Remote sensing image robust watermarking method, device and equipment

Abstract

The invention provides a remote sensing image robust watermarking method, device and equipment, and relates to the technical field of computers, and the method comprises the steps: carrying out the feature extraction of an original image, mapping the extracted image features to a potential feature space through an adapter network, and obtaining the corresponding potential features, and fusing the potential features and watermark information, inputting the fused potential features and watermark information into a generator network, generating a watermark image in combination with a U-shaped structure and a semi-wavelet attention mechanism, performing random distortion simulation on the watermark image through signal processing and geometric transformation, and analyzing distortion features of a distorted image to obtain watermark information. According to the method, an adapter structure is introduced in front of the generator, and the convolution optimization generator is used to pay more attention to effective features, so that the network training cost can be effectively reduced; a semi-wavelet attention module is introduced into the generator, and spatial domain and frequency domain features of the image are fully combined, so that the influence of the watermark on the image quality is reduced, and the robustness of the watermark to complex attacks is improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a remote sensing image robust watermark method, device and equipment. Background Art

[0002] As an important means of obtaining high-resolution images of the earth's surface, remote sensing technology plays a key role in environmental monitoring, disaster warning, land use planning and other fields. However, with the continuous expansion of the application of remote sensing images, the copyright protection and information security issues have become increasingly prominent. Unauthorized copying and use not only infringe intellectual property rights, but may even threaten national security. As an effective means of copyright protection and integrity assurance, digital watermarking technology plays an indispensable role in the field of geographic data. Digital watermarking technology cleverly embeds copyright information into remote sensing images, realizing the close combination of watermarks and remote sensing images. Then, through the extraction and verification of watermarks, the copyright ownership of remote sensing images can be proved and their intellectual property rights can be protected, thereby preventing unauthorized copying and abuse.

[0003] Traditional remote sensing image copyright protection methods mainly rely on encryption technology and digital signature technology to maintain copyright by controlling access rights and preventing illegal dissemination. However, such methods have significant limitations: once the image is decrypted or the signature is tampered with, the original copyright mark is easily destroyed, making it difficult to effectively trace the ownership. At the same time, these methods are prone to failure when facing operations such as image compression and cutting, and cannot guarantee the integrity of copyright information. In contrast, digital watermarking technology based on deep learning embeds copyright information through an end-to-end learning method, showing high robustness. However, due to the complex texture of remote sensing images, such methods are easily disturbed by invalid features during the processing process, resulting in uneven watermark embedding and artifacts. At the same time, the computational cost of the network is increased during the training process due to the influence of invalid features, which significantly reduces the efficiency of actual applications. In addition, many existing methods only process single features in the spatial domain or frequency domain, and fail to achieve effective fusion of the two, further limiting their robustness. Therefore, how to ensure the integrity and stability of watermark information under complex attacks and noise interference has become one of the core issues that need to be urgently solved in current digital watermarking technology. Summary of the invention

[0004] The purpose of the present invention is to solve the problem that it is difficult to ensure the integrity and stability of watermark information under complex attacks and noise interference. The present invention provides a remote sensing image robust watermarking method, device and equipment.

[0005] The technical solution of the embodiment of the present application is implemented as follows: A first aspect of an embodiment of the present application provides a remote sensing image robust watermarking method, comprising: Extract features from the original image, map the extracted image features to the latent feature space through the adapter network, and obtain the corresponding latent features; The latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism; The watermark image is subjected to random distortion simulation through signal processing and geometric transformation, and the distortion characteristics of the distorted image are analyzed to obtain watermark information.

[0006] Optionally, the feature extraction of the original image and mapping the extracted image features to a latent feature space through an adapter network to obtain corresponding latent features include: Performing feature extraction on the original image through convolution operation, batch normalization and ReLU activation function to obtain shallow features; The shallow features are passed through a depth-wise separable convolution module and a point-by-point convolution module to extract the global information of each channel, and residual connections are performed to obtain corresponding residual features; The residual feature is input into a parallel dual attention unit, and the spatial feature and the channel feature of the residual feature are enhanced and adjusted respectively to obtain a fusion feature; The fused features are subjected to convolution processing, average pooling processing and full connection layer processing, and mapped to a potential feature space to obtain corresponding potential features.

[0007] Optionally, the feature extraction of the original image and mapping the extracted image features to a latent feature space through an adapter network to obtain corresponding latent features include: Two convolutional layers and a Sigmoid activation function are used to generate activation values, and the activation values ​​are used to scale the original feature map channel by channel to obtain a channel attention map; Perform global average pooling and maximum pooling operations on the original feature map along the spatial dimension, concatenate the pooling results and pass them through a convolutional layer and a Sigmoid activation function to generate a spatial attention map; The channel attention map and the spatial attention map are fused to obtain a fusion feature.

[0008] Optionally, the fusion of the latent features and the watermark information and inputting them into the generator network, and combining the U-shaped structure and the half-wavelet attention mechanism to generate the watermark image, include: The fused features are input into a downsampling module, and the spatial resolution of the feature map is gradually reduced through a maximum pooling operation to generate multi-level downsampling features; The downsampled features are transferred to the current upsampling layer through a skip connection, the spatial resolution of the feature map is gradually restored using bilinear interpolation, and the final feature map is output; A target watermark image is generated based on the final feature map.

[0009] Optionally, the fusion of the latent features and the watermark information and inputting them into the generator network, and combining the U-shaped structure and the half-wavelet attention mechanism to generate the watermark image, include: The feature map is subjected to depthwise separable convolution and pointwise convolution to extract high-dimensional semantic features, and residual connections are introduced to add the extracted features to the input features; Half-wavelet attention is used to enhance feature representation. The feature map is divided into two parts along the channel dimension. One part retains the original features, and the other part is subjected to discrete wavelet transform. The converted wavelet domain features are input into the dual attention module to obtain weighted wavelet features. The inverse wavelet transform is used to resize the features to their original size, and the wavelet domain features are concatenated with the retained original feature parts. The multi-level feature information is fused to obtain the fused features and generate the watermarked image.

[0010] Optionally, the signal processing and geometric transformation include: Gaussian noise processing, speckle noise processing, stripe noise processing, JPEG compression, image rotation, random cropping and resizing.

[0011] Optionally, the performing random distortion simulation on the watermark image through signal processing and geometric transformation, analyzing the distortion characteristics of the distorted image, and obtaining the watermark information includes: The distorted image obtained by the distortion simulation is subjected to convolution operation, batch normalization and ReLU activation function processing to extract shallow features; The shallow features are extracted through depth-wise separable convolution and point-by-point convolution to obtain global information of each channel, and residual connections are performed to obtain residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features respectively, and then fuse them to obtain fused features; The fused features are processed through convolution, average pooling and full connection layers to obtain watermark information.

[0012] A second aspect of the embodiment of the present application provides a remote sensing image robust watermarking device, comprising: a feature mapping module, a feature fusion module and a distortion simulation module, wherein: The feature mapping module is configured to extract features from the original image, map the extracted image features to a latent feature space through an adapter network, and obtain corresponding latent features; The feature fusion module is configured to fuse the latent features with the watermark information and input them into the generator network, and generate a watermark image by combining the U-shaped structure and the half-wavelet attention mechanism; The distortion simulation module is configured to perform random distortion simulation on the watermark image through signal processing and geometric transformation, analyze the distortion characteristics of the distorted image, and obtain watermark information.

[0013] A third aspect of an embodiment of the present application provides an electronic device, comprising a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the remote sensing image robust watermarking method described in the first aspect.

[0014] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in the first aspect.

[0015] Compared with the prior art, the technical solution provided by this application has the following beneficial effects: The present invention provides a remote sensing image robust watermarking method, device and equipment, which can effectively reduce the network training cost by introducing an adapter structure before a generator and using a convolution optimization generator to pay more attention to effective features; a half-wavelet attention module is introduced in the generator to fully combine the spatial domain and frequency domain features of the image, thereby reducing the impact of the watermark on the image quality and improving the robustness of the watermark against complex attacks. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process flow of a remote sensing image robust watermarking method provided in an embodiment of the present application; Figure 2 A schematic diagram of an image processing network model provided in an embodiment of the present application; Figure 3 A schematic diagram of a network model of an adapter and an extractor module provided in an embodiment of the present application; Figure 4 A schematic diagram of the network architecture of a dual attention unit (DAU) provided in an embodiment of the present application; Figure 5 A schematic diagram of the network architecture of the generator module provided in an embodiment of the present application; Figure 6 A schematic diagram of the structure of a remote sensing image robust watermarking device provided in an embodiment of the present application; Figure 7 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0017] Below, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present application. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present application. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application.

[0018] The terms used herein are only for describing specific embodiments and are not intended to limit the present application. The terms "include", "comprising", etc. used herein indicate the existence of the features, steps, operations and / or components, but do not exclude the existence or addition of one or more other features, steps, operations or components. All terms (including technical and scientific terms) used herein have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification, and should not be interpreted in an idealized or overly rigid manner.

[0019] Some block diagrams and / or flow charts are shown in the accompanying drawings. It should be understood that some blocks or combinations thereof in the block diagrams and / or flow charts may be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that these instructions, when executed by the processor, may create a device for implementing the functions / operations described in these block diagrams and / or flow charts.

[0020] In some embodiments, see Figure 1 and Figure 2 , Figure 1 A schematic diagram of a process flow of a remote sensing image robust watermarking method provided in an embodiment of the present application. Figure 2 A schematic diagram of an image processing network model provided in an embodiment of the present application; a remote sensing image robust watermarking method provided in an embodiment of the present application includes: S110, extracting features from the original image, and mapping the extracted image features to a latent feature space through an adapter network to obtain corresponding latent features.

[0021] Here, the original image is the original remote sensing image. Before extracting features from the original image, the original image can also be preprocessed, such as image enhancement, filtering, etc., and then the extracted image features are mapped to the latent feature space through the adapter network. Please refer to Figure 3 , Figure 3 A schematic diagram of the network model of the adapter and extractor modules provided in an embodiment of the present application.

[0022] In an optional embodiment, the original remote sensing image is adjusted to a tensor of shape Batch_Size×H×W×3. Subsequently, a 3×3 convolution kernel is used to perform a convolution operation on the image, combined with batch normalization and ReLU activation function to extract shallow features. On this basis, the shallow features are input into the depthwise separable convolution module for processing, and the processed output features are residually connected with the input features of the module, and then the results are passed to the point-by-point convolution module. Finally, the feature map is input into the parallel dual attention unit DAU, and the spatial features and channel features are respectively enhanced and adjusted by the spatial attention module SA and the channel attention module CA, and finally fused; please refer to Figure 4 , Figure 4 A schematic diagram of the network architecture of the dual attention unit DAU provided in the embodiment of the present application. On this basis, the fused features are mapped to the potential feature space through a 3×3 convolution layer, an average pooling layer, and a fully connected layer.

[0023] In some embodiments, S110, feature extraction is performed on the original image, and the extracted image features are mapped to a latent feature space through an adapter network to obtain corresponding latent features, including: The original image is subjected to feature extraction through convolution operation, batch normalization and ReLU activation function to obtain shallow features; The shallow features are passed through the depth-wise separable convolution module and the point-by-point convolution module to extract the global information of each channel, and the residual connection is performed to obtain the corresponding residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features of the residual features respectively to obtain the fused features; The fused features are processed by convolution, average pooling and full connection layer, and mapped to the latent feature space to obtain the corresponding latent features.

[0024] In some embodiments, S110, feature extraction is performed on the original image, and the extracted image features are mapped to a latent feature space through an adapter network to obtain corresponding latent features, including: Two convolutional layers and a Sigmoid activation function are used to generate activation values, and the activation values ​​are used to scale the original feature map channel by channel to obtain the channel attention map; Perform global average pooling and maximum pooling operations on the original feature map along the spatial dimension, concatenate the pooling results and pass them through the convolution layer and Sigmoid activation function to generate a spatial attention map; The channel attention map and the spatial attention map are fused to obtain the fused features.

[0025] In this embodiment, the spatial features and channel features are enhanced and adjusted, and finally fused, which may specifically include the following steps: 1) Calculating the channel attention map, for the feature map , use the squeeze operation to perform global average pooling on the spatial dimension to encode the global context, thereby obtaining the feature descriptor . Use the excitation operation to pass d through two convolutional layers, and then generate activation values ​​through the sigmoid activation function , by using the activation value Rescale the feature map F to get the channel attention map; 2): Calculate the spatial attention map and independently perform attention on the features in the channel dimension Apply global average pooling and maximum pooling operations, and concatenate the outputs of the two to form a feature map The feature map is processed by convolution and sigmoid activation function to obtain the spatial attention map , and finally use the attention map to recalibrate the input feature F; 3) Fuse the channel attention map and the spatial attention map to obtain the final feature weight distribution.

[0026] S120, the latent features are fused with the watermark information and input into the generator network, and the watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism.

[0027] Here, the latent features are fused with the watermark information and then input into the generator network, and the watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism. Figure 5 , Figure 5 A schematic diagram of the network architecture of the generator module provided in an embodiment of the present application.

[0028] In some embodiments, S120, the latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism, including: The fused features are input into the downsampling module, and the spatial resolution of the feature map is gradually reduced through the maximum pooling operation to generate multi-level downsampling features; The downsampled features are passed to the current upsampling layer through skip connections, and the spatial resolution of the feature map is gradually restored using bilinear interpolation to output the final feature map; Generate the target watermark image based on the final feature map.

[0029] In this embodiment, the fusion feature Input downsampling module, through the maximum pooling operation, the spatial resolution of the feature map can be gradually reduced to obtain the downsampled feature ; Downsample the features Input upsampling module, pass the corresponding downsampled features directly to the current upsampling layer through skip connection, gradually use bilinear interpolation to restore the spatial resolution of the feature map, and output the final feature map ; The original remote sensing image With feature map After adding, we get the target watermark image , which can be expressed as: . And use image loss To optimize the generator performance, as follows: ,in, represents the potential features generated by the adapter, Indicates watermark information. Represents the parameters of the generator.

[0030] In some embodiments, S120, the latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism, including: The feature map is subjected to depthwise separable convolution and pointwise convolution to extract high-dimensional semantic features, and residual connections are introduced to add the extracted features to the input features; Half-wavelet attention is used to enhance feature representation. The feature map is divided into two parts along the channel dimension. One part retains the original features, and the other part is subjected to discrete wavelet transform. The converted wavelet domain features are input into the dual attention module to obtain weighted wavelet features. The inverse wavelet transform is used to resize the features to their original size, and the wavelet domain features are concatenated with the retained original feature parts. The multi-level feature information is fused to obtain the fused features and generate the watermarked image.

[0031] In this embodiment, the feature map is subjected to depthwise separable convolution and pointwise convolution to extract high-dimensional semantic features, and residual connection is introduced to add the extracted features to the input features; half wavelet attention is used to enhance feature representation, and the input features are added to the residual connection. Along the channel dimension, it is divided into and Two parts. The original features are retained. Perform discrete wavelet transform to obtain wavelet domain features Then the wavelet domain features are input into the dual attention module DAU to obtain the weighted wavelet features , and through inverse wavelet transform Re-adjusted to The shape is consistent, becoming the weighted normal domain feature Finally, the feature and Features Connect and obtain the residual features The processed features are then combined with the original features through 3×3 convolution and ReLU activation function. The results after point convolution are connected to obtain the final output features . The SCConv convolution module is used, and batch normalization and ReLU activation function are combined to enhance feature expression capabilities.

[0032] S130, performing random distortion simulation on the watermark image through signal processing and geometric transformation, analyzing the distortion characteristics of the distorted image, and obtaining watermark information.

[0033] In this embodiment, the watermark image can be convolved to extract preliminary features, and then the features are further extracted through the SCConv convolution module, and the extracted feature map is sequentially input into the convolution layer, average pooling layer and fully connected layer to calculate the similarity probability score between the watermark image and the original image. To optimize the performance of the adversary and the generator, as follows: ; in, represents the original image, represents the watermark image, and They represent the parameters of the discriminator and the generator respectively.

[0034] In some embodiments, the signal processing and geometric transformation include: Gaussian noise processing, speckle noise processing, stripe noise processing, JPEG compression, image rotation, random cropping and resizing.

[0035] For example, during the training process, the selected Gaussian noise standard deviation ranges from 0.03 to 0.05, the speckle noise density ranges from 0.01 to 0.03, the stripe noise intensity ranges from 0.05 to 0.1, and the JPEG compression quality factor ranges from 30 to 70. The image rotation angle ranges from ±30°, the random cropping size ranges from 80% to 100%, and the resizing ratio ranges from 0.8 to 1.2.

[0036] In some embodiments, S130, performing random distortion simulation on the watermark image through signal processing and geometric transformation, analyzing the distortion characteristics of the distorted image, and obtaining watermark information includes: The distorted image obtained by the distortion simulation is subjected to convolution operation, batch normalization and ReLU activation function processing to extract shallow features; The shallow features are extracted through depth-wise separable convolution and point-by-point convolution to obtain the global information of each channel, and residual connections are performed to obtain residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features respectively, and then fuse them to obtain the fused features; The fused features are processed through convolution, average pooling and fully connected layers to obtain watermark information.

[0037] In an optional embodiment, the initial remote sensing image is adjusted to a tensor of shape Batch_Size×HxW×3, and then a 3×3 convolution kernel is used to perform a convolution operation on the image, combined with batch normalization and ReLU activation function to extract shallow features. The shallow features are input into the depth separable convolution module for processing, and the processed output features are residually connected with the input features of the module, and then the results are passed to the point-by-point convolution module. The feature map is input into the parallel dual attention unit DAU, and the spatial features and channel features are respectively enhanced and adjusted by the spatial attention module SA and the channel attention module CA, and finally fused. The fused features are processed through a 3×3 convolution layer, average pooling, and a fully connected layer to output the extracted watermark information.

[0038] In an example, the process of extracting watermark information by the extractor can be expressed as: , and use information loss To optimize the extractor performance, as follows: ; in, Represents the original watermark information, Represents the extracted watermark information, represents a distorted image, Represents the parameters of an extractor.

[0039] The present invention introduces an adapter structure before the generator for remote sensing image watermarking, maps the original image features to the latent feature space, and shares the structure and parameters with the extractor, effectively reducing the redundant feature embedding in the generator and enhancing the concealment and anti-attack ability of the watermark. The generator adopts a U-shaped network structure, combined with deep separable convolution and SCConv convolution, which not only improves the extraction ability of complex texture features, but also significantly reduces the computational cost and the impact of redundant features on training, thereby shortening the network training time. In order to solve the problem that watermarks are prone to introduce artifacts, the generator further introduces a half-wavelet attention module. This module divides the features along the channel dimension, performs discrete wavelet transform on some features, and combines the dual attention mechanism of space and channel to enhance the fusion ability of the watermark in the spatial domain and frequency domain. This design effectively optimizes the efficiency of feature utilization, reduces the impact of watermark embedding on image quality, and improves the robustness of the system to complex attacks.

[0040] In some embodiments, see Figure 6 , Figure 6 A schematic diagram of the structure of a remote sensing image robust watermarking device provided in an embodiment of the present application; The present application provides a remote sensing image robust watermarking device 600, including: a feature mapping module 610, a feature fusion module 620 and a distortion simulation module 630, wherein: A feature mapping module 610 is configured to extract features from the original image, map the extracted image features to a latent feature space through an adapter network, and obtain corresponding latent features; The feature fusion module 620 is configured to fuse the latent features with the watermark information and input them into the generator network, and generate the watermark image by combining the U-shaped structure and the half-wavelet attention mechanism; The distortion simulation module 630 is configured to perform random distortion simulation on the watermark image through signal processing and geometric transformation, analyze the distortion characteristics of the distorted image, and obtain watermark information.

[0041] In some embodiments, the feature mapping module 610 is specifically configured as follows: The original image is subjected to feature extraction through convolution operation, batch normalization and ReLU activation function to obtain shallow features; The shallow features are passed through the depth-wise separable convolution module and the point-by-point convolution module to extract the global information of each channel, and the residual connection is performed to obtain the corresponding residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features of the residual features respectively to obtain the fused features; The fused features are processed by convolution, average pooling and full connection layer, and mapped to the latent feature space to obtain the corresponding latent features.

[0042] In some embodiments, the feature mapping module 610 is further configured to: Two convolutional layers and a Sigmoid activation function are used to generate activation values, and the activation values ​​are used to scale the original feature map channel by channel to obtain the channel attention map; Perform global average pooling and maximum pooling operations on the original feature map along the spatial dimension, concatenate the pooling results and pass them through the convolution layer and Sigmoid activation function to generate a spatial attention map; The channel attention map and the spatial attention map are fused to obtain the fused features.

[0043] In some embodiments, the feature fusion module 620 is specifically configured as follows: The fused features are input into the downsampling module, and the spatial resolution of the feature map is gradually reduced through the maximum pooling operation to generate multi-level downsampling features; The downsampled features are passed to the current upsampling layer through skip connections, and the spatial resolution of the feature map is gradually restored using bilinear interpolation to output the final feature map; Generate the target watermark image based on the final feature map.

[0044] In some embodiments, the feature fusion module 620 is specifically configured as follows: The feature map is subjected to depthwise separable convolution and pointwise convolution to extract high-dimensional semantic features, and residual connections are introduced to add the extracted features to the input features; Half-wavelet attention is used to enhance feature representation. The feature map is divided into two parts along the channel dimension. One part retains the original features, and the other part is subjected to discrete wavelet transform. The converted wavelet domain features are input into the dual attention module to obtain weighted wavelet features. The inverse wavelet transform is used to resize the features to their original size, and the wavelet domain features are concatenated with the retained original feature parts. The multi-level feature information is fused to obtain the fused features and generate the watermarked image.

[0045] In some embodiments, the distortion simulation module 630 is specifically configured as follows: Gaussian noise processing, speckle noise processing, stripe noise processing, JPEG compression, image rotation, random cropping and resizing.

[0046] In some embodiments, the distortion simulation module 630 is specifically configured as follows: The distorted image obtained by the distortion simulation is subjected to convolution operation, batch normalization and ReLU activation function processing to extract shallow features; The shallow features are extracted through depth-wise separable convolution and point-by-point convolution to obtain the global information of each channel, and residual connections are performed to obtain residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features respectively, and then fuse them to obtain the fused features; The fused features are processed through convolution, average pooling and fully connected layers to obtain watermark information.

[0047] The remote sensing image robust watermarking device provided in the embodiment of the present application can implement each process in the embodiment corresponding to the above-mentioned remote sensing image robust watermarking method, and will not be described again here to avoid repetition.

[0048] It should be noted that the remote sensing image robust watermark device provided in the embodiment of the present application and the remote sensing image robust watermark method provided in the embodiment of the present application are based on the same application concept. Therefore, the specific implementation of this embodiment can refer to the implementation of the aforementioned remote sensing image robust watermark method, and the repeated parts will not be repeated.

[0049] In some embodiments, see Figure 7 , Figure 7 The electronic device 700 provided in the embodiment of the present application includes a processor 710 and a memory 720; the memory 720 stores a computer program, wherein the computer program implements the above-mentioned remote sensing image robust watermarking method when executed by the processor.

[0050] Specifically, the processor 710 may include, for example, a general-purpose microprocessor, an instruction set processor and / or a related chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 610 may also include an onboard memory for cache purposes. The processor 710 may be a single processing unit or multiple processing units for executing different actions of the method flow according to the embodiment of the present application.

[0051] The memory 720 may be any medium capable of containing, storing, conveying, propagating or transmitting instructions. For example, the memory 720 may include, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared or semiconductor system, device, component or propagation medium. Specific examples of the memory 720 include: a magnetic storage device, such as a magnetic tape or a hard disk (HDD); an optical storage device, such as a compact disk (CD-ROM); a random access memory (RAM) or flash memory; and / or a wired / wireless communication link.

[0052] The present application also provides a computer-readable medium on which a computer program is stored. When the program is executed by a processor, the above-mentioned remote sensing image robust watermarking method is implemented. The computer-readable medium may be included in the device / apparatus / system described in the above-mentioned embodiment; or it may exist independently without being assembled into the device / apparatus / system. The above-mentioned computer-readable medium carries one or more programs. When the above-mentioned one or more programs are executed, the method according to the embodiment of the present application is implemented.

[0053] According to an embodiment of the present application, a computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium containing or storing a program that may be used by or in combination with an instruction execution system, device or device. In the present application, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, wherein a computer-readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, optical cable, radio frequency signal, etc., or any suitable combination of the above.

[0054] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present application. In particular, without departing from the spirit and teachings of the present application, the features described in the various embodiments and / or claims of the present application may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present application. Therefore, the scope of the present application should not be limited to the above-described embodiments, but should be determined not only by the appended claims, but also by the equivalents of the appended claims.

Claims

1. A remote sensing image robust watermarking method, characterized in that: include: Extract features from the original image, map the extracted image features to the latent feature space through the adapter network, and obtain the corresponding latent features; The latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism; The watermark image is subjected to random distortion simulation through signal processing and geometric transformation, and the distortion characteristics of the distorted image are analyzed to obtain watermark information.

2. The remote sensing image robust watermarking method according to claim 1, characterized in that: The feature extraction of the original image is performed, and the extracted image features are mapped to the latent feature space through the adapter network to obtain the corresponding latent features, including: Performing feature extraction on the original image through convolution operation, batch normalization and ReLU activation function to obtain shallow features; The shallow features are passed through a depth-wise separable convolution module and a point-by-point convolution module to extract the global information of each channel, and residual connections are performed to obtain corresponding residual features; The residual feature is input into a parallel dual attention unit, and the spatial feature and the channel feature of the residual feature are respectively enhanced and adjusted to obtain a fusion feature; The fused features are subjected to convolution processing, average pooling processing and full connection layer processing, and mapped to a potential feature space to obtain corresponding potential features.

3. The remote sensing image robust watermarking method according to claim 2, characterized in that: The feature extraction of the original image is performed, and the extracted image features are mapped to the latent feature space through the adapter network to obtain the corresponding latent features, including: Two convolutional layers and a Sigmoid activation function are used to generate activation values, and the activation values ​​are used to scale the original feature map channel by channel to obtain a channel attention map; Performing global average pooling and maximum pooling operations on the original feature map along the spatial dimension, concatenating the pooling results and passing them through a convolutional layer and a Sigmoid activation function to generate a spatial attention map; The channel attention map and the spatial attention map are fused to obtain a fusion feature.

4. The remote sensing image robust watermarking method according to claim 2, characterized in that: The latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism, including: The fused features are input into a downsampling module, and the spatial resolution of the feature map is gradually reduced through a maximum pooling operation to generate multi-level downsampling features; The downsampled features are transferred to the current upsampling layer through a skip connection, the spatial resolution of the feature map is gradually restored using bilinear interpolation, and the final feature map is output; A target watermark image is generated based on the final feature map.

5. The remote sensing image robust watermarking method according to claim 4, characterized in that: The latent features are fused with the watermark information and then input into the generator network, and a watermark image is generated by combining the U-shaped structure and the half-wavelet attention mechanism, including: The feature map is subjected to depthwise separable convolution and pointwise convolution to extract high-dimensional semantic features, and residual connections are introduced to add the extracted features to the input features; Half-wavelet attention is used to enhance feature representation. The feature map is divided into two parts along the channel dimension. One part retains the original features, and the other part is subjected to discrete wavelet transform. The converted wavelet domain features are input into the dual attention module to obtain weighted wavelet features. The inverse wavelet transform is used to resize the features to their original size, and the wavelet domain features are concatenated with the retained original feature parts. The multi-level feature information is fused to obtain the fused features and generate the watermarked image.

6. The remote sensing image robust watermarking method according to claim 1, characterized in that: The signal processing and geometric transformation include: Gaussian noise processing, speckle noise processing, stripe noise processing, JPEG compression, image rotation, random cropping and resizing.

7. The remote sensing image robust watermarking method according to claim 1, characterized in that: The step of performing random distortion simulation on the watermark image through signal processing and geometric transformation, analyzing the distortion characteristics of the distorted image, and obtaining watermark information includes: The distorted image obtained by the distortion simulation is subjected to convolution operation, batch normalization and ReLU activation function processing to extract shallow features; The shallow features are extracted through depth-wise separable convolution and point-by-point convolution to obtain global information of each channel, and residual connections are performed to obtain residual features; The residual features are input into the parallel dual attention unit to strengthen and adjust the spatial features and channel features respectively, and then fuse them to obtain fused features; The fused features are processed through convolution, average pooling and full connection layers to obtain watermark information.

8. A remote sensing image robust watermarking device, characterized in that: include: Feature mapping module, feature fusion module and distortion simulation module, among which, The feature mapping module is configured to extract features from the original image, map the extracted image features to a latent feature space through an adapter network, and obtain corresponding latent features; The feature fusion module is configured to fuse the latent features with the watermark information and input them into the generator network, and generate a watermark image by combining the U-shaped structure and the half-wavelet attention mechanism; The distortion simulation module is configured to perform random distortion simulation on the watermark image through signal processing and geometric transformation, analyze the distortion characteristics of the distorted image, and obtain watermark information.

9. An electronic device comprising a processor and a memory; the memory stores a computer program, wherein: When the computer program is executed by the processor, the computer program implements the remote sensing image robust watermarking method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, wherein when the program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images