteristics of the target's echo signal and sea clutter are similar, making it difficult to effectively distinguish them from background noise. Traditional methods often fail to meet the needs of accurate detection. At present, researchers have proposed a variety of airborne radar sea surface small target detection methods, such as the Chinese patent with publication number CN114943888B, entitled Sea surface small target detection method based on multi-scale information fusion, which uses a deep learning model to detect sea surface small targets based on multi-scale information fusion. By constructing a simulation data set of small sea surface targets and training the model, the Transformer module is used to decode and detect the target position and category during the detection process.The innovative use of top-down Transformer decoding strategy, combined with deformable convolution and self-attention mechanism, continuously corrects the bounding box by introducing low-level features, significantly improving the accuracy of small target detection; for example, in the Chinese patent with publication number CN105894033B and the name of a weak target detection method and system under sea clutter background, it is proposed to use known sea clutter signals as training data, extract and fuse at least two feature vectors to form a joint feature vector, and train the detection system to achieve an ideal detection accuracy on known samples; although the technical means proposed in these studies can effectively improve the detection effect of weak and small targets, there are still problems such as insufficient information utilization and failure to consider the relationship between the target and the background. Summary of the invention

[0003] The purpose of the present invention is to provide an airborne radar sea surface dynamic weak small target detection method based on airspace perception, so as to solve the problem that the existing methods in the prior art have insufficient information utilization when detecting weak small targets and do not consider the relationship between the target and the background.

[0004] To achieve the above object, the present invention provides a method for detecting dynamic small targets on the sea surface using an airborne radar based on airspace perception, comprising the following steps:

[0005] Step 1, data acquisition and preprocessing; obtaining original sea surface image data through airborne radar; then subjecting the obtained original sea surface image data to denoising and image normalization processing; wherein the original sea surface image contains sea surface echo, environmental noise and target echo;

[0006] Step 2: construct an energy enhancement module to enhance the signal strength of weak targets; wherein the energy enhancement module includes an adaptive signal gain layer and a nonlinear transformation layer;

[0007] Step 3: Construct a spatial perception backbone network module. Based on the scattering difference between the target area and the background area, the target existence probability of each pixel is obtained to identify weak targets. The spatial perception backbone network module includes a multi-scale convolutional network, a spatial context perception network and a self-attention mechanism layer.

[0008] Preferably, the construction process of the adaptive signal gain layer is as follows:

[0009] S21. Design gain function , the gain factor is dynamically adjusted according to the ratio of the signal strength of the target area to the background noise. The expression is as follows:

[0010] ;

[0011] In the formula, Indicates the signal intensity of the target area in the image, represents the noise intensity in the background area, is the gain factor, which adjusts the gain size;

[0012] S22, based on the local contrast of the image Adaptive adjustment of gain factor , the expression is as follows:

[0013] ;

[0014] S23, estimate the local contrast of the image based on the standard deviation within the pixel neighborhood , the expression is as follows:

[0015] ;

[0016] In the formula, Is a neighborhood The mean of is the pixel value in the neighborhood.

[0017] Preferably, the expression of the nonlinear transformation layer is as follows:

[0018] ;

[0019] ;

[0020] In the formula, is the pixel value of the original image, For the enhanced image, is the Gamma coefficient, select To enhance dark targets, is the wavelet basis function, represents the high-frequency information after wavelet decomposition, Represents the enhanced image.

[0021] Preferably, the calculation expression of the multi-scale convolution of the multi-scale convolutional network is as follows:

[0022] ;

[0023] In the formula, Represents the result of energy enhancement of the original input image. Represents a multi-scale convolutional layer, which is used to extract feature information at different scales. Represents the output of a multi-scale convolutional network.

[0024] Preferably, the spatial context-aware network is specifically based on the multi-scale convolutional network, and a residual network (ResNet) module is added. The calculation expression is as follows:

[0025] ;

[0026] In the formula, is the ReLU activation function, is the output of the residual module, Represents the output of a multi-scale convolutional network.

[0027] Preferably, the calculation expression of the self-attention mechanism layer is as follows:

[0028] ;

[0029] In the formula, is the query vector, is the key vector, is a value vector, is the vector dimension.

[0030] Preferably, the calculation expression of the target existence probability is as follows:

[0031] ;

[0032] in, Indicates location The probability of the target existing at is the activation function, and Cbackbone is the output of the backbone network.

[0033] Therefore, the present invention adopts the above-mentioned airborne radar sea surface dynamic weak target detection method based on airspace perception, which has the following beneficial effects:

[0034] (1) The constructed energy enhancement module significantly improves the visibility of small targets in radar images through multi-level signal processing. Specifically, by performing signal gain and nonlinear transformation on the input image, the dynamic range of the target signal is effectively expanded, so that the signal of small targets is significantly enhanced. At the same time, by decomposing the image into low-frequency and high-frequency components, important detail information in the image is extracted, the influence of noise is effectively eliminated, and the target features are highlighted.

[0035] (2) The constructed spatial perception backbone network module effectively improves the method's ability to perceive the scattering difference between the target and the background. Specifically, through the multi-scale convolutional network, the module can extract target features at different scales, thereby better capturing targets of different sizes. The introduction of the residual network further deepens the learning ability of the model, avoids information loss, and enhances the spatial feature representation of the target. The self-attention mechanism can accurately model the long-range dependency between the target and the background in the image, thereby more accurately locating weak targets, especially in complex sea surfaces and dynamic backgrounds. This mechanism adaptively calculates the attention weights, focuses on the target area in the image, suppresses background noise, and significantly improves the accuracy of target detection.

[0036] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 The present invention is an overall flow chart of a method for detecting dynamic small targets on the sea surface using an airborne radar based on airspace perception;

[0038] Figure 2 This is a structural diagram of the airspace awareness backbone network module of an embodiment of the present invention. DETAILED DESCRIPTION

[0039] The following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] See also Figure 1-Figure 2 , a method for detecting dynamic small targets on the sea surface using an airborne radar based on airspace perception, comprising the following steps:

[0041] Step 1, data acquisition and preprocessing; obtaining original sea surface image data through airborne radar; then subjecting the obtained original sea surface image data to denoising and image normalization processing; wherein the original sea surface image contains sea surface echo, environmental noise and target echo;

[0042] Step 2: Construct an energy enhancement module to enhance the signal strength of small targets; by enhancing the energy of the target area, the signal strength of small targets is enhanced, noise interference is reduced, and a clearer input is provided for the subsequent target detection module; wherein the energy enhancement module includes an adaptive signal gain layer and a nonlinear transformation layer;

[0043] The construction process of the adaptive signal gain layer is as follows:

[0044] S21. Design gain function , the gain factor is dynamically adjusted according to the ratio of the signal strength of the target area to the background noise. The expression is as follows:

[0045] ;

[0046] In the formula, Indicates the signal intensity of the target area in the image, represents the noise intensity in the background area, is the gain factor, which adjusts the gain size;

[0047] S22, based on the local contrast of the image Adaptive adjustment of gain factor , the expression is as follows:

[0048] ;

[0049] S23, estimate the local contrast of the image based on the standard deviation within the pixel neighborhood , the expression is as follows:

[0050] ;

[0051] In the formula, Is a neighborhood The mean of is the pixel value in the neighborhood.

[0052] In order to further enhance the weak and small targets, nonlinear transformation is used to adjust the image, which can effectively process the low signal part in the image and avoid the processing problem caused by too small value: For dark targets, this method can further enhance the low brightness area of ​​the image: In addition, this method can effectively extract high-frequency detail information in the image, especially for the edge features of weak and small targets. The expression of the nonlinear transformation layer is as follows:

[0053] ;

[0054] ;

[0055] In the formula, is the pixel value of the original image, For the enhanced image, is the Gamma coefficient, select To enhance dark targets, is the wavelet basis function, represents the high-frequency information after wavelet decomposition, Represents the enhanced image.

[0056] After nonlinear transformation, the image is decomposed into detail images and approximate images at multiple scales. The image is then decomposed using multiscale wavelet decomposition to extract high-frequency information, especially the details in the target area. The high-frequency information is amplified and the image is reconstructed.

[0057] Step 3: Construct a spatial perception backbone network module. Based on the scattering difference between the target area and the background area, the target existence probability of each pixel is obtained to identify weak targets. The spatial perception backbone network module includes a multi-scale convolutional network, a spatial context perception network, and a self-attention mechanism layer. The specific relationship between the three is as follows: First, the multi-scale convolutional network is used to extract image features at different scales. Through multiple convolutional layers and pooling layers, the network can effectively extract low-level and high-level spatial features and preliminarily locate the spatial position of the target. Then, on the basis of the multi-scale convolutional network, a residual network (ResNet) module is added to enhance the network's deep learning ability and avoid information loss. The residual block can help the network learn more complex spatial features, especially when dealing with complex backgrounds and weak targets, which can improve the accuracy of the network; further, by introducing the self-attention mechanism, the modeling ability of dependencies and contextual information is enhanced; finally, by combining the output of the multi-scale convolutional network, the spatial context-aware network and the self-attention mechanism layer, the spatial features of the target and the background are fused, and the probability of the target existence at each pixel is finally obtained; the calculation expression of the multi-scale convolution of the multi-scale convolutional network is as follows:

[0058] ;

[0059] In the formula, Represents the result of energy enhancement of the original input image. Represents a multi-scale convolutional layer, which is used to extract feature information at different scales. Represents the output of a multi-scale convolutional network.

[0060] The spatial context-aware network is specifically based on the multi-scale convolutional network, with a residual network module added. The calculation expression is as follows:

[0061] ;

[0062] In the formula, is the ReLU activation function, is the output of the residual module, Represents the output of a multi-scale convolutional network.

[0063] The calculation expression of the self-attention mechanism layer is as follows:

[0064] ;

[0065] In the formula, is the query vector, is the key vector, is a value vector, is the vector dimension.

[0066] The calculation expression of the target existence probability is as follows:

[0067] ;

[0068] in, Indicates location The probability of the target existing at is the activation function, and Cbackbone is the output of the backbone network.

[0069] Therefore, the present invention adopts the above-mentioned airborne radar sea surface dynamic weak target detection method based on airspace perception. Through the synergistic effect of the energy enhancement module and the airspace perception backbone network module, it can accurately detect weak targets under dynamic sea surface and low signal-to-noise ratio conditions, providing technical support for sea surface monitoring and related applications.

[0070] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for detecting dynamic small targets on the sea surface using airborne radar based on airspace perception, characterized in that: The following steps are involved: Step 1, data acquisition and preprocessing; obtaining original sea surface image data through airborne radar; then subjecting the obtained original sea surface image data to denoising and image normalization processing; wherein the original sea surface image contains sea surface echo, environmental noise and target echo; Step 2: construct an energy enhancement module to enhance the signal strength of weak targets; wherein the energy enhancement module includes an adaptive signal gain layer and a nonlinear transformation layer; Step 3: Construct a spatial perception backbone network module to obtain the target existence probability of each pixel based on the scattering difference between the target area and the background area, and identify weak targets; the spatial perception backbone network module includes a multi-scale convolutional network, a spatial context perception network, and a self-attention mechanism layer; The calculation expression of the target existence probability is as follows: ； in, Indicates location The probability of the target existing at is the activation function, Cbackbone is the output of the backbone network; The calculation expression of multi-scale convolution of multi-scale convolutional network is as follows: ； In the formula, Represents the result of energy enhancement of the original input image. Represents a multi-scale convolutional layer, which is used to extract feature information at different scales. Represents the output of a multi-scale convolutional network; The spatial context-aware network is specifically based on the multi-scale convolutional network, with a residual network module added. The calculation expression is as follows: ； In the formula, is the ReLU activation function, is the output of the residual module, Represents the output of a multi-scale convolutional network; The calculation expression of the self-attention mechanism layer is as follows: ； In the formula, is the query vector, is the key vector, is a value vector, is the vector dimension; The expression of the nonlinear transformation layer is as follows: ； ； In the formula, is the pixel value of the original image, For the enhanced image, is the Gamma value coefficient. Select γ<1 to enhance the dark target. is the wavelet basis function, represents the high-frequency information after wavelet decomposition, Represents the enhanced image.

2. According to the method for detecting dynamic small targets on the sea surface based on airborne radar based on airspace perception in claim 1, it is characterized in that: The construction process of the adaptive signal gain layer is as follows: S21. Design gain function , the gain factor is dynamically adjusted according to the ratio of the signal strength of the target area to the background noise. The expression is as follows: ； In the formula, Indicates the signal intensity of the target area in the image, represents the noise intensity in the background area, is the gain factor, which adjusts the gain size; S22, based on the local contrast of the image Adaptive adjustment of gain factor , the expression is as follows: ； S23, estimate the local contrast of the image based on the standard deviation within the pixel neighborhood , the expression is as follows: ； In the formula, Is a neighborhood The mean of is the pixel value in the neighborhood.

Images