A scattering characteristic perception full polarization sar ship detection method

Abstract

The application discloses a scattering characteristic perception full-polarization SAR ship detection method, which comprises the following steps: S1, four-component decomposition is performed on full-polarization SAR data to obtain specular scattering, double-bounce scattering, ±45° dipole and asymmetric scattering components, and then a feature vector is constructed by using the double-bounce scattering, ±45° dipole and asymmetric scattering components as input data of a ship detection network; S2, a scattering characteristic perception module is constructed according to characteristic differences of targets under different scattering mechanisms to perceive contribution rates of different scattering mechanisms to the targets; and S3, a scattering characteristic perception ship detection network is used to detect targets in a full-polarization SAR image. The scattering characteristic perception module is designed in combination with characteristic differences of targets under different scattering mechanisms, and the module can effectively perceive contribution rates of different scattering mechanisms to the targets, so that the effects of suppressing clutters and highlighting targets are achieved, and the detection precision of the targets is improved.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence and synthetic aperture radar target detection technology, and more specifically, to a fully polarimetric SAR ship detection method based on scattering characteristics. Background Technology

[0002] With the development of satellite technology, Synthetic Aperture Radar (SAR) plays a vital role in Earth observation due to its unique all-weather, all-day advantage. Because of its crucial role in studying surface changes, various countries have launched SAR satellites, such as China's Gaofen-3, Japan's ALOS, Europe's Sentinel-1, and Canada's RADARSAT-2. Using SAR to observe maritime targets is an effective method, and ship detection is a very important topic. Accurate ship position data can help governments address issues such as illegal fishing and maritime traffic safety. However, detecting ships using SAR imagery is a complex task. This complexity is related to the complexity of the marine environment and the ship's structure.

[0003] In recent years, convolutional neural networks (CNNs) have been widely used in computer vision due to their powerful feature extraction capabilities. In object detection, a large number of excellent CNN-based detection algorithms have emerged, mainly divided into three categories: 1) Single-stage object detection: Compared to two-stage object detection algorithms, single-stage algorithms directly calculate and generate detection results from the image, resulting in high detection speed but low accuracy. Representative single-stage object detection algorithms include CenterNet, YOLO, SSD, RetinaNet, and improved versions of YOLO. 2) Two-stage object detection algorithms: Compared to single-stage object detection algorithms, two-stage algorithms first extract candidate boxes from the image, and then perform secondary corrections based on the candidate regions to obtain the detection point results, resulting in higher detection accuracy but slower detection speed. Representative algorithms of this type mainly include R-CNN, Faster R-CNN, and Mask R-CNN. In the field of SAR image object detection, many CNN-based object detection algorithms have also emerged. However, limited by the dataset itself, existing neural networks only focus on the model itself, that is, only improving the model structure, while ignoring the characteristics of the SAR data itself. For example, the SSDD dataset, LS_SSDD dataset, and OpenSARShip dataset are all composed of single-channel intensity values.

[0004] As is well known, fully polarimetric SAR (WPSAR) data contains rich backscattering information, which can more comprehensively describe the scattering characteristics of ships. Therefore, there is an urgent need for a method that can combine the scattering characteristic data in WPSAR images with neural networks, and fully utilize the powerful feature mining capabilities of neural networks to extract the target information contained in the WPSAR data, thereby improving the accuracy of target detection. Summary of the Invention

[0005] The purpose of this invention is to provide a fully polarimetric SAR ship detection method based on scattering characteristics, so as to overcome the defects of the existing technology.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A method for detecting ships using fully polarimetric SAR based on scattering characteristics includes the following steps:

[0008] S1. Decompose the fully polarimetric SAR data into four components to obtain specular scattering, double-hop scattering, ±45° dipole and asymmetric scattering components. Then, use the double-hop scattering, ±45° dipole and asymmetric scattering components to construct a feature vector as the input data for the ship detection network.

[0009] S2. A scattering characteristic perception module, constructed based on the differences in characteristics exhibited by the target under different scattering mechanisms, perceives the contribution rate of different scattering mechanisms to the target and uses the contribution rate as the weight of each pixel in the ship detection network.

[0010] S3. Targets in fully polarimetric SAR images are detected using a ship detection network that senses scattering characteristics.

[0011] Furthermore, the formula for decomposing the fully polarimetric SAR data into four components in step S1 is as follows:

[0012]

[0013] In the formula, f s f d f od and f Asym The expansion coefficient is to be determined, and the specular scattering power P s =f s (1+|β| 2 Double-hop scattering power P d =f d (1+|α| 2 Volume scattering power P Od =2f Od Asymmetric scattering power

[0014]

[0015] Furthermore, the formula for constructing a feature vector as the input data of the network using double-hop scattering, ±45° dipoles, and asymmetric scattering components is as follows:

[0016] Input = [Dbl 512×512 Od 512×512 Asym 512×512 ]

[0017] In the formula, Dbl, Od, and Asym represent the power values ​​of the double-hop scattering, ±45° dipole, and asymmetric scattering components, respectively.

[0018] Further, step S2 includes:

[0019] S20. Segment the feature map and use convolution kernels of different sizes to capture multi-scale information of targets on different feature maps;

[0020] S21. Use the channel attention module to extract the channel attention weights of feature maps at different scales to obtain the channel attention vector at each different scale.

[0021] S22. Use the Softmax function to recalibrate the features of the multi-scale channel attention vector to obtain the attention weights of all channels in the new multi-scale feature map.

[0022] S23. Perform element-wise dot product on the new weights and the feature map that has not been processed by the channel attention module, and output a feature map after attention weighting of multi-scale feature information.

[0023] S24. Add the weighted feature map to the unprocessed feature map to obtain a new feature map with shallow texture features and deep semantic features.

[0024] Furthermore, in step S3, a ship detection network with scattering characteristics perception is constructed using ResNet50 as the backbone network and a feature pyramid network with a scattering characteristic perception module as the feature fusion layer.

[0025] Compared with the prior art, the advantages of the present invention are as follows: The present invention provides a scattering characteristic sensing method for ship detection of fully polarimetric SAR. Combining the characteristic differences of the target under different scattering mechanisms, a scattering characteristic sensing module is designed. Using this module, the contribution rate of the target under different scattering mechanisms can be effectively sensed, so as to suppress clutter and highlight the target, thereby improving the accuracy of target detection. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention 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 only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0027] Figure 1 These are visualizations of ships under different scattering mechanisms.

[0028] Figure 2 This is a flowchart of the fully polarimetric SAR ship detection method based on scattering characteristics of the present invention.

[0029] Figure 3 This is a diagram of the scattering characteristic sensing module of the present invention.

[0030] Figure 4 These are two different pyramid structures characteristic of this invention.

[0031] Figure 5 This invention relates to a fully polarimetric SAR detection network structure for sensing scattering characteristics. Detailed Implementation

[0032] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0033] See Figure 1 As shown, this embodiment discloses a fully polarimetric SAR ship detection method based on scattering characteristics, including the following steps:

[0034] Step S1: Decompose the fully polarimetric SAR data into four components to obtain specular scattering, double-hop scattering, ±45° dipole and asymmetric scattering components. Then, use the double-hop scattering, ±45° dipole and asymmetric scattering components to construct a feature vector as input data for the ship detection network.

[0035] Specifically, step S1 includes the following steps:

[0036] Step S10: Obtain the corresponding coherence matrix [T] using the scattering matrix [S] of the fully polarimetric SAR data, and then perform four-component decomposition using the coherence matrix [T]. The decomposition expression is as follows:

[0037]

[0038] In the formula, f s f d f od and f AsymThe expansion coefficient is to be determined, and the specular scattering power P s =f s (1+|β| 2 Double-hop scattering power P d =f d (1+|α| 2 Volume scattering power P Od =2f Od Helical scattering power

[0039]

[0040] Step S11: Construct a feature vector using double-hop scattering, ±45° dipoles, and asymmetric scattering components as the input data for the network. The corresponding expression is as follows:

[0041] Input = [Dbl 512×512 Od 512×512 Asym 512×512 ]

[0042] In the formula, Dbl, Od, and Asym represent the power values ​​of the double-hop scattering, ±45° dipole, and asymmetric scattering components, respectively.

[0043] Step S2: Based on the differences in the characteristics of the target under different scattering mechanisms, the scattering characteristic perception module perceives the contribution rate of different scattering mechanisms to the target, and uses the contribution rate as the weight of each pixel in the ship detection network.

[0044] like Figure 1 As shown, the visual effects of a target vary significantly under different scattering mechanisms. In other words, the characteristics of the target, such as texture and shape, differ depending on the scattering mechanism. Therefore, based on this characteristic, a scattering characteristic sensing module was designed. Figure 2 As shown, step S2 specifically includes:

[0045] Step S20: Segment the feature maps, use convolution kernels of different sizes to capture multi-scale information of the target on different feature maps, and then reconnect all the feature maps.

[0046] Step S21: Use the channel attention module (which is a structure in a neural network) to extract the channel attention weights of feature maps at different scales, and obtain the channel attention vector at each different scale. The function of this module is to obtain the channel attention vector at different scales.

[0047] Step S22: Use the Softmax function to recalibrate the features of the multi-scale channel attention vector to obtain the attention weights of all channels of the new multi-scale feature map.

[0048] Step S23: Perform element-wise dot product operation on the new weights and the feature map that has not been processed by the channel attention module, and output a feature map after attention weighting of multi-scale feature information.

[0049] Step S24: Add the weighted feature map and the unprocessed feature map to obtain a new feature map with shallow texture features and deep semantic features.

[0050] like Figure 3 As shown, a scattering characteristic perception module is inserted into the feature pyramid to perceive the scattering characteristics of the target under different feature maps. More importantly, this module can effectively fuse shallow detail information and deep semantic information in the neural network to fully explore the target's characteristics.

[0051] Step S3: Use a ship detection network that senses scattering characteristics to detect targets in the fully polarimetric SAR image.

[0052] like Figure 4 As shown, the scattering characteristic-aware fully polarimetric SAR ship detection network uses ResNet50 as the backbone network and a feature pyramid network with a scattering characteristic awareness module as the feature fusion layer. This network can fully mine the scattering information of targets under different scattering mechanisms, thereby improving the target detection accuracy.

[0053] This embodiment utilizes scattering characteristic data from fully polarimetric SAR (WPSAR) data as network input. As mentioned earlier, few have attempted to leverage the scattering information contained in WPSAR data and combine it with neural networks for ship detection. Therefore, this embodiment uses double-hop scattering, ±45° dipole scattering, and asymmetric scattering as neural network inputs. Based on the differences in target characteristics exhibited under these three scattering mechanisms, a scattering characteristic perception module is designed, and a scattering characteristic perception-based WPSAR ship detection network is proposed. This network can fully extract scattering information from the target and effectively fuse shallow detail information and deep semantic information from the neural network, thereby improving the accuracy of target detection.

[0054] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, the patent owner may make various modifications or alterations within the scope of the appended claims, as long as they do not exceed the protection scope described in the claims of the present invention, they shall be within the protection scope of the present invention.

Claims

1. A method for detecting fully polarimetric SAR ships based on scattering characteristics, characterized in that, Includes the following steps: S1. Decompose the fully polarimetric SAR data into four components to obtain specular scattering, double-hop scattering, ±45° dipole and asymmetric scattering components. Then, use the double-hop scattering, ±45° dipole and asymmetric scattering components to construct a feature vector as the input data for the ship detection network. S2. A scattering characteristic perception module, constructed based on the differences in characteristics exhibited by the target under different scattering mechanisms, perceives the contribution rate of different scattering mechanisms to the target and uses the contribution rate as the weight of each pixel in the ship detection network. S3. Targets in fully polarimetric SAR images are detected using a ship detection network that senses scattering characteristics. Step S2 includes: S20. Segment the feature map and use convolution kernels of different sizes to capture multi-scale information of targets on different feature maps; S21. Use the channel attention module to extract the channel attention weights of feature maps at different scales to obtain the channel attention vector at each different scale. S22. Use the Softmax function to recalibrate the features of the multi-scale channel attention vector to obtain the attention weights of all channels of the new multi-scale feature map. S23. Perform element-wise dot product on the new weights and the feature map that has not been processed by the channel attention module, and output a multi-scale feature map after attention weighting. S24. Add the weighted feature map to the unprocessed feature map to obtain a new feature map with shallow texture features and deep semantic features.

2. The fully polarimetric SAR ship detection method based on scattering characteristics as described in claim 1, characterized in that, The formula for decomposing the fully polarimetric SAR data into four components in step S1 is as follows: In the formula, , , and The expansion coefficient is to be determined, and the specular scattering power is... Double-hop scattering power Volume scattering power Asymmetric scattering power .

3. The fully polarimetric SAR ship detection method based on scattering characteristics as described in claim 1, characterized in that, The formula for constructing a feature vector as the input data of the network using double-hop scattering, ±45° dipoles, and asymmetric scattering components is as follows: In the formula, They represent double-hop scattering, Power values ​​of the 45° dipole and asymmetric scattering components.

4. The fully polarimetric SAR ship detection method based on scattering characteristics as described in claim 1, characterized in that, In step S3, a ship detection network with scattering characteristics perception is constructed using ResNet50 as the backbone network and a feature pyramid network with a scattering characteristic perception module as the feature fusion layer.

Images