Remote sensing image hiding and restoration method and apparatus based on space-frequency co-modeling

Abstract

This invention discloses a method and apparatus for hiding and restoring remote sensing images based on space-frequency collaborative modeling. The method includes: acquiring a remote sensing image to be hidden and a carrier image; inputting the remote sensing image to be hidden and the carrier image into a pre-embedded residual attention module to obtain conditional features; inputting the conditional features into an image hiding network to obtain a hidden image; performing discrete wavelet transform on the hidden image to obtain high-frequency and low-frequency features; inputting the high-frequency and low-frequency features into a Gaussian cross-frequency fusion module to obtain dynamic fusion features; and inputting the dynamic fusion features into a reconstruction and restoration module to obtain the secret image. This invention can improve the fidelity, robustness, and anti-detection properties of the secret image.

Description

Technical Field

[0001] This invention relates to the field of secure transmission and information hiding technology for remote sensing data, and in particular to a method, apparatus, storage medium and electronic device for hiding and restoring remote sensing images based on space-frequency collaborative modeling. Background Technology

[0002] Throughout the entire lifecycle of remote sensing imaging—acquisition, download, archiving, and distribution—there is a risk of interception, alteration, or malicious use at any stage. High-resolution imagery not only carries sub-meter spatial coordinates but also implicitly contains information on resource distribution, disaster assessments, land cover, and even the deployment and movements of sensitive targets: military aircraft parking patterns, naval formation structures, and traces of temporary positions can all be deciphered by adversaries. Behind this immense intelligence value lies an unprecedented security vulnerability.

[0003] Traditional encryption converts images into random bitstreams, which, while theoretically strong, significantly disrupts statistical properties and is easily identified as "secret messages" by traffic monitoring, leading to targeted decryption or channel suppression. Information hiding, on the other hand, disguises confidential data as ordinary images for release. Transmission over public channels does not trigger anomalies, and interceptors find it difficult to determine the presence of secrets using visual or statistical methods, thus avoiding targeted attacks. However, remote sensing images have imaging mechanisms that are drastically different from natural photographs: regular textures (fields, ports, road networks), sharp geometric edges (ridges, riverbanks), flat frequency domain energy distribution, and the link is often affected by compression, noise, and channel fading. Therefore, information hiding schemes for remote sensing must simultaneously meet the dual requirements of "structural robustness" and "detail recovery" while maintaining "visual transparency."

[0004] Existing deep learning hiding algorithms can be broadly categorized into spatial domain and frequency domain approaches. The former directly learns pixel mappings, such as end-to-end networks based on the least significant bit (LSB). However, once compression or filtering occurs, the secret information is erased. The latter embeds mid-to-high frequency coefficients into the data using transformations like DCT and DWT. While this improves resistance to attacks, it faces the dual bottlenecks of "frequency domain interference during the embedding stage" and "simple coupling of high and low frequencies during the recovery stage." High-frequency subbands carry edges and textures, crucial for reconstruction quality; low-frequency subbands provide contours and tone, requiring careful utilization as supplementary information. Simply using splicing or weighted fusion cannot dynamically match the importance of these two aspects, causing the reconstruction result to fluctuate drastically with channel conditions, resulting in a simultaneous decrease in hiding stability and accuracy. Summary of the Invention

[0005] This invention provides a method, apparatus, storage medium, and electronic device for hiding and restoring remote sensing images based on space-frequency collaborative modeling, which can improve the fidelity, robustness, and anti-detection properties of secret images.

[0006] This invention provides a remote sensing image hiding and restoration method based on space-frequency co-modeling, comprising: Acquire the remote sensing image to be hidden and the carrier image; The remote sensing image to be hidden and the carrier image are input into a pre-embedded residual attention module to obtain conditional features. The conditional features are then input into an image hiding network to obtain a hidden image. The dense image is subjected to discrete wavelet transform to obtain high-frequency features and low-frequency features. The high-frequency features and low-frequency features are then input into the Gaussian cross-frequency fusion module to obtain dynamic fusion features. The dynamic fusion features are input into the reconstruction and restoration module to obtain the secret image.

[0007] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency collaborative modeling, the remote sensing image to be hidden and the carrier image are input into a pre-embedded residual attention module to obtain conditional features, including: The remote sensing image to be hidden and the carrier image are input into a convolution module for preliminary feature mapping to obtain remote sensing features and carrier features. The remote sensing features and the carrier features are input into multiple residual attention blocks to obtain deep remote sensing feature representations and deep carrier feature representations; Discrete wavelet transform is performed on the deep remote sensing feature representation to obtain multi-scale sub-band features. The multi-scale sub-band features and the deep carrier features are then concatenated to obtain conditional features.

[0008] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency co-modeling, the processing procedure of the residual attention block is expressed by the following formula:

[0009] in, For deep carrier feature representation, This represents the deep remote sensing features. For upsampling, It is a 3×3 convolution. A deep residual network consisting of N residual channel attention blocks stacked together. As a carrier characteristic, These are remote sensing features.

[0010] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency co-modeling, the image hiding network includes an encoder, a decoder, and an output layer, and the processing procedure of the image hiding network includes: The conditional features are input into the multi-layer coding unit of the encoder to obtain the coding features; The encoded features are input into the dual-branch structure of the encoder to extract local spatial structure features and global features, respectively. The local spatial structure features and the global features are concatenated along the channel dimension to obtain a deep fusion representation; The encoded features output by each coding unit are passed to the corresponding layer of the decoder through skip connections to restore the spatial detail features and obtain the spatial detail features. The spatial detail features are input into the encoding unit of the encoder to obtain the restored deep fusion features; The restored deep fusion features are input into the output layer to obtain a dense image.

[0011] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency co-modeling, after obtaining the high-frequency features, the method further includes: The high-frequency features are enhanced to obtain the final high-frequency enhanced representation, which is expressed by the following formula:

[0012]

[0013]

[0014]

[0015]

[0016] in, For edge enhancement features, and These are the horizontal and vertical components of the edge detection operator, respectively. High-frequency characteristics and Let represent the mean and standard deviation of the high-frequency features, respectively. For convolutional blocks, and It is a non-linear activation function. For convolution operations, It follows a standard normal distribution. Let represent a random variable sampled from a standard normal distribution. For adaptive fusion factor, Gaussian characteristics, For the final high-frequency enhancement representation, This indicates element-wise multiplication.

[0017] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency co-modeling, the high-frequency features and the low-frequency features are input into the Gaussian cross-frequency fusion module to obtain dynamic fusion features, including: Global average pooling is performed on the final high-frequency enhanced representation and low-frequency features, and channel attention weights are generated through a multilayer perceptron. Based on the final high-frequency enhanced representation, the low-frequency features, and the channel attention weights, the fused channel features are obtained. Spatial attention is calculated on the fused channel features, spatial weights are obtained through convolution operations and activation functions, and dynamic fused features are obtained based on the final high-frequency enhanced representation, the low-frequency features, and the spatial weights.

[0018] Furthermore, according to the above-mentioned remote sensing image hiding and restoration method based on space-frequency co-modeling, the processing procedure of the reconstruction and restoration module is represented by the following formula:

[0019]

[0020] in, For frequency domain characteristics, As a dynamic fusion feature, For secret images, For separable convolution, For activation function, For batch normalization, For filling edge pixels, This is the inverse wavelet transform.

[0021] The present invention also provides a remote sensing image hiding and restoration device based on space-frequency co-modeling, comprising: The acquisition module is used to acquire the remote sensing image to be hidden and the carrier image; The first processing module is used to perform a pre-embedded residual attention operation on the remote sensing image to be hidden and the carrier image to obtain conditional features, and input the conditional features into the image hiding network to obtain a hidden image; The second processing module is used to perform discrete wavelet transform on the dense image to obtain high-frequency features and low-frequency features, and to perform Gaussian cross-frequency fusion operation on the high-frequency features and the low-frequency features to obtain dynamic fusion features; The reconstruction and restoration module is used to reconstruct and restore the dynamically fused features to obtain the secret image.

[0022] The present invention also provides a computer-readable storage medium storing a plurality of instructions adapted for loading by a processor to execute any of the above-described remote sensing image hiding and restoration methods based on space-frequency co-modeling.

[0023] The present invention also provides an electronic device, including a processor and a memory, wherein the processor is electrically connected to the memory, the memory is used to store instructions and data, and the processor is used in the steps of the remote sensing image hiding and restoration method based on space-frequency co-modeling described in any of the preceding claims.

[0024] This invention provides a method, apparatus, storage medium, and electronic device for remote sensing image hiding and restoration based on space-frequency co-modeling. The invention inputs the remote sensing image to be hidden and the carrier image into a pre-embedded residual attention module. Based on the enhancement mechanism of residual channel attention, prior information of the image is mined to generate high-quality intermediate features. The acquired features are input into an image hiding network to jointly model local texture and global context, generating a visually faithful and statistically imperceptible dense image. Discrete wavelet transform is performed on the dense image to obtain high-frequency and low-frequency sub-bands (high-frequency features and low-frequency features). The high-frequency and low-frequency sub-bands are input into a Gaussian cross-frequency fusion module, where edge enhancement, Gaussian modeling, and channel-space dual attention weighting are performed sequentially to obtain dynamic fusion features. Inverse wavelet transform is performed on the dynamic fusion features to reconstruct a distortion-free secret image. This invention has the following beneficial effects: (1) The present invention hides and restores the “domain” processing. The embedding stage only operates in the pixel space, completely avoiding the modification of frequency domain coefficients and eliminating the hidden danger of traditional frequency domain steganography being detected by traffic monitoring due to energy anomalies.

[0025] (2) Introducing global-local dual-path modeling, the dense image almost overlaps with the original carrier in terms of color, texture, histogram and spectrum, and the detection resistance rate is close to random guessing.

[0026] (3) The decryption end of this invention first performs wavelet decomposition, and then uses Gaussian edge magnifying glass and channel / space bidirectional attention to denoise high-frequency details and preserve low-frequency structure. Compared with simple splicing or weighted fusion, the reconstructed image has sharp edges and no overall distortion. Attached Figure Description

[0027] The technical solution and other beneficial effects of the present invention will become apparent from the following detailed description of specific embodiments of the invention, in conjunction with the accompanying drawings.

[0028] Figure 1 A flowchart of a remote sensing image hiding and restoration method based on space-frequency collaborative modeling provided in an embodiment of the present invention.

[0029] Figure 2Another flowchart of the remote sensing image hiding and restoration method based on space-frequency collaborative modeling provided in the embodiments of the present invention.

[0030] Figure 3 This is a schematic diagram of the structure of the remote sensing image hiding and restoration device based on space-frequency collaborative modeling provided in an embodiment of the present invention.

[0031] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0032] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, 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.

[0033] This invention provides a method, apparatus, storage medium, and electronic device for remote sensing image hiding and restoration based on space-frequency co-modeling. The remote sensing image hiding and restoration apparatus based on space-frequency co-modeling provided in this invention can be integrated into an electronic device, such as a terminal or server. The terminal can include a tablet computer, laptop computer, personal computer (PC), microprocessor box, or other devices.

[0034] Please see Figure 1 and Figure 2 , Figure 1 This is a flowchart of a remote sensing image hiding and restoration method based on space-frequency co-modeling provided in an embodiment of the present invention. Figure 2 Another flowchart of the remote sensing image hiding and restoration method based on space-frequency co-modeling provided in this embodiment of the invention, which is applied in electronic devices, includes the following steps: S1, acquire the remote sensing image to be hidden and the carrier image.

[0035] Specifically, acquire the remote sensing image to be hidden. and carrier images ,in, and For the number of images, This is the image sequence number.

[0036] It should be noted that the remote sensing images to be hidden in this example range in resolution from 300×300 to 1500×900 pixels, including military aircraft, cargo ships, oil tankers, warships, and yachts, covering various scenes such as airports, seaports, nearshore areas, and offshore areas. The carrier image set covers various object types such as people, vehicles, animals, furniture, food, and transportation vehicles.

[0037] S2, input the remote sensing image to be hidden and the carrier image into the pre-embedded residual attention module to obtain conditional features, and input the conditional features into the image hiding network to obtain the hidden image.

[0038] In one embodiment, the remote sensing image to be hidden and the carrier image are input into a pre-embedded residual attention module to obtain conditional features, including the following steps: S21, input the remote sensing image to be hidden and the carrier image into the convolution module for preliminary feature mapping to obtain remote sensing features and carrier features.

[0039] S22, input the remote sensing features and carrier features into multiple residual attention blocks to obtain deep remote sensing feature representations and deep carrier feature representations.

[0040] First, a preliminary feature mapping is performed on the input carrier image and the image to be hidden using a 1×1 convolution to obtain remote sensing features. and carrier characteristics Subsequently, these feature inputs are fed into a deep residual network composed of N stacked residual channel attention blocks. Each residual channel attention block introduces a channel attention mechanism into the local residual structure, effectively highlighting key information channels through global average pooling and adaptive channel weighting. Then, deep feature representations are generated through 3×3 convolutions and PixelShuffle (upsampling), as shown in the following formula:

[0041] in, For deep carrier feature representation, This represents the deep remote sensing features. For upsampling, It is a 3×3 convolution. A deep residual network consisting of N residual channel attention blocks stacked together. As a carrier characteristic, These are remote sensing features.

[0042] S23. Perform discrete wavelet transform on the deep remote sensing feature representation to obtain multi-scale sub-band features. Then, splice the multi-scale sub-band features and the deep carrier features to obtain conditional features.

[0043] The specific formula is shown below:

[0044] in, For discrete wavelet transform, As a conditional feature, This indicates feature splicing.

[0045] In one embodiment, the image hiding network includes an encoder, a decoder, and an output layer. Conditional features are input into the image hiding network to obtain a hidden image, comprising the following steps: S24, input the conditional features into the multi-layer coding unit of the encoder to obtain the coding features.

[0046] Specifically, the multi-layer coding unit gradually extracts multi-level features of the image through multi-layer dense convolutional blocks and downsampling operations, enhancing the network's ability to represent complex textures and fine-grained structures in remote sensing images. The specific formula is shown below:

[0047] in, Indicates the first Layer coding unit, for The output of the layer coding unit and The input of the layer coding unit, for The encoded features output by the layer coding unit. The input of the first layer coding unit is the conditional feature.

[0048] S25, the encoded features are input into the dual-branch structure of the encoder to extract local spatial structure features and global features respectively.

[0049] A dual-branch structure is used to jointly model local spatial features and global contextual information. One branch of the dual-branch structure models the spatial structural features of the local neighborhood (i.e., local spatial structural features) through a standard 3×3 convolution. Another branch flattens the features into a sequence and feeds it into a lightweight Transformer structure to capture global context features across pixels, thus obtaining global features. The specific formula is as follows:

[0050]

[0051] in, For a lightweight Transformer, This represents the operation of flattening features into a sequence. This is the activation function.

[0052] S26, local spatial structural features and global features are spliced ​​together in the channel dimension to obtain a deep fusion representation.

[0053] The specific formula is shown below:

[0054] S27, the encoded features output by each layer of coding unit are passed to the corresponding layer of the decoder through skip connections to restore the spatial detail features and obtain the spatial detail features.

[0055] Specifically, the layer above the decoder ( l +1) upsamples the deep, abstract feature map (i.e., the decoded features) to the current layer l The same spatial resolution is then combined with the shallow detail features (i.e., encoded features) transmitted from the encoder. The specific formula is:

[0056] in, Indicates an upsampling operation. This indicates channel-by-channel splicing. This refers to the output decoding features of the layer above the decoder. For spatial detail features.

[0057] S28, input the spatial detail features into the encoding unit of the encoder to obtain the restored deep fusion features.

[0058] Specifically, spatial detail features are restored step by step through progressive upsampling and multi-layer dense block operations, as shown in the following formula:

[0059] in, This represents the restored deep fusion features.

[0060] S29. The restored deep fusion features are input into the output layer to obtain a dense image.

[0061] Specifically, a dense image is obtained by adjusting the number of channels through convolutional layers. By aggregating global contextual features and preserving local structure, the proposed hidden network can achieve robust and high-quality embedding of secret information while ensuring the visual quality of the carrier image.

[0062] S3 performs discrete wavelet transform on the dense image to obtain high-frequency and low-frequency features, and inputs the high-frequency and low-frequency features into the Gaussian cross-frequency fusion module to obtain dynamic fusion features.

[0063] In one embodiment, after the step of performing discrete wavelet transform on the dense image to obtain high-frequency and low-frequency features, the method includes: The high-frequency features are enhanced to obtain the final high-frequency enhanced representation.

[0064] Specifically, high-frequency features are enhanced through high-frequency branches, which introduce edge enhancement and Gaussian modeling to highlight texture and detail information and amplify potential embedding traces.

[0065] This can be expressed by the following formula:

[0066]

[0067]

[0068]

[0069]

[0070] in, For edge enhancement features, and These are the horizontal and vertical components of the edge detection operator, respectively. High-frequency characteristics and Let represent the mean and standard deviation of the high-frequency features, respectively. For convolutional blocks, and It is a non-linear activation function. For convolution operations, It follows a standard normal distribution. Let represent a random variable sampled from a standard normal distribution. For adaptive fusion factor, Gaussian characteristics, For the final high-frequency enhancement representation, This indicates element-wise multiplication.

[0071] High frequency features and low frequency characteristics The input is fed into the Gaussian cross-frequency fusion module to obtain dynamic fusion features, including: S31, global average pooling is performed on the final high-frequency enhanced representation and low-frequency features, and channel attention weights are generated through a multilayer perceptron. Based on the final high-frequency enhanced representation, low-frequency features and channel attention weights, the fused channel features are obtained.

[0072] Specifically, it is expressed by the following formula:

[0073]

[0074] in, For global average pooling, For channel attention weights, This is the final high-frequency enhancement representation.

[0075] S32 calculates spatial attention for the fused channel features, obtains spatial weights through convolution operations and activation functions, and obtains dynamic fused features based on the final high-frequency enhanced representation, low-frequency features and spatial weights.

[0076] Specifically, it is expressed by the following formula:

[0077]

[0078] in, For activation function, Spatial weights, This is a dynamic fusion feature.

[0079] S4, input the dynamic fusion features into the reconstruction and restoration module to obtain the secret image.

[0080] The reconstruction and restoration module first reconstructs the frequency domain features from the dynamically fused features to further enhance the discriminative power of cross-channel features. Then, it performs an inverse wavelet transform on the frequency domain features to restore them to the spatial domain, resulting in a high-quality secret image. The processing procedure of the reconstruction and restoration module is represented by the following formula:

[0081]

[0082] in, For frequency domain characteristics, As a dynamic fusion feature, For secret images, For separable convolution, For activation function, For batch normalization, For filling edge pixels, This is the inverse wavelet transform.

[0083] In one embodiment, the pre-embedded residual attention module, the hidden network, and the Gaussian cross-frequency fusion module are all jointly trained end-to-end, and the loss function includes image fidelity loss, perceptual loss, and anti-detection loss.

[0084] The formula for image fidelity loss is:

[0085] in, For dense images, For carrier image, It is the square of the L2 norm.

[0086] The perceived loss is:

[0087] in, For layer weights, The first pre-trained network Layer activation graph; Resistance to detection loss:

[0088] in, A binary classifier outputs the "true" probability p∈[0,1].

[0089] Based on the method described in the above embodiments, this embodiment will further describe the remote sensing image hiding and restoration device based on space-frequency collaborative modeling. The remote sensing image hiding and restoration device based on space-frequency collaborative modeling can be implemented as an independent entity or integrated into an electronic device. The electronic device can be a terminal, server, or other devices. The terminal can include a tablet computer, a laptop computer, a personal computer (PC), a microprocessor box, or other devices.

[0090] Please see Figure 3 , Figure 3 This invention specifically describes a remote sensing image hiding and restoration device based on space-frequency co-modeling, which is applied in electronic devices. The device may include: The acquisition module is used to acquire the remote sensing image to be hidden and the carrier image; The first processing module is used to perform a pre-embedded residual attention operation on the remote sensing image to be hidden and the carrier image to obtain conditional features, and input the conditional features into the image hiding network to obtain a hidden image; The second processing module is used to perform discrete wavelet transform on the dense image to obtain high-frequency features and low-frequency features, and to perform Gaussian cross-frequency fusion operation on the high-frequency features and the low-frequency features to obtain dynamic fusion features; The reconstruction and restoration module is used to reconstruct and restore the dynamically fused features to obtain the secret image.

[0091] In specific implementation, the above modules and / or units can be implemented as independent entities, or they can be arbitrarily combined and implemented as the same or several entities. For the specific implementation of the above modules and / or units, please refer to the previous method embodiments. For the specific beneficial effects that can be achieved, please also refer to the beneficial effects in the previous method embodiments, which will not be repeated here.

[0092] In addition, this embodiment of the invention also provides an electronic device, which may be a computer, tablet computer, or other similar device. This electronic device can implement the steps of any embodiment of the remote sensing image hiding and restoration method based on space-frequency co-modeling provided in this embodiment of the invention. Therefore, it can achieve the beneficial effects that any remote sensing image hiding and restoration method based on space-frequency co-modeling provided in this embodiment of the invention can achieve, as detailed in the preceding embodiments, and will not be repeated here.

[0093] Figure 4 A specific structural block diagram of an electronic device provided in an embodiment of the present invention is shown. This electronic device can be used to implement the remote sensing image hiding and restoration method based on space-frequency collaborative modeling provided in the above embodiments. The electronic device 500 can be a terminal, server, or other device. The terminal can include a tablet computer, laptop computer, personal computer (PC), microprocessor box, or other devices.

[0094] The memory 520 can be used to store software programs and modules, such as the program instructions / modules corresponding to those in the above embodiments. The processor 580 executes various functional applications and data processing by running the software programs and modules stored in the memory 520. The memory 520 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 520 may further include memory remotely located relative to the processor 580, and these remote memories can be connected to the electronic device 500 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0095] The input unit 530 can be used to receive input numeric or character information, and to generate a keyboard and mouse related to user settings and function control. Display unit 540 can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces, which can be composed of graphics, text, icons, video, and any combination thereof. Display unit 540 may include display panel 541, which may optionally be configured in the form of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), or other similar forms.

[0096] Electronic device 500, through transmission module 570 (e.g., Wi-Fi module), can help users receive requests, send information, etc., providing users with wireless broadband internet access. Although transmission module 570 is shown in the figure, it is understood that it is not an essential component of electronic device 500 and can be omitted as needed without changing the essence of the invention.

[0097] The processor 580 is the control center of the electronic device 500. It connects to various parts of the phone via various interfaces and lines, and performs various functions and processes data of the electronic device 500 by running or executing software programs and / or modules stored in the memory 520, and by calling data stored in the memory 520, thereby providing overall monitoring of the electronic device. Optionally, the processor 580 may include one or more processing cores; in some embodiments, the processor 580 may integrate an application processor and a modem processor, wherein the application processor mainly handles the operating system, user interface, and applications, and the modem processor mainly handles wireless communication. It is understood that the modem processor may also not be integrated into the processor 580.

[0098] Electronic device 500 also includes a power supply 590 (such as a battery) that supplies power to various components. In some embodiments, the power supply may be logically connected to processor 580 through a power management system, thereby enabling functions such as charging, discharging, and power consumption management through the power management system. The power supply 590 may also include one or more DC or AC power supplies, recharging systems, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components.

[0099] Although not shown, the electronic device 500 also includes cameras (such as front-facing cameras and rear-facing cameras), Bluetooth modules, etc., which will not be described in detail here. Specifically, in this embodiment, the display unit of the electronic device is a touch screen display, and the mobile terminal also includes a memory and one or more programs, wherein one or more programs are stored in the memory and configured to be executed by one or more processors. One or more programs contain instructions for performing the following operations: Acquire the remote sensing image to be hidden and the carrier image; The remote sensing image to be hidden and the carrier image are input into a pre-embedded residual attention module to obtain conditional features. The conditional features are then input into an image hiding network to obtain a hidden image. The dense image is subjected to discrete wavelet transform to obtain high-frequency features and low-frequency features. The high-frequency features and low-frequency features are then input into the Gaussian cross-frequency fusion module to obtain dynamic fusion features. The dynamic fusion features are input into the reconstruction and restoration module to obtain the secret image.

[0100] In practice, the above modules can be implemented as independent entities or combined in any way to be implemented as the same or several entities. For the specific implementation of the above modules, please refer to the previous method implementation examples, which will not be repeated here.

[0101] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by instructions, or by instructions controlling related hardware. These instructions can be stored in a computer-readable storage medium and loaded and executed by a processor. Therefore, embodiments of the present invention provide a storage medium storing multiple instructions that can be loaded by a processor to execute the steps of any embodiment of the remote sensing image hiding and restoration method based on space-frequency co-modeling provided by the present invention.

[0102] The computer-readable storage medium may include: read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0103] Since the instructions stored in the storage medium can execute the steps in any embodiment of the remote sensing image hiding and restoration method based on space-frequency co-modeling provided in the embodiments of the present invention, the beneficial effects that any remote sensing image hiding and restoration method based on space-frequency co-modeling provided in the embodiments of the present invention can achieve can be realized. For details, please refer to the previous embodiments, which will not be repeated here.

[0104] The foregoing has provided a detailed description of a remote sensing image hiding and restoration method, apparatus, storage medium, and electronic device based on space-frequency collaborative modeling provided by embodiments of the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, for those skilled in the art, there will be changes in specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A remote sensing image hiding and restoration method based on space-frequency co-modeling, characterized in that, The method includes: Acquire the remote sensing image to be hidden and the carrier image; The remote sensing image to be hidden and the carrier image are input into a pre-embedded residual attention module to obtain conditional features. These conditional features are then input into an image hiding network to obtain a hidden image. The process of inputting the remote sensing image to be hidden and the carrier image into the pre-embedded residual attention module to obtain conditional features includes: The remote sensing image to be hidden and the carrier image are input into a convolutional module for preliminary feature mapping to obtain remote sensing features and carrier features. The remote sensing features and the carrier features are then input into multiple residual attention blocks to obtain deep remote sensing feature representations and deep carrier feature representations. Discrete wavelet transform is performed on the deep remote sensing feature representations to obtain multi-scale sub-band features. The multi-scale sub-band features and the deep carrier features are then concatenated to obtain conditional features. The processing procedure for the residual attention block is expressed by the following formula: in, For deep carrier feature representation, This represents the deep remote sensing features. For upsampling, It is a 3×3 convolution. A deep residual network consisting of N residual channel attention blocks stacked together. As a carrier characteristic, Remote sensing features; The dense image is subjected to discrete wavelet transform to obtain high-frequency features and low-frequency features. The high-frequency features and low-frequency features are then input into the Gaussian cross-frequency fusion module to obtain dynamic fusion features. The process, following the step of obtaining high-frequency features, also includes: The high-frequency features are enhanced to obtain the final high-frequency enhanced representation, which is expressed by the following formula: in, For edge enhancement features, and These are the horizontal and vertical components of the edge detection operator, respectively. High-frequency characteristics and Let represent the mean and standard deviation of the high-frequency features, respectively. For convolutional blocks, and It is a non-linear activation function. For convolution operations, It follows a standard normal distribution. Let represent a random variable sampled from a standard normal distribution. For adaptive fusion factor, Gaussian characteristics, For the final high-frequency enhancement representation, This indicates element-wise multiplication; The dynamic fusion features are input into the reconstruction and restoration module to obtain the secret image.

2. The remote sensing image hiding and restoration method based on space-frequency co-modeling according to claim 1, characterized in that, The image hiding network includes an encoder, a decoder, and an output layer. The processing procedure of the image hiding network includes: The conditional features are input into the multi-layer coding unit of the encoder to obtain the coding features; The encoded features are input into the dual-branch structure of the encoder to extract local spatial structure features and global features, respectively. The local spatial structure features and the global features are concatenated along the channel dimension to obtain a deep fusion representation; The encoded features output by each coding unit are passed to the corresponding layer of the decoder through skip connections to restore the spatial detail features and obtain the spatial detail features. The spatial detail features are input into the encoding unit of the encoder to obtain the restored deep fusion features; The restored deep fusion features are input into the output layer to obtain a dense image.

3. The remote sensing image hiding and restoration method based on space-frequency co-modeling according to claim 1, characterized in that, The high-frequency features and the low-frequency features are input into the Gaussian cross-frequency fusion module to obtain dynamic fused features, including: Global average pooling is performed on the final high-frequency enhanced representation and low-frequency features, and channel attention weights are generated through a multilayer perceptron. Based on the final high-frequency enhanced representation, the low-frequency features, and the channel attention weights, the fused channel features are obtained. Spatial attention is calculated on the fused channel features, spatial weights are obtained through convolution operations and activation functions, and dynamic fused features are obtained based on the final high-frequency enhanced representation, the low-frequency features, and the spatial weights.

4. The remote sensing image hiding and restoration method based on space-frequency co-modeling according to claim 1, characterized in that, The processing procedure of the reconstruction and recovery module is represented by the following formula: in, For frequency domain characteristics, As a dynamic fusion feature, For secret images, For separable convolution, For activation function, For batch normalization, For filling edge pixels, This is the inverse wavelet transform.

5. A remote sensing image hiding and restoration device based on space-frequency co-modeling, wherein the remote sensing image hiding and restoration device based on space-frequency co-modeling is used to execute the remote sensing image hiding and restoration method based on space-frequency co-modeling as described in any one of claims 1-4, characterized in that, include: The acquisition module is used to acquire the remote sensing image to be hidden and the carrier image; The first processing module is used to perform a pre-embedded residual attention operation on the remote sensing image to be hidden and the carrier image to obtain conditional features, and input the conditional features into the image hiding network to obtain a hidden image; The second processing module is used to perform discrete wavelet transform on the dense image to obtain high-frequency features and low-frequency features, and to perform Gaussian cross-frequency fusion operation on the high-frequency features and the low-frequency features to obtain dynamic fusion features; The reconstruction and restoration module is used to reconstruct and restore the dynamically fused features to obtain the secret image.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a plurality of instructions adapted for loading by a processor to execute the remote sensing image hiding and restoration method based on space-frequency co-modeling as described in any one of claims 1 to 4.

7. An electronic device, characterized in that, The device includes a processor and a memory, the processor being electrically connected to the memory, the memory being used to store instructions and data, and the processor being used to execute the steps in the remote sensing image hiding and restoration method based on space-frequency co-modeling as described in any one of claims 1 to 4.

Images