A remote sensing image transmission method and device based on channel feedback and target orientation, a terminal and a medium

Abstract

The application provides a remote sensing image transmission method and device based on channel feedback and target orientation, a terminal and a medium, the method comprising: performing semantic segmentation on an obtained original remote sensing image; determining a region in the remote sensing image most relevant to an application task based on a current application task to be executed and a real-time feedback channel signal-to-noise ratio state and using the semantic segmentation image to obtain an optimized image to be transmitted; inputting the optimized image into a UAV transmitting end to perform an encoding operation on the optimized image; and sending an image encoding result to a ground receiving end through a low-altitude wireless channel to perform a decoding operation on the image encoding result to obtain a target image. The application reduces redundant data by preferentially transmitting image content related to a task and actively adapts to a real-time feedback channel signal-to-noise ratio state to maintain image recovery quality under varying channel conditions, thereby realizing high-quality image transmission of remote sensing images under a bandwidth-limited and channel-varying scene.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a remote sensing image transmission method, apparatus, terminal, and medium based on channel feedback and target guidance. Background Technology

[0002] In the field of low-altitude communication, it is difficult to meet the high-efficiency transmission requirements under dynamic channels and limited bandwidth. Traditional communication usually focuses on the accurate transmission of raw bits without exploring the semantic value of data, resulting in the following problems: First, the bandwidth utilization is extremely low. Traditional solutions require the transmission of complete raw data, leading to a high proportion of redundant data, which can easily cause spectrum resource waste or congestion. The root cause is the failure to distinguish the semantic importance of data and the indiscriminate transmission of all information. Second, the ability to resist channel fluctuations is weak. Due to the dynamic changes in channel SNR, traditional fixed channel coding schemes cannot adaptively adjust, and the bit error rate increases significantly in low SNR scenarios, thereby compromising data integrity.

[0003] While existing technologies attempt to combine semantic communication with low-altitude communication, they still suffer from insufficient adaptability under limited bandwidth conditions. For example, existing solutions only optimize for the Doppler effect and do not address the semantic selection problem under bandwidth constraints. Although existing models can adapt to multi-SNR scenarios, they do not distinguish the priority of semantic content and still need to transmit non-critical semantic information, resulting in wasted bandwidth.

[0004] In summary, existing low-altitude communication systems suffer from low data transmission efficiency due to limited bandwidth resources and degraded received image quality due to dynamic fluctuations in the channel signal-to-noise ratio. Therefore, providing a solution to these technical problems is a current challenge for those skilled in the art. Summary of the Invention

[0005] In view of this, the purpose of the present invention is to provide a remote sensing image transmission method, device, terminal and medium based on channel feedback and target guidance, which aims to solve the problems of low data transmission efficiency caused by limited bandwidth resources and degraded received image quality due to dynamic fluctuations in channel signal-to-noise ratio in existing low-altitude communication.

[0006] The technical solution adopted by this invention to solve the technical problem is as follows: In a first aspect, the present invention discloses a remote sensing image transmission method based on channel feedback and target guidance, wherein the method includes: Semantic segmentation is performed on the acquired raw remote sensing image to obtain a semantically segmented image; Based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and using the semantic segmentation image to determine the region in the original remote sensing image that is most relevant to the application task, an optimized image to be transmitted is obtained. The optimized image is input to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain the image encoding result; The image encoding result is transmitted to a ground receiver via a low-altitude wireless channel, and the ground receiver performs a decoding operation on the image encoding result to obtain the target image.

[0007] Optionally, the step of performing semantic segmentation on the acquired original remote sensing image to obtain a semantically segmented image includes: The acquired raw remote sensing images are input into a pre-trained semantic segmentation model for pixel-level classification to obtain semantic segmentation images containing multi-class labels; The semantic segmentation model includes an encoder, a decoder, and a classification layer. The encoder is used to extract features from the remote sensing image using standard two-dimensional convolution and perform max pooling to obtain the corresponding feature map. The decoder is used to upsample the feature map output by the encoder using the pooling index stored in the encoder to obtain the reconstructed upsampled result. The classification layer is used to classify pixels to obtain the category label of each pixel.

[0008] Optionally, the step of determining the region in the original remote sensing image most relevant to the application task based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and using the semantic segmentation image to obtain the optimized image to be transmitted, includes: Determine the target category of the application task to be executed, and perform a masking operation on the semantic segmentation image according to the target category of the application task to be executed to obtain a binary mask image; Based on the binary mask image, regions in the original remote sensing image that are masked as being irrelevant to the application task are determined, thus obtaining non-important regions; The channel signal-to-noise ratio (SNR) status is evaluated using a preset evaluation function to obtain the evaluation result corresponding to the channel SNR status. Based on the evaluation results, an adaptive mean blurring filter operation is performed on the non-important regions in the original remote sensing image to obtain the processed remote sensing image. A mask correction operation is performed on the processed remote sensing image to obtain an optimized image to be transmitted.

[0009] Optionally, performing a masking operation on the semantic segmentation image according to the target category of the application task to be performed to obtain a binary mask image includes: The target category of the application task to be executed is mapped and matched with the multi-category label in the semantic segmentation image to obtain the corresponding matching result; A masking operation is performed on the semantic segmentation image based on the matching result to obtain a binary mask image.

[0010] Optionally, performing a masking operation on the semantic segmentation image based on the matching result includes: When the matching result indicates that the target category of the application task to be executed matches a category label in the semantic segmentation image, the category pixel value corresponding to the category label that matches the target category in the semantic segmentation image is set to 1. If the matching result indicates that the target category of the application task to be executed does not match a category label in the semantic segmentation image, then the category pixel value corresponding to the category label in the semantic segmentation image that does not match the target category is set to 0.

[0011] Optionally, performing the encoding operation on the optimized image includes: A semantic coding operation is performed on the optimized image to obtain a semantic coding result, and a channel coding operation is performed on the semantic coding result.

[0012] Optionally, performing the decoding operation on the image encoding result includes: The image encoding result is subjected to channel decoding to obtain a channel decoding result, and the channel decoding result is subjected to semantic decoding.

[0013] Secondly, the present invention also discloses a remote sensing image transmission device based on channel feedback and target guidance, wherein the device comprises: The image segmentation module is used to perform semantic segmentation on the acquired raw remote sensing images to obtain semantically segmented images; The segmentation feature selection module is used to determine the region in the original remote sensing image that is most relevant to the application task based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and to obtain the optimized image to be transmitted. An image encoding module is used to input the optimized image to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain an image encoding result; The image decoding module is used to send the image encoding result to a ground receiver via a low-altitude wireless channel, so that the ground receiver can perform a decoding operation on the image encoding result to obtain the target image.

[0014] Thirdly, the present invention discloses a terminal, comprising: a memory, a processor, and a remote sensing image transmission program based on channel feedback and target orientation stored in the memory and executable on the processor, wherein the remote sensing image transmission program based on channel feedback and target orientation implements the steps of the remote sensing image transmission method based on channel feedback and target orientation as described above when executed by the processor.

[0015] Fourthly, the present invention discloses a computer-readable storage medium storing a computer program that can be executed to implement the steps of the remote sensing image transmission method based on channel feedback and target orientation as described above.

[0016] This invention provides a remote sensing image transmission method, apparatus, terminal, and medium based on channel feedback and target orientation. The remote sensing image transmission method based on channel feedback and target orientation includes: performing semantic segmentation on the acquired original remote sensing image to obtain a semantically segmented image; determining the region in the original remote sensing image most relevant to the application task based on the current application task and the real-time feedback channel signal-to-noise ratio (SNR) status, and using the semantically segmented image to obtain an optimized image to be transmitted; inputting the optimized image to a UAV transmitter to perform encoding operations on the optimized image, obtaining an image encoding result; and transmitting the image encoding result to a ground receiver via a low-altitude wireless channel, where the ground receiver performs decoding operations on the image encoding result to obtain the target image. Therefore, this invention decomposes the remote sensing image through semantic segmentation, prioritizes the transmission of image content relevant to the current application task to reduce redundant data, and actively adapts to the real-time feedback channel SNR status to maintain image recovery quality under varying channel conditions, ultimately achieving high-quality image transmission of remote sensing images in bandwidth-constrained and channel-variable scenarios. Attached Figure Description

[0017] Figure 1 This is a flowchart of a preferred embodiment of the remote sensing image transmission method based on channel feedback and target guidance in this invention; Figure 2 This is a schematic diagram of a specific interleaved window mechanism disclosed in this invention; Figure 3 This is a schematic diagram of a specific channel codec structure disclosed in this invention; Figure 4 This is a schematic block diagram of a preferred embodiment of a remote sensing image transmission method based on channel feedback and target guidance disclosed in this invention. Figure 5 This is a functional principle block diagram of a preferred embodiment of the remote sensing image transmission device based on channel feedback and target guidance in this invention; Figure 6 This is a functional principle block diagram of a preferred embodiment of the terminal in this invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer and more explicit, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] In the field of low-altitude communication, although existing technologies have explored channel adaptability, bandwidth utilization and semantic transmission, they all have significant shortcomings and are difficult to meet the high-efficiency transmission requirements under dynamic channels and limited bandwidth.

[0020] Traditional communication typically focuses on accurately transmitting raw bits without exploring the semantic value of data, resulting in the following problems: First, bandwidth utilization is extremely low. Traditional solutions require the transmission of complete raw data, leading to a high proportion of redundant data and easily causing spectrum resource waste or congestion. The root cause is the failure to distinguish the semantic importance of data, transmitting all information indiscriminately. Second, it has weak resistance to channel fluctuations. Due to the dynamic changes in channel SNR, traditional fixed channel coding schemes cannot adaptively adjust, and the bit error rate increases significantly in low SNR scenarios, thereby compromising data integrity.

[0021] While existing terrestrial semantic communication technologies can improve transmission efficiency through semantic compression, they present the following problems when directly applied to low-altitude scenarios: First, semantic evaluation bias. The semantic weight models of these schemes are trained based on terrestrial channels and do not consider the noise and interference patterns unique to low-altitude channels, leading to inaccurate semantic importance judgments and a sharp drop in transmission efficiency in low-altitude scenarios. Second, excessive computational overhead. Terrestrial solutions often employ complex semantic perception strategies such as gradient analysis and entropy modeling, while UAVs have limited computing power and energy budgets, making it impossible to support real-time semantic reasoning and model parameter updates.

[0022] Furthermore, although existing technologies attempt to combine semantic communication with low-altitude communication, they still suffer from insufficient adaptability under limited bandwidth conditions. For example, existing solutions only optimize for the Doppler effect and do not address the semantic selection problem under bandwidth constraints. Although existing models can adapt to multi-SNR scenarios, they do not distinguish the priority of semantic content and still need to transmit non-critical semantic information, resulting in bandwidth waste.

[0023] In summary, existing low-altitude communication systems suffer from low data transmission efficiency due to limited bandwidth resources and degraded received image quality due to dynamic fluctuations in the channel signal-to-noise ratio. Therefore, providing a solution to these technical problems is a current challenge for those skilled in the art.

[0024] To this end, this application provides a remote sensing image transmission scheme based on channel feedback and target guidance, which can achieve high-quality image transmission of remote sensing images in scenarios with limited bandwidth and variable channels.

[0025] Please see Figure 1 , Figure 1 This is a flowchart of the remote sensing image transmission method based on channel feedback and target guidance in this invention. For example... Figure 1 As shown in the embodiments of the present invention, the remote sensing image transmission method based on channel feedback and target guidance includes: Step S11: Perform semantic segmentation on the acquired original remote sensing image to obtain a semantically segmented image.

[0026] In this embodiment, the original remote sensing image to be transmitted is acquired, and semantic segmentation is performed on the original remote sensing image. Specifically, the acquired original remote sensing image is input into a pre-trained semantic segmentation model for pixel-level classification to obtain a semantically segmented image containing multi-class labels. It is understandable that in low-altitude communication, especially in remote sensing image analysis tasks, UAVs face challenges in data transmission capabilities. A selective transmission strategy can be adopted, first performing semantic segmentation on different types of targets in the remote sensing image. The semantic segmentation model then performs semantic segmentation processing on the remote sensing data. Semantic segmentation helps distinguish different types of task targets, enabling the assessment of the importance of each region and prioritizing transmission. Furthermore, it allows for efficient feature encoding and decoding during subsequent transmission, preserving spatial information while compressing features.

[0027] For example, remote sensing images Input a semantic segmentation model to process remote sensing images using the semantic segmentation model. Perform semantic segmentation and generate corresponding output. ,Right now: ; in, These are the model parameters for the semantic segmentation model.

[0028] Specifically, the image segmentation module may include an encoder, a decoder, and a classification layer. The encoder is used to extract features from the remote sensing image using standard two-dimensional convolution and perform max pooling to obtain the corresponding feature map. The decoder is used to upsample the feature map output by the encoder using the pooling index stored in the encoder to obtain the reconstructed upsampled result. The classification layer is used to classify the pixels to obtain the category label of each pixel.

[0029] For example, the encoder can use the first 13 convolutional layers of the VGG16 encoder for feature extraction, or it can use lighter networks such as MobileNet or EfficientNet to further reduce the computational power consumption of the UAV. Feature extraction from remote sensing images is performed using standard two-dimensional convolution, i.e.: ; in, The feature map corresponding to the input remote sensing image. For convolution kernel, For bias terms, For the output feature map, M is the number of pixels in the height (vertical) direction of the convolution kernel, and N is the number of pixels in the width (horizontal) direction of the convolution kernel.

[0030] Then, by using 2×2 max pooling, the resolution can be reduced to decrease the computational cost, i.e.: ; Simultaneously, the encoder stores pooling indexes, which assist the decoder in restoring resolution for subsequent upsampling operations. That is, in the decoder stage, the image segmentation module uses the pooling indexes stored in the encoder to perform upsampling to obtain the reconstructed upsampled result, thereby restoring boundary information and improving segmentation accuracy. ; in, This represents the feature map obtained from the encoder. This represents the reconstructed upsampling result in the decoder.

[0031] Compared with traditional deconvolution or bilinear interpolation methods, this mechanism has a stronger ability to preserve geometric structure and can restore spatial resolution without introducing redundant information. In other words, image segmentation relies on the pooling index upsampling mechanism to avoid the introduction of redundant information in traditional deconvolution.

[0032] Furthermore, the classification layer can specifically use Softmax to classify the pixels in the upsampled results and output the category label for each pixel, i.e.: ; in, For pixels, the category they belong to The probability, For category The score, This represents the total number of categories.

[0033] Step S12: Based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and using the semantic segmentation image, determine the region in the original remote sensing image that is most relevant to the application task, and obtain the optimized image to be transmitted.

[0034] In this embodiment, based on goal orientation (i.e., the application task to be executed) and real-time feedback of the channel signal-to-noise ratio, semantic image segmentation is used to determine the region in the original remote sensing image most relevant to the task, while other regions are suppressed to prioritize the transmission of key region information, thereby achieving segmentation feature selection. ; in, For the optimized image to be transmitted, This refers to the channel signal-to-noise ratio (SNR) status. This refers to the application task (i.e., the task objective) that needs to be executed at the moment.

[0035] Specifically, the target category of the application task to be executed is determined, and a masking operation is performed on the semantic segmentation image according to the target category of the application task to be executed, resulting in a binary mask image. Based on the binary mask image, regions in the original remote sensing image that are masked and are irrelevant to the application task are identified, resulting in non-important regions. A preset selection evaluation function is used to evaluate the real-time feedback channel signal-to-noise ratio (SNR) state, resulting in an evaluation result corresponding to the channel SNR state. Based on the evaluation result, an adaptive mean fuzzy filtering operation is performed on the non-important regions in the original remote sensing image, resulting in a processed remote sensing image. A mask correction operation is performed on the processed remote sensing image to obtain an optimized image to be transmitted. It can be understood that, in order to solve the problem of UAV data transmission under bandwidth-limited and unstable channel conditions, a semantic selection strategy based on selection evaluation is used to analyze the application task to be executed and the channel SNR state, select appropriate transmission content, and maximize the use of bandwidth resources while ensuring task completion. For example, the selection evaluation model can be used to obtain the image restoration result evaluation value based on channel feedback and target orientation, and further select task-related transmission content according to the semantic selection model to meet diverse task requirements and adapt to different channel conditions, thereby optimizing transmission efficiency and improving reconstruction performance. In other words, a semantic segmentation model is used to perform semantic segmentation on the input raw remote sensing image to identify different target tasks in the raw remote sensing image. Then, a masking operation is used to filter out the target regions in the raw remote sensing image that are related to the task and generate a binary mask image. Further, the evaluation results corresponding to the channel signal-to-noise ratio state are used to determine the adaptive mean fuzzy filtering scheme. The adaptive mean fuzzy filtering scheme is then applied to non-important regions in the raw remote sensing image other than the target regions to enhance the visual effect and improve the accuracy of the analysis. Finally, combined with masking correction technology, an optimized image to be transmitted is generated to prepare for subsequent data transmission.

[0036] Furthermore, in this embodiment, a masking operation is performed on the semantic segmentation image according to the target category of the application task to be executed, resulting in a binary mask image. Specifically, this may include: mapping and matching the target category of the application task to be executed with the multi-category labels in the semantic segmentation image to obtain the corresponding matching results; performing a masking operation on the semantic segmentation image based on the matching results to obtain a binary mask image. For example, if the matching results show that the target category of the application task to be executed matches a category label in the semantic segmentation image, then the category pixel value corresponding to the category label that matches the target category in the semantic segmentation image is set to 1; if the matching results show that the target category of the application task to be executed does not match a category label in the semantic segmentation image, then the category pixel value corresponding to the category label that does not match the target category in the semantic segmentation image is set to 0. That is, based on the matching results, the category pixel value corresponding to the target category of the application task to be executed in the semantic segmentation image is set to 1, and the other category pixel values ​​are set to 0.

[0037] It should be noted that the preset selection evaluation function is used to assess the transmission quality of remote sensing images under bandwidth constraints and noise interference, thereby avoiding reliance on subjective visual evaluation. Simultaneously, it helps to identify and prioritize important image content, enabling UAVs to transmit high-value data within limited communication bandwidth while maximizing information output. This preset selection evaluation function can be: ; in, For the evaluation results, Represents an image or target task. This indicates external conditions (such as the channel signal-to-noise ratio).

[0038] To better capture the nonlinear relationships between factors such as task performance and signal-to-noise ratio, and improve prediction accuracy, linear regression is used for modeling. This allows semantic selection to be based on the linear regression model, linking the task, channel, and transmission performance, achieving on-demand transmission and saving bandwidth. The model's goal is to find the optimal... This makes the estimated value Close to the true value Alternatively, non-linear models such as gradient boosting trees can be used to improve the accuracy of semantic selection; First, establish a linear regression model: ; in, For the model parameters that need to be optimized, This is noise error.

[0039] Then, the mean squared error is used as the loss function, that is: ; in, For the sample size, for the parameter The gradient of the loss function is: ; in, These are the input features for the corresponding samples.

[0040] Finally, the weights are updated using the gradient descent algorithm, i.e.: ; in, This is the learning rate.

[0041] By selecting an evaluation method, when new image data is input, the important segmentation features of the image are first extracted. Then, the current channel conditions are input, and based on these factors, an evaluation value of the image reconstruction result is output. Finally, based on this evaluation value, key areas can be prioritized for transmission. That is, in the segmentation feature selection stage, a linear regression evaluation model is first constructed, and an evaluation value is output by combining the task type and channel state. Then, key transmission content is filtered through masking operations and adaptive fuzzy filtering.

[0042] Step S13: Input the optimized image to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain the image encoding result.

[0043] In this embodiment, after semantic segmentation of the original remote sensing image and selection of segmentation features are completed, the optimized image to be transmitted can be transmitted to the UAV transmitter. The UAV transmitter then encodes the optimized image, specifically performing semantic encoding operations to obtain the semantic encoding result, and then performing channel encoding operations on the semantic encoding result. It should be noted that the UAV transmitter can include a semantic encoder and a channel encoder, i.e.: ; ; in, This is the semantic encoding result. These are the model parameters for the semantic encoder. These are the model parameters for the channel encoder.

[0044] Step S14: The image encoding result is sent to the ground receiving end via a low-altitude wireless channel, so that the ground receiving end can perform a decoding operation on the image encoding result to obtain the target image.

[0045] In this embodiment, after the UAV transmitter outputs the image encoding result, the image encoding result y is transmitted to a low-altitude channel, that is, it is sent to the ground receiver via a low-altitude wireless channel. The ground receiver then performs a decoding operation on the image encoding result. Specifically, it performs channel decoding on the image encoding result to obtain the channel decoding result, and performs semantic decoding on the channel decoding result. It should be noted that the ground receiver includes a channel decoder and a semantic decoder, and due to channel fading and noise, the signal... After passing through the channel, it will become The fading process is modeled as follows: ; in, It is random Gaussian noise and , This represents the channel gain.

[0046] at last, The data is sequentially fed into the channel decoder and semantic decoder for decoding to generate the target image. ,Right now: ; ; in, The result of channel decoding. These are the model parameters for the channel decoder. These are the model parameters for the semantic decoder.

[0047] It should also be noted that the semantic encoder and decoder can use the InnerLeave Transformer model as a pre-trained model. This model employs depthwise separable convolution and staggered windowing, allowing pixels from different regions to be received during attention calculations in each window, achieving a balance between computational complexity and performance. Alternatively, the semantic encoder and decoder can be a lightweight variant of ViT, combined with knowledge distillation compression parameters. Meanwhile, the channel encoder and decoder can employ a channel signal-to-noise ratio adjustment module and convolutional operations to implement channel encoding and decoding. The channel encoding and decoding can adopt a hybrid architecture of LDPC codes and neural networks to enhance adaptability to complex interference such as Doppler shift and rain attenuation.

[0048] Furthermore, in the semantic encoder section, the role of the semantic encoder is to extract compressed features, mainly composed of an interleaved window mechanism, depthwise separable convolutions, and a multilayer perceptron. The interleaved window mechanism adopts a Reshape-Transpose-Reshape strategy, which can obtain global attention in a single computation, and the depthwise separable convolutions reduce complexity, such as... Figure 2As shown. In this model, after `Transpose`, window self-attention is used to calculate self-attention within the local window to improve computational efficiency. Compared to `SwinTransformer`, which requires two consecutive blocks to complete global attention, the staggered window mechanism only needs one block to obtain global attention, greatly reducing model complexity. For the semantic decoder, its role is to restore the image structure through the inverse staggered window mechanism, and to improve the restoration quality with post-processing optimization. The semantic decoder mainly includes the inverse staggered window mechanism, depthwise separable convolution, and an output layer. The inverse staggered window mechanism is used to gradually restore the spatial structure of the image, the depthwise separable convolution is used to supplement detailed information, and the output layer restores the original remote sensing image. By establishing a semantic encoding and decoding framework, not only can bit information be accurately transmitted, but the core meaning of the information can also be extracted, ensuring that the receiver can correctly understand the meaning of the information, thereby optimizing communication efficiency and reducing redundant data transmission. See also... Figure 3 As shown, the channel encoder consists of four convolutional layers and the channel decoder consists of four deconvolutional layers. That is, the channel encoder and decoder adopt a stacked structure, dynamically adjust the coding depth according to SNR, and combine layer-by-layer training with parameter freezing strategy to solve the problem of parameter mismatch between layers, so as to ensure full SNR scenario adaptation.

[0049] As can be seen, in this embodiment of the invention, remote sensing images are decomposed by semantic segmentation, and image content related to the application task to be performed is transmitted first to reduce redundant data. The system also actively adapts to the real-time feedback of the channel signal-to-noise ratio to maintain image recovery quality under varying channel conditions, ultimately achieving high-quality image transmission of remote sensing images in bandwidth-constrained and variable channel scenarios.

[0050] See Figure 4As shown, for low-altitude communication scenarios with limited bandwidth and dynamically changing channels, the remote sensing data processing stage first performs image segmentation, i.e., decomposing the image through semantic segmentation. Then, segmentation feature selection is performed to prioritize the transmission of image content related to the current application task. By actively adapting to real-time signal-to-noise ratio fluctuations, transmission efficiency is improved and image restoration quality is maintained under varying channel conditions. For example, a selection evaluation model based on channel feedback and target orientation is used to obtain the image restoration result evaluation value. Further, based on the semantic selection model, task-related transmission content is selected to meet diverse task requirements and adapt to different channel conditions, thereby optimizing transmission efficiency and improving reconstruction performance. Then, the optimized image to be transmitted is transmitted to the UAV transmitter. The UAV transmitter performs semantic encoding on the optimized image to obtain the semantic encoding result, and performs channel encoding on the semantic encoding result. After the UAV transmitter outputs the image encoding result, it is sent to the ground receiver through a low-altitude wireless channel. The ground receiver performs channel decoding on the image encoding result to obtain the channel decoding result, and performs semantic decoding on the channel decoding result to generate the target image. The technical solution of this invention achieves high-quality image transmission in scenarios with limited bandwidth and variable channels by coordinating remote sensing data processing, UAV launch to ground reception, and combining semantic perception, dynamic adaptation and efficient encoding and decoding technologies.

[0051] In practical applications, the technical solution of this invention can be trained and validated using a high-resolution remote sensing image target detection dataset (NWPU VHR-10) and an agricultural remote sensing dataset. Specifically, after the UAV acquires images, they undergo preprocessing, segmentation, selection, encoding, and transmission. The ground receiver then performs decoding, restoration, and post-processing, ultimately outputting images that meet the requirements of tasks such as agricultural monitoring. Furthermore, an AWGN simulated channel is used, covering multiple scenarios with an SNR range of -10dB to 10dB, and a learning rate of 0.0001. All configurations are implemented based on the PyTorch framework and run on an NVIDIA 3090 GPU. PSNR and SSIM are used as performance metrics to evaluate model performance.

[0052] Deep JSCC, WITT, and SS-DDPM were used as comparison schemes. The WITT scheme uses a WinTransformer to extract hierarchical semantic features of the image and performs feature compression and mapping through an attention mechanism, thereby enhancing the image structure modeling capability. The SS-DDPM scheme introduces a non-Gaussian diffusion framework, directly sampling multiple noisy states from the original data distribution without defining successive transition probabilities. The PSNR and SSIM results of these methods with different SNR values ​​are shown in Tables 1 and 2.

[0053] Table 1

[0054] Table 2

[0055] By comparing the PSNR of different methods under different SNR in Table 1 and the SSIM of different methods under different SNR in Table 2, it can be seen that the technical solution of the present invention has high PSNR and SSIM under different SNR conditions. In typical scenarios such as agricultural monitoring, it can accurately recover information of key areas of the task, verifying the efficiency and robustness of the method of the present invention and demonstrating the effectiveness of the method of the present invention.

[0056] In one embodiment, such as Figure 5 As shown, based on the above-described remote sensing image transmission method based on channel feedback and target guidance, the present invention also provides a remote sensing image transmission device based on channel feedback and target guidance, comprising: Image segmentation module 11 is used to perform semantic segmentation on the acquired original remote sensing image to obtain a semantically segmented image; The segmentation feature selection module 12 is used to determine the region in the original remote sensing image that is most relevant to the application task based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and to obtain the optimized image to be transmitted. Image encoding module 13 is used to input the optimized image to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain an image encoding result; The image decoding module 14 is used to send the image encoding result to the ground receiving end through a low-altitude wireless channel, so that the ground receiving end can perform a decoding operation on the image encoding result to obtain the target image.

[0057] Furthermore, it is worth noting that the working process of the remote sensing image transmission device based on channel feedback and target guidance provided in this embodiment is the same as the working process of the remote sensing image transmission method based on channel feedback and target guidance described above. Therefore, it will not be repeated here. For details, please refer to the working process of the remote sensing image transmission method based on channel feedback and target guidance described above.

[0058] Figure 6 A schematic diagram of the structure of a terminal provided in an embodiment of this application. The terminal may include: The memory 501, the processor 502, and the computer program stored on the memory 501 and capable of running on the processor 502.

[0059] When the processor 502 executes the program, it implements the remote sensing image transmission method based on channel feedback and target guidance provided in the above embodiments.

[0060] Furthermore, the terminal also includes: Communication interface 503 is used for communication between memory 501 and processor 502.

[0061] The memory 501 is used to store computer programs that can run on the processor 502.

[0062] Memory 501 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0063] If the memory 501, processor 502, and communication interface 503 are implemented independently, they can be interconnected via a bus to communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, only one line is used in the diagram, but this does not imply that there is only one bus or one type of bus.

[0064] Optionally, in a specific implementation, if the memory 501, processor 502, and communication interface 503 are integrated on a single chip, then the memory 501, processor 502, and communication interface 503 can communicate with each other through an internal interface.

[0065] Processor 502 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0066] This embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described remote sensing image transmission method based on channel feedback and target guidance.

[0067] Other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of the invention are indicated by the claims.

[0068] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0069] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can read and execute instructions from and from an instruction execution system, apparatus or device).

[0070] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0071] It should be understood that the application of the present invention is not limited to the examples above. Those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A remote sensing image transmission method based on channel feedback and target guidance, characterized in that, The method includes: Semantic segmentation is performed on the acquired raw remote sensing image to obtain a semantically segmented image; Based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and using the semantic segmentation image to determine the region in the original remote sensing image that is most relevant to the application task, an optimized image to be transmitted is obtained. The optimized image is input to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain the image encoding result; The image encoding result is transmitted to a ground receiver via a low-altitude wireless channel, and the ground receiver performs a decoding operation on the image encoding result to obtain the target image.

2. The remote sensing image transmission method based on channel feedback and target guidance according to claim 1, characterized in that, The step of performing semantic segmentation on the acquired original remote sensing image to obtain a semantically segmented image includes: The acquired raw remote sensing images are input into a pre-trained semantic segmentation model for pixel-level classification to obtain semantic segmentation images containing multi-class labels; The semantic segmentation model includes an encoder, a decoder, and a classification layer. The encoder is used to extract features from the remote sensing image using standard two-dimensional convolution and perform max pooling to obtain the corresponding feature map. The decoder is used to upsample the feature map output by the encoder using the pooling index stored in the encoder to obtain the reconstructed upsampled result. The classification layer is used to classify pixels to obtain the category label of each pixel.

3. The remote sensing image transmission method based on channel feedback and target guidance according to claim 2, characterized in that, The optimized image to be transmitted is obtained by using the semantic segmentation image to determine the region in the original remote sensing image most relevant to the application task based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and by using the semantic segmentation image to determine the region in the original remote sensing image most relevant to the application task. Determine the target category of the application task to be executed, and perform a masking operation on the semantic segmentation image according to the target category of the application task to be executed to obtain a binary mask image; Based on the binary mask image, regions in the original remote sensing image that are masked as being irrelevant to the application task are determined, thus obtaining non-important regions; The channel signal-to-noise ratio (SNR) status is evaluated using a preset evaluation function to obtain the evaluation result corresponding to the channel SNR status. Based on the evaluation results, an adaptive mean blurring filter operation is performed on the non-important regions in the original remote sensing image to obtain the processed remote sensing image. A mask correction operation is performed on the processed remote sensing image to obtain an optimized image to be transmitted.

4. The remote sensing image transmission method based on channel feedback and target guidance according to claim 3, characterized in that, The step of performing a masking operation on the semantic segmentation image according to the target category of the application task to be executed, to obtain a binary mask image, includes: The target category of the application task to be executed is mapped and matched with the multi-category label in the semantic segmentation image to obtain the corresponding matching result; A masking operation is performed on the semantic segmentation image based on the matching result to obtain a binary mask image.

5. The remote sensing image transmission method based on channel feedback and target guidance according to claim 4, characterized in that, The step of performing a masking operation on the semantic segmentation image based on the matching result includes: When the matching result indicates that the target category of the application task to be executed matches a category label in the semantic segmentation image, the category pixel value corresponding to the category label that matches the target category in the semantic segmentation image is set to 1. If the matching result indicates that the target category of the application task to be executed does not match a category label in the semantic segmentation image, then the category pixel value corresponding to the category label in the semantic segmentation image that does not match the target category is set to 0.

6. The remote sensing image transmission method based on channel feedback and target guidance according to any one of claims 1 to 5, characterized in that, The encoding operation performed on the optimized image includes: The optimized image is subjected to semantic coding to obtain a semantic coding result, and then a channel coding operation is performed on the semantic coding result.

7. The remote sensing image transmission method based on channel feedback and target guidance according to claim 6, characterized in that, The decoding operation on the image encoding result includes: The image encoding result is subjected to channel decoding to obtain a channel decoding result, and the channel decoding result is subjected to semantic decoding.

8. A remote sensing image transmission device based on channel feedback and target guidance, characterized in that, The device includes: The image segmentation module is used to perform semantic segmentation on the acquired raw remote sensing images to obtain semantically segmented images; The segmentation feature selection module is used to determine the region in the original remote sensing image that is most relevant to the application task based on the current application task to be executed and the real-time feedback of the channel signal-to-noise ratio, and to obtain the optimized image to be transmitted. An image encoding module is used to input the optimized image to the UAV transmitter so that the UAV transmitter can perform an encoding operation on the optimized image to obtain an image encoding result; The image decoding module is used to send the image encoding result to a ground receiver via a low-altitude wireless channel, so that the ground receiver can perform a decoding operation on the image encoding result to obtain the target image.

9. A terminal, characterized in that, include: The system includes a memory, a processor, and a channel-feedback and target-oriented remote sensing image transmission program stored in the memory and executable on the processor. When executed by the processor, the channel-feedback and target-oriented remote sensing image transmission program implements the steps of the channel-feedback and target-oriented remote sensing image transmission method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that can be executed to implement the steps of the remote sensing image transmission method based on channel feedback and target orientation as described in any one of claims 1 to 7.

Images