Method and device for detecting target in synthetic aperture radar image against complex interference

Abstract

The application provides a synthetic aperture radar image target detection method and device against complex interference, and relates to the technical field of remote sensing. The method comprises the following steps: acquiring a synthetic aperture radar (SAR) image to be detected; performing adaptive amplitude mapping on the SAR image to obtain a dark background radiation enhancement channel and a bright target radiation enhancement channel corresponding to the SAR image; performing coherent speckle noise suppression on the SAR image to obtain a coherent speckle noise suppression channel corresponding to the SAR image; performing channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel and the coherent speckle noise suppression channel to obtain a pseudo-color SAR image, and performing target detection based on the pseudo-color SAR image. By using the technical scheme provided in the application, the coherent speckle noise can be suppressed while the amplitude distribution of the SAR image is stabilized, so that the robust generalization capability of target detection is improved.

Description

Technical Field

[0001] This application relates to the field of remote sensing technology, and in particular to a method and apparatus for target detection in synthetic aperture radar images that is resistant to complex interference. Background Technology

[0002] Synthetic Aperture Radar (SAR) acquires high-resolution images by actively emitting electromagnetic waves and can detect targets obscured by clouds, fog, or even sand and gravel. Target detection relies on the radiometric and visual features contained in SAR images, which play a crucial role in distinguishing targets from the background.

[0003] However, the amplitude distribution of SAR images fluctuates due to changes in scene and imaging conditions, resulting in unstable target radiation characteristics. In addition, speckle noise, which is prevalent in SAR images, damages the target's structure, texture, and other features, causing a decrease in visual feature quality. Ultimately, this leads to an increase in the variance of the target's intra-class feature distribution and a decrease in the generalization performance of the target detection model.

[0004] Therefore, when performing target detection based on SAR images, how to suppress speckle noise while stabilizing the amplitude distribution of SAR images and improve the robust generalization ability of target detection is a technical problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] This application provides a synthetic aperture radar image target detection method and apparatus that resists complex interference. It can suppress speckle noise while stabilizing the amplitude distribution of SAR images, thereby improving the robust generalization capability of target detection.

[0006] This application provides a synthetic aperture radar image target detection method resistant to complex interference, including:

[0007] Acquire synthetic aperture radar (SAR) images of the target object;

[0008] Adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image;

[0009] The SAR image is subjected to speckle noise suppression to obtain the speckle noise suppression channel corresponding to the SAR image;

[0010] The dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel are fused to obtain a pseudo-color SAR image, and target detection is performed based on the pseudo-color SAR image.

[0011] According to the synthetic aperture radar (SAR) image target detection method for resisting complex interference provided in this application, the step of adaptively mapping the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image includes:

[0012] Based on the amplitude distribution probability density distribution model of the SAR image, an amplitude mapping function corresponding to the SAR image is constructed;

[0013] Using the amplitude mapping function, and based on the amplitude value of the target that appears most frequently in the SAR image, the amplitude boundary point for long trailing truncation, the amplitude mapping boundary point for dark background enhancement channel, and the amplitude mapping boundary point for bright target enhancement channel are determined respectively.

[0014] Based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel.

[0015] According to the synthetic aperture radar (SAR) image target detection method for resisting complex interference provided in this application, the adaptive amplitude mapping of the SAR image based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel, includes:

[0016] The amplitude of the SAR image that is greater than the long trailing truncation amplitude boundary is set as the long trailing truncation amplitude boundary, thus obtaining the image amplitude range.

[0017] The amplitude values ​​within the image amplitude range that are smaller than the boundary point of the dark background enhancement channel amplitude mapping are mapped to a first amplitude interval, and the amplitude values ​​that are greater than or equal to the boundary point of the dark background enhancement channel amplitude mapping are mapped to a second amplitude interval, thus obtaining the dark background radiation enhancement channel; wherein, the amplitude value in the second amplitude interval is greater than the amplitude value in the first amplitude interval;

[0018] The amplitudes within the image amplitude range that are smaller than the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the first amplitude range, and the amplitudes that are greater than or equal to the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the second amplitude range, thus obtaining the brightness target radiation enhancement channel.

[0019] According to the synthetic aperture radar (SAR) image target detection method for resisting complex interference provided in this application, the step of performing speckle noise suppression on the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image includes:

[0020] The SAR image is homomorphically transformed to obtain an additively modeled image;

[0021] Multiple Bernoulli samples are performed on the additive modeling image to obtain multiple randomly sampled images;

[0022] For each of the randomly sampled images, the randomly sampled images are input into a pre-trained speckle noise suppression network, and speckle noise is suppressed on the randomly sampled images through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image;

[0023] Based on multiple reconstructed noise-suppressed SAR images, the speckle noise suppression channel is determined.

[0024] According to the synthetic aperture radar image target detection method for resisting complex interference provided in this application, the method involves inputting the randomly sampled image into a pre-trained speckle noise suppression network, and performing speckle noise suppression on the randomly sampled image through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image, comprising:

[0025] The randomly sampled image is input into the block processing module in the speckle noise suppression network, and the block processing module performs block processing on the randomly sampled image to obtain multiple image blocks.

[0026] The multiple image patches are input into the feature embedding module in the speckle noise suppression network, and the feature embedding module performs feature embedding on the multiple image patches to obtain an embedded feature space vector;

[0027] The embedded feature space vector is input into the feature encoding module in the speckle noise suppression network. The embedded feature space vector is encoded by the feature encoding module, and speckle noise suppression is performed during the encoding process to obtain a deep semantic feature vector.

[0028] The deep semantic feature vector is input into the feature decoding module in the speckle noise suppression network. The feature decoding module decodes the deep semantic feature vector to obtain the reconstructed noise-suppressed SAR image.

[0029] According to the synthetic aperture radar image target detection method for resisting complex interference provided in this application, the training process of the speckle noise suppression network includes:

[0030] Modeling SAR image samples based on ideal noise-free images and multiplicative speckle noise;

[0031] The SAR image samples are homomorphically transformed to obtain additive modeling image samples; and the additive modeling image samples are Bernoulli sampled to obtain random sampled image samples.

[0032] The randomly sampled image samples are input into an initial speckle noise suppression network, and speckle noise is suppressed on the randomly sampled image samples through the initial speckle noise suppression network to obtain reconstructed noise-suppressed SAR image samples;

[0033] Based on the loss between the additively modeled image samples and the reconstructed noise-suppressed SAR image samples, the backpropagation algorithm iteratively prioritizes the model parameters in the initial speckle noise suppression network to train the speckle noise suppression network.

[0034] According to the synthetic aperture radar image target detection method for resisting complex interference provided in this application, the step of determining the speckle noise suppression channel based on multiple reconstructed noise-suppressed SAR images includes:

[0035] Determine the average SAR image corresponding to the plurality of reconstructed noise-suppressed SAR images;

[0036] The average SAR image is subjected to inverse homomorphic transformation to obtain the speckle noise suppression channel.

[0037] According to the synthetic aperture radar image target detection method for resisting complex interference provided in this application, the target detection based on the pseudo-color SAR image includes:

[0038] The pseudo-color SAR image is input into the target detection model to obtain the target detection result corresponding to the SAR image;

[0039] The target detection model is trained based on the pseudo-color SAR image samples corresponding to each of the multiple SAR image samples, and the target detection result labels corresponding to each of the SAR image samples.

[0040] This application also provides a synthetic aperture radar image target detection device resistant to complex interference, comprising:

[0041] The acquisition unit is used to acquire the synthetic aperture radar (SAR) image to be detected.

[0042] The mapping unit is used to perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image.

[0043] The suppression unit is used to suppress speckle noise in the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image;

[0044] The fusion detection unit is used to perform channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and to perform target detection based on the pseudo-color SAR image.

[0045] This application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the synthetic aperture radar image target detection method against complex interference as described in any of the preceding claims.

[0046] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the synthetic aperture radar image target detection method against complex interference as described in any of the preceding claims.

[0047] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the synthetic aperture radar image target detection method against complex interference as described in any of the preceding claims.

[0048] The synthetic aperture radar (SAR) image target detection method and apparatus provided in this application, which are resistant to complex interference, performs adaptive amplitude mapping on the SAR image to be detected when performing target detection based on SAR images, to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image; performs speckle noise suppression on the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image; and then performs channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and performs target detection based on the pseudo-color SAR image. By adaptively mapping the amplitude of the SAR image, we obtain the corresponding dark background radiation enhancement channel and bright target radiation enhancement channel. This effectively solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, thus stabilizing the amplitude distribution of the SAR image. Furthermore, by suppressing speckle noise in the SAR image, we obtain the corresponding speckle noise suppression channel. This effectively solves the problem of visual feature quality degradation caused by speckle noise in the SAR image damaging target structure, texture, and other features. By suppressing speckle noise, we improve the visual feature quality. Finally, by combining the pseudo-color SAR image obtained by fusing the dark background radiation enhancement channel, bright target radiation enhancement channel, and speckle noise suppression channel, we can effectively improve the robust generalization ability of target detection. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of a framework for target detection based on SAR images, provided in an embodiment of this application.

[0051] Figure 2 This is a flowchart illustrating a synthetic aperture radar image target detection method for resisting complex interference, provided in an embodiment of this application.

[0052] Figure 3 This is a schematic diagram of an adaptive amplitude mapping process for SAR images provided in an embodiment of this application.

[0053] Figure 4 This is a schematic diagram of a dark background radiation enhancement channel and a bright target radiation enhancement channel obtained through adaptive amplitude mapping, as provided in an embodiment of this application.

[0054] Figure 5 This is a schematic diagram of a process for suppressing speckle noise in SAR images, provided as an embodiment of this application.

[0055] Figure 6 This is a schematic diagram of the framework of a speckle noise suppression network provided in an embodiment of this application.

[0056] Figure 7 This is a schematic diagram comparing a SAR image to be detected and a speckle noise suppression channel, provided as an embodiment of this application.

[0057] Figure 8 This is a schematic diagram comparing a SAR image to be detected and a pseudo-color SAR image provided in an embodiment of this application.

[0058] Figure 9 This is a schematic diagram illustrating the target detection effect provided in an embodiment of this application.

[0059] Figure 10 This is a schematic diagram of a synthetic aperture radar image target detection device that resists complex interference, provided as an embodiment of this application.

[0060] Figure 11 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] In the embodiments of this application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone, where A and B can be singular or plural. In the textual description of this application, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0063] The technical solutions provided in this application can be applied to scenarios such as marine monitoring, disaster monitoring and assessment, agriculture, and forestry. For example, SAR images can effectively detect ship targets in the ocean, identify and track vessels at sea, and improve marine resource management and security capabilities. SAR images also have important applications in the monitoring and assessment of natural disasters (such as floods, earthquakes, and landslides). Through target detection methods, the damage to critical infrastructure (such as bridges and roads) in disaster-stricken areas can be quickly identified, providing data support for post-disaster relief. Furthermore, SAR images can be used for crop classification, forest resource monitoring, and pest and disease detection. By analyzing target features in SAR images, vegetation health can be assessed or illegal logging can be monitored.

[0064] Currently, when performing target detection based on SAR images, to address the issue of fluctuating amplitude distribution in SAR images due to changes in scene and imaging conditions, leading to unstable target radiometric features, quantification or percentage enhancement methods are commonly used. Quantification, which emphasizes background regions with lower amplitude values, often results in insufficient coding bits for the target region, weakening the target's radiometric features. Percentage enhancement, on the other hand, uses a fixed percentage to truncate the amplitude distribution, failing to adaptively select the truncation threshold, thus causing amplitude distribution instability and increased variance in the intra-class distribution of radiometric features between the target and the background.

[0065] To address the problem of speckle noise in SAR images damaging target structure, texture, and other features, thus degrading visual quality, speckle noise suppression algorithms are typically employed. Most existing speckle noise suppression algorithms are based on assumptions about the probability density function of the ideal SAR amplitude distribution or ergodicity; however, these assumptions are not strictly valid, limiting the performance of speckle noise suppression. Furthermore, since ideal SAR amplitude images unaffected by speckle noise are unavailable, supervised algorithms are also difficult to apply to speckle noise suppression.

[0066] It is evident that existing algorithms cannot adequately address the instability of target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, nor can they effectively address the issue of visual feature quality degradation caused by speckle noise in SAR images damaging target structure, texture, and other features. Consequently, they fail to improve the robust generalization capability of target detection.

[0067] To suppress speckle noise while stabilizing the amplitude distribution of SAR images and improve the robust generalization capability of target detection, embodiments of this application provide a synthetic aperture radar image target detection method resistant to complex interference. For example, see [link to relevant documentation]. Figure 1 As shown, Figure 1 This is a schematic diagram of a framework for target detection based on SAR images provided in an embodiment of this application. First, a synthetic aperture radar (SAR) image to be detected is acquired; then, adaptive amplitude mapping is performed on the SAR image to obtain a dark background radiation enhancement channel and a bright target radiation enhancement channel corresponding to the SAR image; speckle noise suppression is performed on the SAR image to obtain a speckle noise suppression channel corresponding to the SAR image; then, channel fusion is performed on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and target detection is performed based on the pseudo-color SAR image.

[0068] By adaptively mapping the amplitude of the SAR image, we obtain the corresponding dark background radiation enhancement channel and bright target radiation enhancement channel. This effectively solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, thus stabilizing the amplitude distribution of the SAR image. Furthermore, by suppressing speckle noise in the SAR image, we obtain the corresponding speckle noise suppression channel. This effectively solves the problem of visual feature quality degradation caused by speckle noise in the SAR image damaging target structure, texture, and other features. By suppressing speckle noise, we improve the visual feature quality. Finally, by combining the pseudo-color SAR image obtained by fusing the dark background radiation enhancement channel, bright target radiation enhancement channel, and speckle noise suppression channel, we can effectively improve the robust generalization ability of target detection.

[0069] It is understood that the execution subject of the synthetic aperture radar image target detection method against complex interference provided in this application can be an electronic device such as a computer, server, or specially set SAR image target detection equipment, or it can be a synthetic aperture radar image target detection device against complex interference set in such electronic device. The synthetic aperture radar image target detection device against complex interference can be implemented by software, hardware, or a combination of both, and can be set according to actual needs.

[0070] The synthetic aperture radar image target detection method against complex interference provided in this application will be described in detail below through several specific embodiments. It is understood that these specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0071] Figure 2 This application provides a schematic flowchart of a synthetic aperture radar image target detection method for resisting complex interference. For example, please refer to... Figure 2 As shown, this synthetic aperture radar image target detection method resistant to complex interference may include:

[0072] S201. Acquire the synthetic aperture radar (SAR) image to be detected.

[0073] For example, when acquiring a SAR image to be detected, it can be retrieved from local storage; it can also be received from other devices; or it can be retrieved from a third-party database, etc. The specific settings can be configured according to actual needs.

[0074] After acquiring the SAR image to be detected, on the one hand, in order to solve the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, in this embodiment of the application, adaptive amplitude mapping can be performed on the SAR image, that is, the following S202 is executed, thereby stabilizing the amplitude distribution of the SAR image; on the other hand, in order to solve the problem of visual feature quality degradation caused by speckle noise in the SAR image destroying the target structure, texture and other features, in this embodiment of the application, speckle noise suppression can be performed on the SAR image, that is, the following S203 is executed, thereby suppressing speckle noise and improving visual feature quality.

[0075] S202. Perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image.

[0076] Among them, adaptive amplitude mapping is an amplitude mapping method designed based on the amplitude distribution characteristics of SAR images. It forms a stable and refined dark background radiation enhancement channel and a bright target radiation enhancement channel, which can effectively solve the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, and stabilize the amplitude distribution of SAR images.

[0077] For example, the dark background radiation enhancement channel can be denoted as... The enhanced radiation channel for bright targets can be denoted as .

[0078] S203. Perform speckle noise suppression on the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image.

[0079] For example, the speckle noise suppression channel can be denoted as... .

[0080] It should be noted that in the embodiments of this application, there is no sequential order between S202 and S203. The embodiments of this application are only used as an example of executing S202 first and then S203, but this does not mean that the embodiments of this application are limited to this.

[0081] S204. Channel fusion is performed on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and target detection is performed based on the pseudo-color SAR image.

[0082] For example, a false-color SAR image can be denoted as .

[0083] The SAR images to be detected are subjected to adaptive amplitude mapping and speckle noise suppression to obtain the feature enhancement channel, namely the dark background radiation enhancement channel. The radiation enhancement channel for bright targets can be denoted as... The coherence speckle noise suppression channel can be denoted as... Enhanced background radiation channels in dark areas Bright target radiation enhancement channel and speckle noise suppression channel Multi-channel fusion is performed to obtain a pseudo-color SAR image with enhanced radiometric and visual features. This is achieved by fusing adaptive amplitude mapping results and speckle noise suppression results, forming a multi-channel pseudo-color SAR image with enhanced radiometric and visual features. This optimizes the radiometric and visual features that target detection depends on at the data source level, thereby improving the generalization performance of SAR image target detection.

[0084] For example, in the embodiments of this application, target detection based on pseudo-color SAR images can include at least two of the following possible implementation methods:

[0085] In one possible implementation, target detection can be performed based on a pseudo-color enhancement detection method.

[0086] Techniques for converting grayscale SAR images to color images enhance the visual information of the image through nonlinear coding and color mapping. For example, utilizing... Transforming and separating the color and grayscale information of a pseudo-color image can better highlight target features, thereby improving the accuracy and robustness of target detection.

[0087] In another possible implementation, object detection is performed using a deep learning-based object detection method.

[0088] The pseudo-color SAR image is input into the target detection model to obtain the target detection result corresponding to the SAR image; the target detection model is trained based on the pseudo-color SAR image samples corresponding to multiple SAR image samples and the target detection result labels corresponding to each SAR image sample.

[0089] For example, the object detection model can be a convolutional neural network (CNN) or a region-based convolutional neural network (R-CNN), and the specific configuration can be determined according to actual needs.

[0090] For example, in this embodiment of the application, the target detection training set consisting of multiple SAR image samples can be denoted as: , Indicates the first SAR image samples, , , , , and All represent model parameters. Each SAR image sample undergoes adaptive amplitude mapping and speckle noise suppression, resulting in a target detection training set composed of pseudo-color SAR image samples enhanced with multi-channel features. An initial object detection model was constructed, and based on the object detection training set... The pseudo-color SAR image samples and the corresponding target detection result labels of the SAR image samples are used to train and optimize the model weights of the initial target detection model. and model weights Load into object detection model In this process, the target detection model is obtained.

[0091] It is understood that when performing target detection based on pseudo-color SAR images, the embodiments of this application are only used as examples of the two possible implementation methods described above, but do not mean that the embodiments of this application are limited to these.

[0092] As can be seen in the embodiments of this application, when performing target detection based on SAR images, adaptive amplitude mapping is performed on the SAR image to be detected to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image; speckle noise suppression is performed on the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image; then, channel fusion is performed on the dark background radiation enhancement channel, the bright target radiation enhancement channel and the speckle noise suppression channel to obtain a pseudo-color SAR image, and target detection is performed based on the pseudo-color SAR image. By adaptively mapping the amplitude of the SAR image, we obtain the corresponding dark background radiation enhancement channel and bright target radiation enhancement channel. This effectively solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, thus stabilizing the amplitude distribution of the SAR image. Furthermore, by suppressing speckle noise in the SAR image, we obtain the corresponding speckle noise suppression channel. This effectively solves the problem of visual feature quality degradation caused by speckle noise in the SAR image damaging target structure, texture, and other features. By suppressing speckle noise, we improve the visual feature quality. Finally, by combining the pseudo-color SAR image obtained by fusing the dark background radiation enhancement channel, bright target radiation enhancement channel, and speckle noise suppression channel, we can effectively improve the robust generalization ability of target detection.

[0093] Based on the above Figure 2 In the illustrated embodiment, for example, the specific implementation of adaptive amplitude mapping of the SAR image in S202 above to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image can be found in the following... Figure 3 The example shown.

[0094] Figure 3 This application provides a schematic diagram of an adaptive amplitude mapping process for SAR images. For example, please refer to [link to relevant documentation]. Figure 3 As shown, the method may include:

[0095] S301. Based on the amplitude distribution probability density distribution model of SAR images, construct the amplitude mapping function corresponding to the SAR images.

[0096] For example, the amplitude distribution probability density distribution model can be denoted as: , and based on By constructing an amplitude mapping function corresponding to the SAR image, and combining this with the amplitude distribution of the SAR image, the amplitude mapping function can be constructed in a targeted manner, which can effectively improve the adaptability and accuracy of the amplitude mapping function, thereby obtaining stable and detailed radiation characteristics.

[0097] S302. Using the amplitude mapping function, and based on the amplitude value of the target that appears most frequently in the SAR image, determine the amplitude boundary point for long trailing truncation, the amplitude mapping boundary point for dark background enhancement channel, and the amplitude mapping boundary point for bright target enhancement channel.

[0098] For example, when determining the amplitude value of the most frequently occurring target in a SAR image, the amplitude distribution statistical histogram of the SAR image can be calculated. and search histogram The amplitude value that appears most frequently Therefore, based on the target amplitude value that appears most frequently. An amplitude mapping function is used, and after adaptive amplitude mapping processing, the amplitude boundary points for long trailing truncation, dark background enhancement channel amplitude mapping, and bright target enhancement channel amplitude mapping are determined respectively.

[0099] For example, based on the target amplitude value that appears most frequently. The determined cutoff point for the long trailing edge can be denoted as: , The boundary point of the amplitude mapping for the dark background enhancement channel can be denoted as: , The boundary point for the amplitude mapping of the enhancement channel for bright targets can be denoted as... , .

[0100] It should be noted that, in this embodiment, although the mathematical models describing the amplitude distribution of SAR images differ for different scenarios and imaging conditions, these mathematical models all possess the basic characteristic of a single peak and a long trailing tail. Therefore, in this embodiment, a specific threshold for truncating the amplitude of the long trailing tail is determined. The cutoff point of the long trail This can be understood as a threshold used to "trim" the extreme parts of the long trailing edge, that is, all points in the SAR image that exceed the truncation threshold for the long trailing edge. The amplitude is set to This effectively limits the impact of extreme brightness values, preparing for subsequent contrast stretching. It's equivalent to setting a "saturation point" for the SAR image, preventing individual extremely bright pixels from compressing the display range of all other pixels.

[0101] S303. Based on the amplitude boundary of the long trailing truncation, the amplitude mapping boundary of the dark background enhancement channel, and the amplitude mapping boundary of the bright target enhancement channel, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel.

[0102] For example, in this embodiment of the application, when performing adaptive amplitude mapping on the SAR image based on the long trailing truncation amplitude boundary, the dark background enhancement channel amplitude mapping boundary, and the bright target enhancement channel amplitude mapping boundary, in conjunction with the description in S302 above, in order to effectively limit the influence of extreme high brightness values ​​and prepare for subsequent contrast stretching, the amplitude of the SAR image can first be adjusted based on the long trailing truncation amplitude boundary, taking values ​​greater than the long trailing truncation amplitude boundary. The amplitude is set to the long trailing truncation amplitude boundary point to obtain the image amplitude range; and within the image amplitude range, the value smaller than the dark background enhancement channel amplitude mapping boundary point is used. The amplitude is mapped to the first amplitude range, for example, [0, 127], which is greater than or equal to the amplitude mapping boundary point of the dark background enhancement channel. The amplitude is mapped to a second amplitude range, for example, [128, 255] to obtain the dark background radiation enhancement channel. ; and map the boundary point of the enhancement channel amplitude for targets smaller than the image amplitude range. The amplitude is mapped to the first amplitude interval, and is greater than or equal to the amplitude mapping boundary point of the bright target enhancement channel. The amplitude is mapped to the second amplitude range to obtain the radiation enhancement channel for bright targets. For example, see Figure 4 As shown, Figure 4 This is a schematic diagram of a dark background radiation enhancement channel and a bright target radiation enhancement channel obtained through adaptive amplitude mapping, which solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, thereby stabilizing the amplitude distribution of SAR images.

[0103] The amplitude value in the second amplitude interval is greater than the amplitude value in the first amplitude interval.

[0104] As can be seen, in this embodiment of the application, when performing adaptive amplitude mapping on SAR images, an amplitude mapping function corresponding to the SAR image can be constructed in a targeted manner based on the amplitude distribution probability density distribution model of the SAR image. Using the amplitude mapping function, and based on the target amplitude value that appears most frequently in the SAR image, the amplitude boundary point of the long trailing truncation, the amplitude mapping boundary point of the dark background enhancement channel, and the amplitude mapping boundary point of the bright target enhancement channel are determined respectively. Then, based on the amplitude boundary point of the long trailing truncation, the amplitude mapping boundary point of the dark background enhancement channel, and the amplitude mapping boundary point of the bright target enhancement channel, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel. This solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, thereby stabilizing the amplitude distribution of the SAR image better.

[0105] Based on any of the above embodiments, for example, in S203 above, the specific implementation of speckle noise suppression of the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image can be found below. Figure 5 The example shown.

[0106] Figure 5 This application provides a schematic diagram of a process for suppressing speckle noise in SAR images. For example, please refer to [link to relevant documentation]. Figure 5 As shown, the method may include:

[0107] S501. Perform homomorphic transformation on the SAR image to obtain an additive modeling image.

[0108] For example, homomorphic transformation algorithms can include logarithmic transformation algorithms, exponential transformation algorithms, etc., and the specific algorithms can be set according to actual needs.

[0109] Logarithmic transformation is one of the most commonly used homomorphic transformation methods. The multiplicative noise characteristics in SAR images can be transformed into additive noise through logarithmic transformation, thereby simplifying the noise model and facilitating subsequent filtering and feature extraction. Exponential transformation is another common homomorphic transformation method, usually used to restore additively modeled images to the original SAR images. This method is often used in conjunction with logarithmic transformation to complete model transformations in SAR image processing.

[0110] Considering that speckle noise is a type of multiplicative noise with complex properties and is difficult to suppress, homomorphic transformation of SAR images can convert multiplicative noise into additive noise, thereby changing the nature of speckle noise. This can effectively reduce the impact of multiplicative noise on image quality, thereby improving the usability and analysis accuracy of SAR images.

[0111] S502. Perform multiple Bernoulli samplings on the additive modeling image to obtain multiple randomly sampled images.

[0112] Bernoulli sampling is a random sampling method based on the Bernoulli distribution, in which the selection of a sample is determined with a fixed probability in each sampling.

[0113] For example, when performing Bernoulli sampling on an additively modeled image, you can perform pixel-by-pixel Bernoulli sampling, region Bernoulli sampling, or weighted Bernoulli sampling, etc., which can be set according to actual needs.

[0114] Pixel-wise Bernoulli sampling involves selecting each pixel in an additively modeled image with a set probability p, while unselected pixels are set to zero or other default values. Repeating this process multiple times yields multiple randomly sampled images. This method is simple and direct, suitable for image downsampling or sparse representation. Region Bernoulli sampling divides the image into regions, with each region randomly selected for retention according to a Bernoulli distribution. The sampling density can be controlled by adjusting the sampling probability p. This method is suitable for scenarios requiring preservation of local image structure. Weighted Bernoulli sampling assigns different sampling probabilities based on pixel values ​​or region characteristics during the sampling process. For example, highlighted regions are given a higher sampling probability to enhance the preservation of key information. This method offers greater flexibility to adapt to different image processing needs.

[0115] Based on the above description, multiple random sampled images can be obtained by performing multiple Bernoulli samplings on the additive modeling image.

[0116] S503. For each randomly sampled image, the randomly sampled image is input into a pre-trained speckle noise suppression network. The speckle noise suppression network is used to suppress speckle noise in the randomly sampled image to obtain a reconstructed SAR image with noise suppression.

[0117] For example, in the embodiments of this application, a self-supervised optimization algorithm based on homomorphic transformation processing and Bernoulli sampling can be used to train the speckle noise suppression network. The specific training process may include:

[0118] Based on an ideal noise-free image and multiplicative speckle noise, SAR image samples are modeled. Homomorphic transformation is performed on the SAR image samples to obtain additively modeled image samples. Bernoulli sampling is then performed on the additively modeled image samples, for example, 10 Bernoulli samplings, to obtain randomly sampled image samples. These randomly sampled image samples are input into an initial speckle noise suppression network, which suppresses speckle noise in the randomly sampled image samples, resulting in reconstructed noise-suppressed SAR image samples. Based on the loss between the additively modeled image samples and the reconstructed noise-suppressed SAR image samples, the backpropagation algorithm iteratively prioritizes the model parameters in the initial speckle noise suppression network to train the speckle noise suppression network. This allows for the training of the speckle noise suppression network using a self-supervised optimization algorithm based on homomorphic transformation and Bernoulli sampling.

[0119] For example, the initial speckle noise suppression network can be denoted as: First, based on an ideal noise-free image and multiplicative speckle noise, a SAR image sample is modeled, as shown in Equation 1 below:

[0120] Formula 1

[0121] in, This represents a SAR image sample used for modeling. Represents an ideal noise-free image. This represents multiplicative speckle noise.

[0122] Homomorphic transformation is performed on SAR image samples to obtain additively modeled image samples, as shown in Equation 2 below:

[0123] Formula 2

[0124] in, H(represents an additively modeled image sample) ) indicates that a homomorphic transformation is performed on the SAR image sample.

[0125] When performing Bernoulli sampling on additive modeling image samples, for example, a Bernoulli sampling with a probability of 1 / 16 can be performed, that is, the additive modeling image is sampled with a probability of 15 / 16. The amplitude value is set to 0, and then random sampled image samples are obtained. and randomly sampled image samples Input initial speckle noise suppression network This yields reconstructed SAR image samples with noise suppression, which are then combined with additive modeling images. Calculate the self-supervised loss function The backpropagation algorithm iteratively prioritizes the initialization of model parameters in the speckle noise suppression network. Thus, a speckle noise suppression network is trained to perform speckle noise suppression.

[0126] For example, in an embodiment of this application, a randomly sampled image is input into a pre-trained speckle noise suppression network. When the speckle noise suppression network suppresses speckle noise on the randomly sampled image, it combines... Figure 6 As shown, Figure 6 This is a schematic diagram of the framework of a speckle noise suppression network provided in an embodiment of this application. First, a randomly sampled image is input to the block processing module in the speckle noise suppression network. The block processing module divides the randomly sampled image into blocks, obtaining multiple image blocks. These multiple image blocks are then input to the feature embedding module in the speckle noise suppression network. The feature embedding module embeds features into the multiple image blocks, obtaining an embedded feature space vector. Next, the embedded feature space vector is input to the feature encoding module in the speckle noise suppression network. The feature encoding module encodes the embedded feature space vector and removes redundant noise during the encoding process to suppress speckle noise, obtaining a deep semantic feature vector. Finally, the deep semantic feature vector is input to the feature decoding module in the speckle noise suppression network. The feature decoding module decodes the deep semantic feature vector to obtain a reconstructed, noise-suppressed SAR image. For example, the feature decoding module can consist of a single fully connected layer.

[0127] After performing speckle noise suppression on the randomly sampled image using a pre-trained speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image, the following S504 can be executed:

[0128] S504. Based on multiple reconstructed noise-suppressed SAR images, determine the speckle noise suppression channel.

[0129] For example, in the embodiments of this application, when determining the speckle noise suppression channel based on multiple reconstructed noise-suppressed SAR images, the average SAR image corresponding to the multiple reconstructed noise-suppressed SAR images can be determined first; then, the average SAR image can be subjected to inverse homomorphic transformation to obtain the speckle noise suppression channel. This can effectively solve the problem of visual feature quality degradation caused by speckle noise in SAR images damaging target structure, texture and other features, thereby suppressing speckle noise and improving visual feature quality.

[0130] It should be noted that, in the above-mentioned acquisition of speckle noise suppression channels, this embodiment of the application only takes the example of first determining the average SAR image corresponding to multiple reconstructed noise-suppressed SAR images, and then performing an inverse homomorphic transformation on the average SAR image to obtain the speckle noise suppression channels. Of course, it is also possible to perform inverse homomorphic transformations on multiple reconstructed noise-suppressed SAR images separately, and then average the results of multiple transformations to obtain the speckle noise suppression channels. The specific settings can be made according to actual needs.

[0131] As can be seen from the embodiments of this application, when suppressing speckle noise in SAR images, the SAR images can first undergo homomorphic transformation to obtain additive modeling images; then, multiple Bernoulli samplings are performed on the additive modeling images to obtain multiple random sampling images; for each random sampling image, the random sampling image is input into a pre-trained speckle noise suppression network, and speckle noise is suppressed on the random sampling images through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image; then, based on multiple reconstructed noise-suppressed SAR images, the speckle noise suppression channel is determined. In this way, by performing homomorphic transformation on the SAR image, multiplicative noise can be converted into additive noise, thus changing the nature of speckle noise. On this basis, a self-supervised optimization speckle noise suppression network is designed to enhance visual features such as structure and texture while suppressing speckle noise, thereby restoring the key visual features of the target while suppressing speckle noise.

[0132] In combination with the above Figure 3 and Figure 5 As shown, after obtaining the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel, multi-channel fusion can be performed on these channels, and target detection can be performed on the pseudo-color SAR image after multi-channel feature enhancement. To facilitate understanding of the SAR image-based target detection method provided in this application, a specific embodiment will be used to describe the SAR image-based target detection method in detail below.

[0133] First, acquire the SAR image to be detected.

[0134] Secondly, adaptive amplitude mapping and speckle noise suppression are performed on the SAR images to be detected.

[0135] In this process, adaptive amplitude mapping is performed on the SAR image to be detected to obtain the dark background radiation enhancement channel. and the radiation enhancement channel for bright targets The specific process is as follows:

[0136] 1) Calculate the amplitude distribution statistical histogram of the input SAR image and search for the amplitude value that appears most frequently in the histogram. 2) Based on the amplitude value that appears most frequently Calculate three amplitude mapping threshold values, namely: long trailing truncation amplitude threshold. Dark background enhancement channel amplitude mapping boundary point and the boundary point of the amplitude mapping of the bright target enhancement channel. ;3) Based on the long trailing amplitude boundary point Truncate the long trailing amplitude distribution of a SAR image, that is, truncate all points in the SAR image whose amplitude exceeds the threshold of the long trailing amplitude. The amplitude is set to ;4) Map the boundary point based on the amplitude of the dark background enhancement channel. Boundary point for amplitude mapping of bright target enhancement channel The truncated amplitude range is segmented and mapped twice. In the first segmentation mapping, the amplitude mapping boundary point of the dark background enhancement channel is set to be smaller than that of the dark background enhancement channel. The amplitude value is mapped to [0, 127], and is greater than or equal to the amplitude mapping boundary point of the dark background enhancement channel. The amplitude value is mapped to [128, 255] to obtain the dark background radiation enhancement channel. In the second segmented mapping, the mapping boundary point is used to map the amplitude of the enhancement channel for targets smaller than those in the bright areas. The amplitude values ​​are mapped to [0, 127], and the boundary points for amplitude mapping of the highlight target enhancement channel are set to be greater than or equal to the boundary points. The amplitude values ​​are mapped to [128, 255] to obtain the radiation enhancement channel for bright targets. This adaptive mapping function, based on the probability density model of SAR image amplitude distribution, obtains the dark background radiation enhancement channel and the bright target radiation enhancement channel through two amplitude mappings, respectively. This solves the problem of unstable target radiation characteristics caused by fluctuations due to changes in scene and imaging conditions, and thus can better stabilize the amplitude distribution of SAR images.

[0137] Speckle noise suppression is performed on the SAR image to be detected to obtain the speckle noise suppression channel. The specific process is as follows:

[0138] 1) The SAR image to be detected is homomorphically transformed to convert multiplicative noise into additive noise, thus changing the speckle noise nature; 2) Multiple Bernoulli samplings are performed on the homomorphically transformed image to obtain multiple randomly sampled images; 3) These multiple randomly sampled images are input into a speckle noise suppression network. 4) Determine the average SAR image corresponding to the multiple reconstructed noise-suppressed SAR images; then perform an inverse homomorphic transformation on the average SAR image to obtain the speckle noise suppression channel. For example, see [link to example]. Figure 7 As shown, Figure 7 This is a comparative schematic diagram of a SAR image to be detected and a speckle noise suppression channel provided in an embodiment of this application. In this way, multiplicative noise is converted into additive noise, the nature of speckle noise is transformed, and a self-supervised optimization speckle noise suppression network is designed on this basis. While suppressing speckle noise, visual features such as structure and texture are enhanced, thereby achieving the restoration of key visual features of the target while suppressing speckle noise.

[0139] Furthermore, enhance the dark background radiation channel. Bright target radiation enhancement channel and speckle noise suppression channel Channel fusion is performed to obtain a pseudo-color SAR image with enhanced multi-channel features. A pseudo-color SAR image with multi-channel feature enhancement is obtained by fusing the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel. For example, see [link to example]. Figure 8 As shown, Figure 8 This is a comparative diagram of a SAR image to be detected and a pseudo-color SAR image provided in an embodiment of this application, which provides a basis for improving the generalization performance of SAR image target detection in the future.

[0140] Finally, the pseudo-color SAR image enhanced with multi-channel features is input into the target detection model to obtain the target detection results. For example, see [link to example]. Figure 9 As shown, Figure 9 This is a schematic diagram of the target detection effect provided in an embodiment of this application. Compared with the prior art, the target detection method based on SAR images provided in this embodiment of the application can improve the average accuracy of target detection by 7.2%, indicating that it achieves better target detection generalization performance under the condition that the radiation characteristics are unstable and the visual characteristics are interfered with by speckle noise.

[0141] As can be seen, the target detection method based on SAR images provided in this application not only enhances the stability of the target radiation features in the SAR image by performing adaptive amplitude mapping and speckle noise suppression on the SAR image to be detected, but also suppresses speckle noise and restores the visual features destroyed by speckle noise to a certain extent, thereby enhancing key visual features such as target structure and texture. It optimizes the data source quality of SAR image target detection from both radiation and visual features, and ultimately improves the generalization performance of SAR image target detection.

[0142] The synthetic aperture radar image target detection device against complex interference provided in this application is described below. The synthetic aperture radar image target detection device against complex interference described below can be referred to in correspondence with the synthetic aperture radar image target detection method against complex interference described above.

[0143] Figure 10 This application provides a schematic diagram of a synthetic aperture radar image target detection device for resisting complex interference. For example, please refer to... Figure 10 As shown, the synthetic aperture radar image target detection device 100, which is resistant to complex interference, may include:

[0144] Acquisition unit 1001 is used to acquire the synthetic aperture radar (SAR) image to be detected;

[0145] The mapping unit 1002 is used to perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image.

[0146] The suppression unit 1003 is used to suppress speckle noise in the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image;

[0147] The fusion detection unit 1004 is used to perform channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel and the speckle noise suppression channel to obtain a pseudo-color SAR image, and to perform target detection based on the pseudo-color SAR image.

[0148] For example, in this embodiment of the application, the mapping unit 1002 is used to perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image, including:

[0149] Based on the amplitude distribution probability density distribution model of the SAR image, an amplitude mapping function corresponding to the SAR image is constructed;

[0150] Using the amplitude mapping function, and based on the amplitude value of the target that appears most frequently in the SAR image, the amplitude boundary point for long trailing truncation, the amplitude mapping boundary point for dark background enhancement channel, and the amplitude mapping boundary point for bright target enhancement channel are determined respectively.

[0151] Based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel.

[0152] For example, in an embodiment of this application, the mapping unit 1002 is used to perform adaptive amplitude mapping on the SAR image based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel, including:

[0153] The amplitude of the SAR image that is greater than the long trailing truncation amplitude boundary is set as the long trailing truncation amplitude boundary, thus obtaining the image amplitude range.

[0154] The amplitude values ​​within the image amplitude range that are smaller than the boundary point of the dark background enhancement channel amplitude mapping are mapped to a first amplitude interval, and the amplitude values ​​that are greater than or equal to the boundary point of the dark background enhancement channel amplitude mapping are mapped to a second amplitude interval, thus obtaining the dark background radiation enhancement channel; wherein, the amplitude value in the second amplitude interval is greater than the amplitude value in the first amplitude interval;

[0155] The amplitudes within the image amplitude range that are smaller than the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the first amplitude range, and the amplitudes that are greater than or equal to the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the second amplitude range, thus obtaining the brightness target radiation enhancement channel.

[0156] For example, in an embodiment of this application, the suppression unit 1003 is used to suppress speckle noise in the SAR image to obtain a speckle noise suppression channel corresponding to the SAR image, including:

[0157] The SAR image is homomorphically transformed to obtain an additively modeled image;

[0158] Multiple Bernoulli samples are performed on the additive modeling image to obtain multiple randomly sampled images;

[0159] For each of the randomly sampled images, the randomly sampled images are input into a pre-trained speckle noise suppression network, and speckle noise is suppressed on the randomly sampled images through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image;

[0160] Based on multiple reconstructed noise-suppressed SAR images, the speckle noise suppression channel is determined.

[0161] For example, in an embodiment of this application, the suppression unit 1003 is used to input the randomly sampled image into a pre-trained speckle noise suppression network, and to perform speckle noise suppression on the randomly sampled image through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image, including:

[0162] The randomly sampled image is input into the block processing module in the speckle noise suppression network, and the block processing module performs block processing on the randomly sampled image to obtain multiple image blocks.

[0163] The multiple image patches are input into the feature embedding module in the speckle noise suppression network, and the feature embedding module performs feature embedding on the multiple image patches to obtain an embedded feature space vector;

[0164] The embedded feature space vector is input into the feature encoding module in the speckle noise suppression network. The embedded feature space vector is encoded by the feature encoding module, and speckle noise suppression is performed during the encoding process to obtain a deep semantic feature vector.

[0165] The deep semantic feature vector is input into the feature decoding module in the speckle noise suppression network. The feature decoding module decodes the deep semantic feature vector to obtain the reconstructed noise-suppressed SAR image.

[0166] For example, in an embodiment of this application, the training process of the speckle noise suppression network includes:

[0167] Modeling SAR image samples based on ideal noise-free images and multiplicative speckle noise;

[0168] The SAR image samples are homomorphically transformed to obtain additive modeling image samples; and the additive modeling image samples are Bernoulli sampled to obtain random sampled image samples.

[0169] The randomly sampled image samples are input into an initial speckle noise suppression network, and speckle noise is suppressed on the randomly sampled image samples through the initial speckle noise suppression network to obtain reconstructed noise-suppressed SAR image samples;

[0170] Based on the loss between the additively modeled image samples and the reconstructed noise-suppressed SAR image samples, the backpropagation algorithm iteratively prioritizes the model parameters in the initial speckle noise suppression network to train the speckle noise suppression network.

[0171] For example, in an embodiment of this application, the suppression unit 1003 is used to determine the speckle noise suppression channel based on multiple reconstructed noise-suppressed SAR images, including:

[0172] Determine the average SAR image corresponding to the plurality of reconstructed noise-suppressed SAR images;

[0173] The average SAR image is subjected to inverse homomorphic transformation to obtain the speckle noise suppression channel.

[0174] For example, in an embodiment of this application, the fusion detection unit 1004 is used to perform target detection based on the pseudo-color SAR image, including:

[0175] The pseudo-color SAR image is input into the target detection model to obtain the target detection result corresponding to the SAR image;

[0176] The target detection model is trained based on the pseudo-color SAR image samples corresponding to each of the multiple SAR image samples, and the target detection result labels corresponding to each of the SAR image samples.

[0177] The synthetic aperture radar image target detection device 100 with anti-complex interference provided in this application embodiment can execute the technical solution of the synthetic aperture radar image target detection method with anti-complex interference in any of the above embodiments. Its implementation principle and beneficial effects are similar to those of the synthetic aperture radar image target detection method with anti-complex interference. Please refer to the implementation principle and beneficial effects of the synthetic aperture radar image target detection method with anti-complex interference, which will not be repeated here.

[0178] Figure 11 This is a schematic diagram of the physical structure of an electronic device provided in an embodiment of this application, such as... Figure 11 As shown, the electronic device may include a processor 1110, a communications interface 1120, a memory 1130, and a communication bus 1140, wherein the processor 1110, the communications interface 1120, and the memory 1130 communicate with each other via the communication bus 1140. The processor 1110 can call logical instructions in the memory 1130 to execute a synthetic aperture radar (SAR) image target detection method resistant to complex interference. The method includes: acquiring a SAR image to be detected; performing adaptive amplitude mapping on the SAR image to obtain a dark background radiation enhancement channel and a bright target radiation enhancement channel corresponding to the SAR image; performing speckle noise suppression on the SAR image to obtain a speckle noise suppression channel corresponding to the SAR image; performing channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and performing target detection based on the pseudo-color SAR image.

[0179] Furthermore, the logical instructions in the aforementioned memory 1130 can be implemented as software functional units and, when sold or used as independent products, 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 a part of the technical solution, can be 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 (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0180] On the other hand, this application also provides a computer program product, which includes a computer program that can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the synthetic aperture radar image target detection method against complex interference provided by the above methods. The method includes: acquiring a synthetic aperture radar (SAR) image to be detected; performing adaptive amplitude mapping on the SAR image to obtain a dark background radiation enhancement channel and a bright target radiation enhancement channel corresponding to the SAR image; performing speckle noise suppression on the SAR image to obtain a speckle noise suppression channel corresponding to the SAR image; performing channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image, and performing target detection based on the pseudo-color SAR image.

[0181] In another aspect, this application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a synthetic aperture radar (SAR) image target detection method against complex interference provided by the methods described above. The method includes: acquiring a SAR image to be detected; performing adaptive amplitude mapping on the SAR image to obtain a dark background radiation enhancement channel and a bright target radiation enhancement channel corresponding to the SAR image; performing speckle noise suppression on the SAR image to obtain a speckle noise suppression channel corresponding to the SAR image; performing channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel to obtain a pseudo-color SAR image; and performing target detection based on the pseudo-color SAR image.

[0182] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0183] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0184] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A synthetic aperture radar image target detection method resistant to complex interference, characterized in that, include: Acquire synthetic aperture radar (SAR) images of the target object; Adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image; The SAR image is subjected to speckle noise suppression to obtain the speckle noise suppression channel corresponding to the SAR image; The dark background radiation enhancement channel, the bright target radiation enhancement channel, and the speckle noise suppression channel are fused to obtain a pseudo-color SAR image, and target detection is performed based on the pseudo-color SAR image. The step of performing adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image includes: Based on the amplitude distribution probability density distribution model of the SAR image, an amplitude mapping function corresponding to the SAR image is constructed; Using the amplitude mapping function, and based on the amplitude value of the target that appears most frequently in the SAR image, the amplitude boundary point for long trailing truncation, the amplitude mapping boundary point for dark background enhancement channel, and the amplitude mapping boundary point for bright target enhancement channel are determined respectively. Based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel.

2. The method according to claim 1, characterized in that, The adaptive amplitude mapping of the SAR image based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel, includes: The amplitude of the SAR image that is greater than the long trailing truncation amplitude boundary is set as the long trailing truncation amplitude boundary, thus obtaining the image amplitude range. The amplitude values ​​within the image amplitude range that are smaller than the boundary point of the dark background enhancement channel amplitude mapping are mapped to a first amplitude interval, and the amplitude values ​​that are greater than or equal to the boundary point of the dark background enhancement channel amplitude mapping are mapped to a second amplitude interval, thus obtaining the dark background radiation enhancement channel; wherein, the amplitude value in the second amplitude interval is greater than the amplitude value in the first amplitude interval; The amplitudes within the image amplitude range that are smaller than the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the first amplitude range, and the amplitudes that are greater than or equal to the boundary point of the brightness target enhancement channel amplitude mapping are mapped to the second amplitude range, thus obtaining the brightness target radiation enhancement channel.

3. The method according to claim 1 or 2, characterized in that, The step of performing speckle noise suppression on the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image includes: The SAR image is homomorphically transformed to obtain an additively modeled image; Multiple Bernoulli samples are performed on the additive modeling image to obtain multiple randomly sampled images; For each of the randomly sampled images, the randomly sampled images are input into a pre-trained speckle noise suppression network, and speckle noise is suppressed on the randomly sampled images through the speckle noise suppression network to obtain a reconstructed noise-suppressed SAR image; Based on multiple reconstructed noise-suppressed SAR images, the speckle noise suppression channel is determined.

4. The method according to claim 3, characterized in that, The step of inputting the randomly sampled image into a pre-trained speckle noise suppression network, and using the speckle noise suppression network to suppress speckle noise in the randomly sampled image to obtain a reconstructed noise-suppressed SAR image includes: The randomly sampled image is input into the block processing module in the speckle noise suppression network, and the block processing module performs block processing on the randomly sampled image to obtain multiple image blocks. The multiple image patches are input into the feature embedding module in the speckle noise suppression network, and the feature embedding module performs feature embedding on the multiple image patches to obtain an embedded feature space vector; The embedded feature space vector is input into the feature encoding module in the speckle noise suppression network. The embedded feature space vector is encoded by the feature encoding module, and speckle noise suppression is performed during the encoding process to obtain a deep semantic feature vector. The deep semantic feature vector is input into the feature decoding module in the speckle noise suppression network. The feature decoding module decodes the deep semantic feature vector to obtain the reconstructed noise-suppressed SAR image.

5. The method according to claim 3, characterized in that, The training process of the speckle noise suppression network includes: Modeling SAR image samples based on ideal noise-free images and multiplicative speckle noise; The SAR image samples are homomorphically transformed to obtain additive modeling image samples; and the additive modeling image samples are Bernoulli sampled to obtain random sampled image samples. The randomly sampled image samples are input into an initial speckle noise suppression network, and speckle noise is suppressed on the randomly sampled image samples through the initial speckle noise suppression network to obtain reconstructed noise-suppressed SAR image samples; Based on the loss between the additively modeled image samples and the reconstructed noise-suppressed SAR image samples, the backpropagation algorithm iteratively prioritizes the model parameters in the initial speckle noise suppression network to train the speckle noise suppression network.

6. The method according to claim 3, characterized in that, The determination of the speckle noise suppression channel based on multiple reconstructed noise-suppressed SAR images includes: Determine the average SAR image corresponding to the plurality of reconstructed noise-suppressed SAR images; The average SAR image is subjected to inverse homomorphic transformation to obtain the speckle noise suppression channel.

7. The method according to claim 1 or 2, characterized in that, The target detection based on the pseudo-color SAR image includes: The pseudo-color SAR image is input into the target detection model to obtain the target detection result corresponding to the SAR image; The target detection model is trained based on the pseudo-color SAR image samples corresponding to each of the multiple SAR image samples, and the target detection result labels corresponding to each of the SAR image samples.

8. A synthetic aperture radar image target detection device resistant to complex interference, characterized in that, include: The acquisition unit is used to acquire the synthetic aperture radar (SAR) image to be detected. The mapping unit is used to perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image. The suppression unit is used to suppress speckle noise in the SAR image to obtain the speckle noise suppression channel corresponding to the SAR image; The fusion detection unit is used to perform channel fusion on the dark background radiation enhancement channel, the bright target radiation enhancement channel and the speckle noise suppression channel to obtain a pseudo-color SAR image, and to perform target detection based on the pseudo-color SAR image; The mapping unit is used to perform adaptive amplitude mapping on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel corresponding to the SAR image, including: Based on the amplitude distribution probability density distribution model of the SAR image, an amplitude mapping function corresponding to the SAR image is constructed; Using the amplitude mapping function, and based on the amplitude value of the target that appears most frequently in the SAR image, the amplitude boundary point for long trailing truncation, the amplitude mapping boundary point for dark background enhancement channel, and the amplitude mapping boundary point for bright target enhancement channel are determined respectively. Based on the long trailing truncation amplitude boundary point, the dark background enhancement channel amplitude mapping boundary point, and the bright target enhancement channel amplitude mapping boundary point, adaptive amplitude mapping is performed on the SAR image to obtain the dark background radiation enhancement channel and the bright target radiation enhancement channel.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the synthetic aperture radar image target detection method for resisting complex interference as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the synthetic aperture radar image target detection method against complex interference as described in any one of claims 1 to 7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the synthetic aperture radar image target detection method against complex interference as described in any one of claims 1 to 7.

Images