Novel high-precision synthetic aperture radar ship target detection method and system

Abstract

The invention discloses a novel high-precision synthetic aperture radar ship target detection method, which comprises the steps of obtaining original SAR image data for preprocessing, performing multi-scale feature extraction and polarization feature fusion, performing sea clutter adaptive suppression, performing multi-level attention enhancement, performing target positioning and classification and performing post-processing optimization. And finally, outputting a final ship target detection result, wherein the final ship target detection result comprises bounding box coordinates, category labels and confidence scores. According to the method, an end-to-end ship target detection process is constructed, multi-dimensional information of the SAR image can be fully utilized, different sea conditions and target scales can be adaptively coped with, the detection precision and robustness are remarkably improved, the false alarm rate under the complex sea condition is effectively reduced, and the method is particularly suitable for ship detection under the high sea condition.

Description

Technical Field

[0001] This invention belongs to the field of synthetic aperture radar image processing and target detection, specifically relating to a novel high-precision synthetic aperture radar method and system for ship target detection. Background Technology

[0002] Synthetic Aperture Radar (SAR) has all-weather, all-time operating capabilities and is widely used in fields such as marine monitoring and ship inspection. Existing SAR methods for ship target detection mainly suffer from the following problems:

[0003] Synthetic aperture radar (SAR) for ship target detection faces challenges such as large target size variations, strong sea clutter interference, and difficulty in detecting small targets.

[0004] Traditional constant false alarm rate (CFAR) detection methods rely on fixed statistical models and thresholds, which can easily generate a large number of false alarms under complex sea conditions, and have insufficient detection capabilities for small targets and nearshore targets.

[0005] While deep learning-based detection methods have improved accuracy, they suffer from problems such as poor adaptability to multi-scale targets, insufficient feature extraction, and slow detection speed.

[0006] Furthermore, existing methods often only utilize the intensity information of SAR images, ignoring the polarization information and sea clutter statistical characteristics of SAR images, which limits detection performance. Summary of the Invention

[0007] To address the aforementioned problems, the present invention aims to provide a novel high-precision synthetic aperture radar (SAR) method and system for ship target detection, which addresses the issues of high false alarm rate, weak small target detection capability, and poor adaptability to multi-scale targets in traditional methods.

[0008] The specific technical solution for achieving the objective of this invention is as follows:

[0009] A novel high-precision synthetic aperture radar method for ship target detection includes the following steps:

[0010] Step 1: Acquire raw SAR image data, including intensity images and polarization channel data, and preprocess the SAR images;

[0011] Step 2: Extract shallow, medium, and deep multi-scale features of the image using the multi-scale feature extraction module, and then perform pyramid fusion.

[0012] Step 3: Extract the polarization features of the SAR image through the polarization feature fusion module and fuse them with multi-scale features;

[0013] Step 4: Estimate the local sea state using the sea clutter adaptive suppression module, establish a clutter statistical model, and adaptively suppress sea clutter;

[0014] Step 5: Generate spatial attention and channel attention weights through a multi-level attention enhancement module to enhance target features;

[0015] Step 6: Generate multi-scale anchor boxes through the target localization and classification module, perform feature alignment, and predict bounding box coordinates and class probabilities;

[0016] Step 7: Perform confidence screening, non-maximum suppression, and scale consistency verification through the post-processing optimization module;

[0017] Step 8: Output the final ship target detection results, including bounding box coordinates, category labels, and confidence scores.

[0018] This solution also provides a novel high-precision synthetic aperture radar ship target detection system, including the following modules:

[0019] Preprocessing unit: Used to acquire raw SAR image data, including intensity image and polarization channel data, and to preprocess the SAR image;

[0020] Multi-scale feature extraction module: used to extract shallow, medium and deep multi-scale features of the image and perform pyramid fusion;

[0021] Polarization feature fusion module: used to extract polarization features from SAR images and fuse them with multi-scale features;

[0022] Adaptive Sea Clutter Suppression Module: Used to estimate local sea state, establish a statistical model of clutter, and adaptively suppress sea clutter;

[0023] Multi-level attention enhancement module: used to generate spatial attention and channel attention weights to enhance target features;

[0024] The target localization and classification module is used to generate multi-scale anchor boxes, perform feature alignment, and predict bounding box coordinates and class probabilities.

[0025] Post-processing optimization module: used for confidence filtering, non-maximum suppression and scale consistency verification, outputting the final ship target detection results, including bounding box coordinates, category labels and confidence scores.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] (1) The present invention adopts a multi-scale feature extraction module. Through the branch extraction and pyramid fusion of shallow, medium and deep features, it can effectively detect multi-scale ship targets from small fishing boats to large cargo ships, and significantly improve the adaptability to targets of different scales.

[0028] (2) The invention introduces a polarization feature fusion module, which makes full use of the polarization information of SAR images. Through polarization decomposition and multi-modal feature fusion, the distinction between the target and sea clutter is enhanced, and the detection accuracy is improved.

[0029] (3) The present invention designs a sea clutter adaptive suppression module, which adaptively establishes a clutter statistical model based on local sea conditions and suppresses it, effectively reducing the false alarm rate under complex sea conditions, and is particularly suitable for ship detection under high sea conditions.

[0030] (4) The present invention employs a multi-level attention enhancement module, which highlights target features and suppresses background noise through the synergistic effect of spatial attention and channel attention, thereby further improving the detection capability of small and weak targets.

[0031] (5) This invention improves the accuracy of target positioning and the reliability of detection results by using techniques such as anchor frame optimization, feature alignment and post-processing optimization, and reduces repeated detection and false detection.

[0032] (6) The novel high-precision synthetic aperture radar (SAR) ship target detection method and system proposed in this invention constructs an end-to-end ship target detection system by organically combining multi-scale feature extraction, polarization feature fusion, sea clutter adaptive suppression, and multi-level attention enhancement techniques. This system can fully utilize the multi-dimensional information of SAR images, adaptively respond to different sea conditions and target scales, and significantly improve detection accuracy and robustness.

[0033] The present invention will be further described below with reference to specific embodiments. Attached Figure Description

[0034] Figure 1 This is a schematic diagram of the process for the novel high-precision synthetic aperture radar ship target detection method of the present invention.

[0035] Figure 2 This is a schematic diagram of the multi-scale feature extraction module architecture of the present invention.

[0036] Figure 3 This is a schematic diagram of the architecture of the novel high-precision synthetic aperture radar ship target detection system of the present invention. Detailed Implementation

[0037] Example

[0038] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. The described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] As indicated in this application and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of explicitly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0040] Unless otherwise specifically stated, the relative arrangement, numerical expressions, and values ​​of the components and steps described in these embodiments do not limit the scope of this application. It should also be understood that, for ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0041] Combination Figure 1 A novel high-precision synthetic aperture radar method for detecting ship targets includes the following steps:

[0042] Step 1: Acquire raw SAR image data, including intensity images and polarization channel data, and preprocess the SAR images;

[0043] The preprocessing includes radiometric calibration, geometric correction, noise reduction filtering, and polarization channel separation;

[0044] Step 2: Extract shallow, medium, and deep multi-scale features of the image using the multi-scale feature extraction module, and then perform pyramid fusion.

[0045] The multi-scale feature extraction module includes a backbone feature extraction network, shallow feature branches, mid-level feature branches, deep feature branches, and a feature pyramid fusion unit, such as... Figure 2 As shown;

[0046] The backbone feature extraction network employs an improved residual network structure to extract the basic features of SAR images.

[0047] The shallow feature branch extracts detailed features of the image for detecting small-scale ship targets; the middle feature branch extracts intermediate semantic features of the image for detecting medium-scale ship targets; the deep feature branch extracts high-level semantic features of the image for detecting large-scale ship targets; the feature pyramid fusion unit fuses features from different levels through top-down and bottom-up feature fusion paths to generate a multi-scale fused feature map.

[0048] Specifically, in this embodiment, the backbone feature extraction network uses an improved ResNet50 as its backbone network, containing five stages of convolutional layers, progressively reducing the feature map resolution and increasing the number of feature channels. The shallow feature branch extracts features from the second stage of the backbone network, with a feature map size of 1 / 4 of the input image, containing rich detail information, suitable for small-scale object detection. The mid-level feature branch extracts features from the third stage of the backbone network, with a feature map size of 1 / 8 of the input image, containing mid-level semantic information, suitable for medium-scale object detection. The deep feature branch extracts features from the fourth stage of the backbone network, with a feature map size of 1 / 16 of the input image, containing high-level semantic information, suitable for large-scale object detection.

[0049] The Feature Pyramid Fusion Unit employs a Feature Pyramid Network (FPN) structure, upsampling deep features and fusing them with shallow features through a top-down path and lateral connections. Specifically, deep features are upsampled by a factor of 2 after 1×1 convolution for dimensionality reduction and then added to and fused with mid-level features; the fused features are then upsampled by a factor of 2 again and added to and fused with shallow features. Simultaneously, a bottom-up path enhancement is used, employing 3×3 convolution and downsampling to propagate shallow features to deeper layers, enhancing the expressive power of features at each level. The final output consists of three fused feature maps, P3, P4, and P5, at different scales, with feature map sizes of 1 / 4, 1 / 8, and 1 / 16 of the input image, respectively.

[0050] Step 3: Extract the polarization features of the SAR image through the polarization feature fusion module and fuse them with multi-scale features;

[0051] The polarization feature fusion module includes a polarization channel separation unit, a polarization decomposition unit, a polarization feature encoding unit, and a multimodal feature fusion unit;

[0052] Specifically, the polarization channel separation unit separates the fully polarimetric SAR image into four polarization channels: HH, HV, VH, and VV; the polarization decomposition unit uses Pauli decomposition or HA-Alpha decomposition to extract polarization scattering characteristic parameters; the polarization feature encoding unit encodes polarization features through convolutional layers and normalization layers to generate polarization feature vectors; and the multimodal feature fusion unit uses an attention-weighted fusion mechanism to adaptively fuse the multi-scale intensity features and polarization features obtained in step 2 to enhance the contrast between the target and sea clutter.

[0053] Step 4: Estimate the local sea state using the sea clutter adaptive suppression module, establish a clutter statistical model, and adaptively suppress sea clutter;

[0054] The adaptive sea clutter suppression module includes a sea state estimation unit, a clutter statistical modeling unit, an adaptive filtering unit, and a feature enhancement unit;

[0055] The sea state estimation unit estimates the current sea state level and wind speed through statistical analysis using a local window (e.g., 64×64 pixels); the clutter statistical modeling unit establishes a K-distribution or G0-distribution sea clutter statistical model based on the sea state parameters; the adaptive filtering unit suppresses sea clutter and preserves target information by using an adaptive threshold segmentation method based on the clutter model parameters; and the feature enhancement unit enhances the saliency of target features through residual connection and feature normalization.

[0056] The adaptive threshold is set as: T = μ + k×σ, where μ is the local mean, σ is the local standard deviation, and k is adaptively adjusted according to the sea state level, with a larger value of k for higher sea state levels.

[0057] Step 5: Generate spatial attention and channel attention weights through a multi-level attention enhancement module to enhance target features;

[0058] The multi-level attention enhancement module includes: a spatial attention submodule, a channel attention submodule, a multi-scale perception unit, and a feature weighted fusion unit;

[0059] The spatial attention submodule generates a spatial attention weight map through global pooling and convolution operations to highlight the spatial location of the target; the channel attention submodule generates a channel attention weight vector through global average pooling and fully connected layers to strengthen the feature channels related to the target; the multi-scale perception unit uses dilated convolution and a multi-branch structure to perceive target features at different receptive field scales; and the feature weighted fusion unit weights and fuses spatial attention features and channel attention features to generate an attention-enhanced feature map.

[0060] Step 6: Generate multi-scale anchor boxes through the target localization and classification module, perform feature alignment, and predict bounding box coordinates and class probabilities;

[0061] The target localization and classification module includes: an anchor box generation unit, a feature alignment unit, a bounding box regression unit, and a target classification unit;

[0062] The anchor frame generation unit generates candidate anchor frames with multiple scales and aspect ratios based on the feature map size and target scale distribution; the feature alignment unit uses deformable convolution or RoI Align method to extract features for each candidate anchor frame region; the bounding box regression unit predicts the position offset of the anchor frame through a fully connected layer to accurately regress the target bounding box coordinates; and the target classification unit outputs the class probability of each anchor frame, including ship type and background category, through a fully connected layer and Softmax activation function.

[0063] In this embodiment, anchor frame generation employs a combination of 9 scales and 3 aspect ratios, resulting in 27 anchor frame types, covering a scale range from small vessels (10-50 meters) to large vessels (100-300 meters). Bounding box regression uses the Smooth L1 loss function, and target classification uses the Focal Loss loss function, balancing positive and negative samples and easy and difficult samples.

[0064] Step 7: Perform confidence screening, non-maximum suppression, and scale consistency verification through the post-processing optimization module;

[0065] The post-processing optimization module includes a confidence screening unit, a non-maximum suppression unit, a scale consistency verification unit, and a result output unit.

[0066] The confidence screening unit filters out low-confidence detection boxes based on a preset confidence threshold; the non-maximum suppression unit removes redundant detection boxes with high overlap based on an IoU threshold, retaining the detection results with the highest confidence; the scale consistency verification unit verifies the consistency between the scale of the detection boxes and the prior scale of the ship, removing unreasonable detection results; and the result output unit integrates the position coordinates, category labels, and confidence scores of the detection boxes to output the final ship target detection result.

[0067] In this embodiment, the confidence threshold is set to 0.5 and the IoU threshold is set to 0.4. Scale consistency verification checks whether the aspect ratio of the detection box is within a reasonable range (0.1-10) and whether the area is within the prior scale range (100-90000 pixels), removing unreasonable detection results.

[0068] Step 8: Output the final ship target detection results, including bounding box coordinates, category labels, and confidence scores.

[0069] The proposed detection model has been validated on multiple publicly available SAR ship detection datasets. Compared to traditional CFAR methods, it improves average accuracy by 35% and reduces false alarm rate by 60%. Compared to existing deep learning methods, it improves average accuracy by 12% and increases the recall rate for small targets (area less than 32×32 pixels) by 18%. Under high sea state conditions (sea state 5 and above), the false alarm rate of the proposed method is only 40% of that of traditional methods, demonstrating excellent robustness.

[0070] Combination Figure 3 This solution also provides a novel high-precision synthetic aperture radar ship target detection system, including the following modules:

[0071] SAR image preprocessing unit: used to acquire raw SAR image data, including intensity image and polarization channel data, and to preprocess the SAR image;

[0072] Multi-scale feature extraction module: used to extract shallow, medium and deep multi-scale features of the image and perform pyramid fusion;

[0073] Polarization feature fusion module: used to extract polarization features from SAR images and fuse them with multi-scale features;

[0074] Adaptive Sea Clutter Suppression Module: Used to estimate local sea state, establish a statistical model of clutter, and adaptively suppress sea clutter;

[0075] Multi-level attention enhancement module: used to generate spatial attention and channel attention weights to enhance target features;

[0076] The target localization and classification module is used to generate multi-scale anchor boxes, perform feature alignment, and predict bounding box coordinates and class probabilities.

[0077] Post-processing optimization module: used for confidence filtering, non-maximum suppression and scale consistency verification, outputting the final ship target detection results, including bounding box coordinates, category labels and confidence scores.

[0078] The output of the SAR image preprocessing unit 1 is connected to the input of the multi-scale feature extraction module 2, and the auxiliary output of the SAR image preprocessing unit 1 is connected to the polarization channel input of the polarization feature fusion module 3. The output of the multi-scale feature extraction module 2 is connected to the intensity feature input of the polarization feature fusion module 3. The output of the polarization feature fusion module 3 is connected to the input of the sea clutter adaptive suppression module 4. The output of the sea clutter adaptive suppression module 4 is connected to the input of the multi-level attention enhancement module 5. The output of the multi-level attention enhancement module 5 is connected to the input of the target localization and classification module 6. The output of the target localization and classification module 6 is connected to the input of the post-processing optimization module 7. The post-processing optimization module 7 outputs the final ship target detection result.

[0079] The SAR image preprocessing unit 1 receives raw SAR image data and performs radiometric calibration, geometric correction, noise reduction, and polarization channel separation. The multi-scale feature extraction module 2 extracts image features at different scales to meet the needs of multi-scale target detection. The polarization feature fusion module 3 extracts and fuses the polarization features of SAR images to enhance the distinction between targets and background. The adaptive sea clutter suppression module 4 adaptively suppresses sea clutter according to local sea conditions to reduce the false alarm rate. The multi-level attention enhancement module 5 highlights target features and suppresses background interference through an attention mechanism. The target localization and classification module 6 accurately locates the target bounding box and classifies ship types. The post-processing optimization module 7 removes redundant detections and false detections and outputs the final result.

[0080] In this detection model, the raw SAR image data first undergoes radiometric calibration, geometric correction, and noise reduction in the SAR image preprocessing unit 1, while simultaneously separating the data from each polarization channel. The preprocessed image then enters the multi-scale feature extraction module 2, which extracts multi-level, multi-scale feature representations. The multi-scale intensity features and polarization channel data are jointly input into the polarization feature fusion module 3, which generates an enhanced feature map through polarization decomposition and feature fusion. The enhanced feature map then enters the sea clutter adaptive suppression module 4, which adaptively suppresses sea clutter interference based on local sea state statistical characteristics. The suppressed feature map then passes through the multi-level attention enhancement module 5, which further highlights target features through spatial and channel attention mechanisms. The enhanced features are input into the target localization and classification module 6, which generates candidate detection boxes and category predictions. Finally, the post-processing optimization module 7 performs confidence filtering, non-maximum suppression, and scale consistency verification on the detection results, outputting the final ship target detection result.

[0081] The embodiments described above are merely one implementation method of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A novel high-precision synthetic aperture radar method for detecting ship targets, characterized in that, Includes the following steps: Step 1: Acquire raw SAR image data, including intensity images and polarization channel data, and preprocess the SAR images; Step 2: Extract shallow, medium, and deep multi-scale features of the image using the multi-scale feature extraction module, and then perform pyramid fusion. Step 3: Extract the polarization features of the SAR image through the polarization feature fusion module and fuse them with multi-scale features; Step 4: Estimate the local sea state using the sea clutter adaptive suppression module, establish a clutter statistical model, and adaptively suppress sea clutter; Step 5: Generate spatial attention and channel attention weights through a multi-level attention enhancement module to enhance target features; Step 6: Generate multi-scale anchor boxes through the target localization and classification module, perform feature alignment, and predict bounding box coordinates and class probabilities; Step 7: Perform confidence screening, non-maximum suppression, and scale consistency verification through the post-processing optimization module; Step 8: Output the final ship target detection results, including bounding box coordinates, category labels, and confidence scores.

2. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The multi-scale feature extraction module in step 2 includes a backbone feature extraction network, shallow feature branches, medium feature branches, deep feature branches, and a feature pyramid fusion unit; The backbone feature extraction network employs an improved residual network structure to extract the basic features of SAR images. The shallow feature branch extracts detailed features of the image for detecting small-scale ship targets; the middle feature branch extracts intermediate semantic features of the image for detecting medium-scale ship targets; the deep feature branch extracts high-level semantic features of the image for detecting large-scale ship targets; the feature pyramid fusion unit fuses features from different levels through top-down and bottom-up feature fusion paths to generate a multi-scale fused feature map.

3. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The polarization feature fusion module in step 3 includes a polarization channel separation unit, a polarization decomposition unit, a polarization feature encoding unit, and a multimodal feature fusion unit; Specifically, the polarization channel separation unit separates the fully polarimetric SAR image into four polarization channels: HH, HV, VH, and VV; the polarization decomposition unit uses Pauli decomposition or HA-Alpha decomposition to extract polarization scattering characteristic parameters; the polarization feature encoding unit encodes polarization features through convolutional layers and normalization layers to generate polarization feature vectors; and the multimodal feature fusion unit uses an attention-weighted fusion mechanism to adaptively fuse the multi-scale intensity features and polarization features obtained in step 2 to enhance the contrast between the target and sea clutter.

4. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The sea clutter adaptive suppression module in step 4 includes a sea state estimation unit, a clutter statistical modeling unit, an adaptive filtering unit, and a feature enhancement unit. The sea state estimation unit estimates the current sea state level and wind speed through local window statistical analysis; the clutter statistical modeling unit establishes a K-distribution or G0-distribution sea clutter statistical model based on the sea state parameters; the adaptive filtering unit suppresses sea clutter and retains target information by using an adaptive threshold segmentation method based on the clutter model parameters; and the feature enhancement unit enhances the saliency of target features through residual connection and feature normalization.

5. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 4, characterized in that, The adaptive threshold is set as: T = μ + k×σ, where μ is the local mean, σ is the local standard deviation, and k is adaptively adjusted according to the sea state level. The higher the sea state level, the larger the value of k.

6. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The multi-level attention enhancement module in step 5 includes: a spatial attention submodule, a channel attention submodule, a multi-scale perception unit, and a feature weighted fusion unit; The spatial attention submodule generates a spatial attention weight map through global pooling and convolution operations to highlight the spatial location of the target; the channel attention submodule generates a channel attention weight vector through global average pooling and fully connected layers to strengthen the feature channels related to the target; the multi-scale perception unit uses dilated convolution and a multi-branch structure to perceive target features at different receptive field scales; and the feature weighted fusion unit weights and fuses spatial attention features and channel attention features to generate an attention-enhanced feature map.

7. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The target localization and classification module in step 6 includes: an anchor box generation unit, a feature alignment unit, a bounding box regression unit, and a target classification unit; The anchor frame generation unit generates candidate anchor frames with multiple scales and aspect ratios based on the feature map size and target scale distribution; the feature alignment unit uses deformable convolution or RoI Align method to extract features for each candidate anchor frame region; the bounding box regression unit predicts the position offset of the anchor frame through a fully connected layer to accurately regress the target bounding box coordinates; and the target classification unit outputs the class probability of each anchor frame, including ship type and background category, through a fully connected layer and Softmax activation function.

8. The novel high-precision synthetic aperture radar method for detecting ship targets according to claim 1, characterized in that, The post-processing optimization module includes a confidence screening unit, a non-maximum suppression unit, a scale consistency verification unit, and a result output unit. The confidence screening unit filters out low-confidence detection boxes based on a preset confidence threshold; the non-maximum suppression unit removes redundant detection boxes with high overlap based on an IoU threshold, retaining the detection results with the highest confidence; the scale consistency verification unit verifies the consistency between the scale of the detection boxes and the prior scale of the ship, removing unreasonable detection results; and the result output unit integrates the position coordinates, category labels, and confidence scores of the detection boxes to output the final ship target detection result.

9. A novel high-precision synthetic aperture radar (SAR) system for detecting ship targets, characterized in that, Includes the following modules: SAR image preprocessing unit: used to acquire raw SAR image data, including intensity image and polarization channel data, and to preprocess the SAR image; Multi-scale feature extraction module: used to extract shallow, medium and deep multi-scale features of the image and perform pyramid fusion; Polarization feature fusion module: used to extract polarization features from SAR images and fuse them with multi-scale features; Adaptive Sea Clutter Suppression Module: Used to estimate local sea state, establish a statistical model of clutter, and adaptively suppress sea clutter; Multi-level attention enhancement module: used to generate spatial attention and channel attention weights to enhance target features; The target localization and classification module is used to generate multi-scale anchor boxes, perform feature alignment, and predict bounding box coordinates and class probabilities. Post-processing optimization module: used for confidence filtering, non-maximum suppression and scale consistency verification, outputting the final ship target detection results, including bounding box coordinates, category labels and confidence scores.

Images