Unmanned aerial vehicle spectral data compression method and device

Abstract

This invention relates to the fields of remote sensing image processing and agricultural and forestry monitoring technology, and provides a method and apparatus for compressing spectral data from unmanned aerial vehicles (UAVs). The method includes: acquiring spectral image data of a target area and determining a set of multi-indicator functional traits to be inverted in the target area; calculating the joint importance score of each spectral band to all functional traits based on the set of multi-indicator functional traits using a spectral sensitivity analysis model; dividing each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; employing deep learning-based feature-preserving encoding for the core region, transform encoding for the transition region, and feature extraction encoding for the redundant region; and decoding and reconstructing the encoded streams of each region to obtain the target spectral data used for functional trait inversion. This achieves maximum preservation of key spectral information required for multi-indicator collaborative inversion under high compression ratio conditions, ensuring the accuracy of functional trait inversion in the reconstructed data.

Description

Technical Field

[0001] This invention relates to the fields of remote sensing image processing and agricultural and forestry monitoring technology, and in particular to a method and apparatus for compressing spectral data from unmanned aerial vehicles (UAVs). Background Technology

[0002] UAV hyperspectral imaging technology has become an important tool for obtaining detailed vegetation information in fields such as precision agriculture and ecological monitoring, enabling the retrieval of a series of key functional traits such as chlorophyll, water content, and biomass. However, hyperspectral data comprises hundreds of bands and is massive in volume, while the onboard computing power, storage space, and wireless data transmission bandwidth of UAVs are severely limited. This poses a fundamental challenge to achieving large-scale, real-time or near-real-time monitoring and analysis.

[0003] Existing hyperspectral data compression technologies are mainly divided into two categories: one is lossless / lossy compression based on traditional image coding, which can guarantee the quality of reconstructed images or achieve a high compression ratio, but its optimization target is weakly related to the accuracy indicators of subsequent specific remote sensing inversion tasks, and compression distortion may lead to the loss of subtle spectral features or spatial textures that are crucial to inversion; the other is compression methods based on transform domain or deep learning, which have high computational complexity and are difficult to run in real time on resource-constrained UAV embedded platforms.

[0004] Therefore, there is currently a lack of a hyperspectral data compression method that can be executed in real time on the UAV, is driven by the needs of the front-end inversion task, and intelligently distinguishes and prioritizes the preservation of key information required for the inversion of different traits. This lack results in UAVs either being unable to transmit complete data or transmitting data of insufficient quality to meet the accuracy requirements of multi-index collaborative inversion. Summary of the Invention

[0005] This invention provides a method and apparatus for compressing spectral data from unmanned aerial vehicles (UAVs), which addresses the shortcomings of existing general compression methods that are disconnected from subsequent functional trait inversion tasks, resulting in the loss of key spectral information and a significant decrease in inversion accuracy, and are difficult to implement in real-time processing under resource-constrained UAV environments. The invention maximizes the retention of key spectral information required for multi-index collaborative inversion under high compression ratio conditions, ensuring the accuracy of functional trait inversion of reconstructed data.

[0006] This invention provides a method for compressing spectral data from unmanned aerial vehicles (UAVs), comprising: Collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area. The spectral image data is collected by a hyperspectral imager set on a UAV. Based on the aforementioned set of multi-index functional traits, the joint importance score of each spectral band to all functional traits is calculated using a spectral sensitivity analysis model. Based on the joint importance score, each spectral band is divided into a core region, a transition region, and a redundant region; The core region is encoded using deep learning-based feature preservation coding, the transition region is encoded using transform coding, and the redundant region is encoded using feature extraction coding. The encoded streams of each region are decoded and reconstructed to obtain the target spectral data for functional trait inversion.

[0007] In one possible implementation, the method further includes: A spectral sensitivity analysis model is pre-built and trained. The spectral sensitivity analysis model includes a multi-task inversion neural network. The input of the multi-task inversion neural network is a spectral vector, and the output is the predicted value of each functional trait. For each spectral band, the average contribution of the input spectral vector of each spectral band to the predicted value of each functional trait is calculated by the integral gradient method based on the multi-task inversion neural network, and the average contribution is used as the joint importance score.

[0008] In one possible implementation, the method further includes: The division of each spectral band is based on a preset statistical threshold, and the formula is as follows: ; in, b For bands, As the core area, As a transition zone, This is a redundant area. The mean of the joint importance scores for all bands. The standard deviation of the joint importance score for all bands.

[0009] In one possible implementation, the method further includes: Construct a deep learning-based encoder and decoder, wherein the encoder is used to compress and encode the spectral data of the core region, and the decoder is used to decode and reconstruct the encoded bitstream; The encoder and the decoder are jointly trained with the overall loss function as the optimization objective. The overall loss function includes reconstruction loss, task fidelity loss and bit rate estimation loss. The task fidelity loss is determined based on the difference between the first inversion result of the pre-trained multi-task inversion network on the original core region spectral data and the second inversion result on the decoded and reconstructed data.

[0010] In one possible implementation, the method further includes: The bitrate estimation loss is obtained by estimating the probability distribution entropy of the quantized latent features, as shown in the formula: ; in, Loss estimation for bitrate For parameters The fitted marginal probability distribution model is jointly trained with the encoder. For quantized hidden features, For the spectral data of the core area, For encoder.

[0011] In one possible implementation, the method further includes: The spectral data of the transition region is compressed using transform coding with a preset compression ratio; Principal component analysis was used to encode the spectral data of the redundant region, followed by dimensionality reduction and quantization.

[0012] In one possible implementation, the method further includes: Based on the band index information, the reconstructed data of each region are reorganized in the original band order to obtain complete target spectral data; The target spectral data is input into a multi-task inversion network to generate a spatial distribution map of the multi-index functional traits.

[0013] The present invention also provides a UAV spectral data compression device, comprising the following modules: The data acquisition and transmission module is used to collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area. The spectral image data is collected by a hyperspectral imager set on the UAV. An airborne edge processing module is used to calculate the joint importance score of each spectral band to all functional traits based on the multi-index functional trait set using a spectral sensitivity analysis model. The airborne edge processing module is also used to divide each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; The airborne edge processing module is also used to apply deep learning-based feature preservation coding to the core area, transform coding to the transition area, and feature extraction coding to the redundant area. The ground station decoding and inversion module is used to decode and reconstruct the encoded streams of each region to obtain target spectral data for functional trait inversion.

[0014] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the UAV spectral data compression method as described above.

[0015] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the UAV spectral data compression method as described above.

[0016] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the UAV spectral data compression method as described above.

[0017] The present invention provides a method and apparatus for compressing UAV spectral data. This method involves acquiring spectral image data of a target region and determining a set of multi-indicator functional traits to be inverted within that region. The spectral image data is acquired using a hyperspectral imager mounted on the UAV. Based on the set of multi-indicator functional traits, a spectral sensitivity analysis model is used to calculate the joint importance score of each spectral band for all functional traits. According to the joint importance score, each spectral band is divided into a core region, a transition region, and a redundant region. Deep learning-based feature-preserving encoding is applied to the core region, transform encoding to the transition region, and feature extraction encoding to the redundant region. The encoded streams of each region are decoded and reconstructed to obtain the target spectral data for functional trait inversion. Compared to existing general compression methods that are disconnected from subsequent functional trait inversion tasks, leading to the loss of key spectral information and a significant decrease in inversion accuracy, and are difficult to implement in real-time processing under resource-constrained UAV environments, this solution maximizes the retention of key spectral information required for multi-indicator collaborative inversion under high compression ratios, ensuring the accuracy of functional trait inversion in the reconstructed data. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is one of the flowcharts illustrating the UAV spectral data compression method provided by the present invention.

[0020] Figure 2 This is the second flowchart of the UAV spectral data compression method provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the structure of the UAV spectral data compression device provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0024] To facilitate understanding of the embodiments of the present invention, further explanations and descriptions will be provided below with reference to the accompanying drawings and specific embodiments. These embodiments do not constitute a limitation on the embodiments of the present invention.

[0025] Figure 1 This is one of the flowcharts illustrating the UAV spectral data compression method provided by the present invention, such as... Figure 1 As shown, the method includes the following: S11. Collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area.

[0026] This embodiment is primarily applicable to scenarios involving UAV spectral data compression. It acquires raw hyperspectral image data gathered by the UAV. (Width × Height × Number of Bands), and determine the set of multi-index functional traits of mangroves to be inverted. .

[0027] Among them, the spectral image data is collected by a hyperspectral imager set on the drone; the set of multi-index functional traits may include one or more of the following: chlorophyll content, leaf water content, leaf nitrogen content, and biomass of vegetation in the target area.

[0028] S12. Based on the multi-index functional trait set, calculate the joint importance score of each spectral band to all functional traits using a spectral sensitivity analysis model.

[0029] Construct a multi-task sensing spectral sensitivity analysis model Based on historical datasets Calculate each band in the original hyperspectral image For all functional traits Joint importance score .

[0030] S13. Divide each spectral band into a core region, a transition region, and a redundant region based on the joint importance score.

[0031] Based on the joint importance score And based on a preset threshold, all bands are divided into core areas. Transition Zone and redundant areas .

[0032] S14. Deep learning-based feature preservation coding is used for the core area, transform coding is used for the transition area, and feature extraction coding is used for the redundant area.

[0033] Design and implement adaptive coding for different regions: For the core area It employs a deep learning-based encoder. Its training objective is to minimize the task fidelity loss. ; For the transition zone and redundant areas High compression ratio transform encoders are used respectively. and feature extraction encoder .

[0034] S15. Decode and reconstruct the encoded streams of each region to obtain the target spectral data for functional trait inversion.

[0035] The compressed bitstreams generated in each region are packaged and transmitted to the ground terminal, where the corresponding decoder is used for reconstruction, resulting in reconstructed hyperspectral data that can be used for high-precision functional trait inversion. .

[0036] The UAV spectral data compression method provided by this invention involves acquiring spectral image data of a target region and determining the set of multi-indicator functional traits to be inverted in the target region. The spectral image data is acquired using a hyperspectral imager mounted on the UAV. Based on the set of multi-indicator functional traits, a spectral sensitivity analysis model is used to calculate the joint importance score of each spectral band for all functional traits. According to the joint importance score, each spectral band is divided into a core region, a transition region, and a redundant region. Deep learning-based feature-preserving encoding is used for the core region, transform encoding for the transition region, and feature extraction encoding for the redundant region. The encoded streams of each region are decoded and reconstructed to obtain the target spectral data for functional trait inversion. Compared to existing general compression methods that are disconnected from subsequent functional trait inversion tasks, leading to the loss of key spectral information and a significant decrease in inversion accuracy, and are difficult to implement in real-time processing under resource-constrained UAV environments, this method maximizes the retention of key spectral information required for multi-indicator collaborative inversion under high compression ratio conditions, ensuring the accuracy of functional trait inversion in the reconstructed data.

[0037] Figure 2 This is the second flowchart illustrating the UAV spectral data compression method provided by the present invention, as shown below. Figure 2 As shown, the method includes the following: S21. Collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area.

[0038] This embodiment uses the synergistic monitoring of chlorophyll content (Chl), leaf water content (LWC), and leaf nitrogen content (LNC) in mangroves as a scenario for detailed explanation.

[0039] Data on mangroves in the target area was acquired using a hyperspectral imager mounted on a drone, covering a spectral range of 400-1000 nm and a total of 300 bands. A set of traits to be inverted was defined. Simultaneously collect ground samples, measure the corresponding true values ​​of traits, and construct a training sample set. .

[0040] S22. A spectral sensitivity analysis model is pre-constructed and trained. The spectral sensitivity analysis model includes a multi-task inversion neural network. The input of the multi-task inversion neural network is a spectral vector, and the output is the predicted value of each functional trait.

[0041] S23. For each spectral band, the average contribution of the input spectral vector of each spectral band to the predicted value of each functional trait is calculated by the integral gradient method based on the multi-task inversion neural network, and the average contribution is used as the joint importance score.

[0042] Construct a fully connected neural network with a shared bottom-level feature extraction layer and three independent output heads. Train the network using a sample set so that it can extract features from spectral vectors. Accurate prediction .

[0043] Calculate the joint importance score Specifically, this is achieved through the following methods: Construct a multi-task inversion neural network Its input is the spectral vector of a single pixel, and its output is the predicted value of K traits; Using the trained network, calculate the importance score of the b-th band. The average contribution of the input value in this band to the total change in the multi-task output is obtained by the integral gradient method: ; in, For interpolation path, For the spectrum of the nth sample, For baseline spectrum, Define weights for the user with the k-th trait.

[0044] Furthermore, using the trained network The importance of each band b is calculated using the integral gradient method. The baseline spectrum is then set. The average spectrum for all samples. For each sample... Calculate the path integral of its contribution to the prediction of the three traits.

[0045] For example, its contribution to chlorophyll prediction is: ; Based on the three traits, the importance of band b in this sample is determined as follows: .

[0046] Furthermore, the average of all N samples is taken to obtain the final joint importance score for each band: , forming a fractional vector .

[0047] S24. The division of each spectral band is based on the preset statistical threshold.

[0048] calculate mean and standard deviation According to the formula in claim 3, the core area, transition area and redundancy area are divided.

[0049] Specifically, the division of each spectral band is based on a preset statistical threshold, and the formula is as follows: ; in, b For bands, As the core area, As a transition zone, This is a redundant area. The mean of the joint importance scores for all bands. The standard deviation of the joint importance score for all bands.

[0050] S25. Deep learning-based feature preservation coding is used for the core area, transform coding is used for the transition area, and feature extraction coding is used for the redundant area.

[0051] We construct a deep learning-based encoder and decoder. The encoder is used to compress and encode the spectral data of the core region, and the decoder is used to decode and reconstruct the encoded bitstream.

[0052] The encoder and decoder are jointly trained with the overall loss function as the optimization objective. The overall loss function includes reconstruction loss, task fidelity loss, and bit rate estimation loss. The task fidelity loss is determined based on the difference between the first inversion result of the pre-trained multi-task inversion network on the original core region spectral data and the second inversion result on the decoded and reconstructed data.

[0053] Specifically, the core area encoder The training is conducted in an end-to-end manner, and its overall loss function is... Defined as: ; in, For the spectral data of the core area, For the corresponding decoder, For the k-th output head in the pre-trained multi-task inversion network, To estimate the loss for bitrate, , and To balance hyperparameters.

[0054] Bitrate estimation loss By estimating the quantized latent features The probability distribution is approximated by calculating its entropy: in, Loss estimation for bitrate For parameters The fitted marginal probability distribution model is jointly trained with the encoder. For quantized hidden features, For the spectral data of the core area, For encoder.

[0055] Core region compression: Construct a lightweight convolutional autoencoder as and Training is crucial; end-to-end training is employed, with the loss function defined above. The pre-trained inversion network... The parameters are frozen at this point. This step forces the encoder to learn a compressed representation that retains as much as possible after decoding. Its inversion capability.

[0056] Transition zone compression: for The band data within the range is subjected to moderate lossy compression using the 3D-SPHIT algorithm.

[0057] Redundancy compression: for The band data within the range were subjected to significant dimensionality reduction using principal component analysis, retaining the first three principal components, and then the principal component coefficients were encoded using low-bit quantization.

[0058] S26. Based on the band index information, reconstruct the data of each region in the original band order to obtain complete target spectral data.

[0059] S27. Input the target spectral data into a multi-task inversion network to generate a spatial distribution map of the multi-index functional traits.

[0060] The ground station receives packetized data containing three bitstreams, calls the corresponding decoders to reconstruct the data for each, and then reassembles them in their original order into a complete hyperspectral data cube. Finally, the data is reconstructed at the receiving end and used for high-precision inversion.

[0061] The method of this invention significantly reduces the amount of data while maintaining the accuracy of multi-index inversion to the greatest extent, and its performance is significantly better than traditional compression methods that aim at visual fidelity.

[0062] This invention also provides a UAV hyperspectral data compression system for multi-index functional trait inversion, comprising: 1. Airborne Unit Sensors: Hyperspectral imager, POS system.

[0063] Computational core: flight control and data acquisition; 3D-SPHIT transform coding; execution of lightweight multi-task sensitivity analysis models and core region deep learning encoders. .

[0064] Workflow: After data acquisition, the band importance of the current scene is quickly calculated, and partitioning and core area encoding are completed. Transition and redundant area data are accelerated and encoded. Finally, the integrated bitstreams are transmitted via wireless data link.

[0065] 2. Ground station unit Receiver-borne bitstream.

[0066] Run the corresponding decoder on a high-performance workstation or server. The 3D-SPHIT decoder and PCA reconstruction module complete the data reconstruction.

[0067] Call upon a complete, higher-precision multi-task inversion network (which can be used as an airborne network). The enhanced version generates the final trait distribution map.

[0068] The UAV spectral data compression method provided by this invention involves acquiring spectral image data of a target region and determining the set of multi-indicator functional traits to be inverted in the target region. The spectral image data is acquired using a hyperspectral imager mounted on the UAV. Based on the set of multi-indicator functional traits, a spectral sensitivity analysis model is used to calculate the joint importance score of each spectral band for all functional traits. According to the joint importance score, each spectral band is divided into a core region, a transition region, and a redundant region. Deep learning-based feature-preserving encoding is used for the core region, transform encoding for the transition region, and feature extraction encoding for the redundant region. The encoded streams of each region are decoded and reconstructed to obtain the target spectral data for functional trait inversion. Compared to existing general compression methods that are disconnected from subsequent functional trait inversion tasks, leading to the loss of key spectral information and a significant decrease in inversion accuracy, and are difficult to implement in real-time processing under resource-constrained UAV environments, this method maximizes the retention of key spectral information required for multi-indicator collaborative inversion under high compression ratio conditions, ensuring the accuracy of functional trait inversion in the reconstructed data.

[0069] The UAV spectral data compression device provided by the present invention is described below. The UAV spectral data compression device described below can be referred to in correspondence with the UAV spectral data compression method described above.

[0070] Figure 3 This is a schematic diagram of the structure of the UAV spectral data compression device provided by the present invention, specifically including: The data acquisition and transmission module 301 is used to collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area. The spectral image data is collected by a hyperspectral imager mounted on a UAV. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0071] The airborne edge processing module 302 is used to calculate the joint importance score of each spectral band to all functional traits based on the multi-index functional trait set using a spectral sensitivity analysis model. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0072] The airborne edge processing module 302 is further configured to divide each spectral band into a core region, a transition region, and a redundant region based on the joint importance score. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0073] The airborne edge processing module 302 is further configured to employ deep learning-based feature-preserving encoding for the core region, transform encoding for the transition region, and feature extraction encoding for the redundant region. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0074] The ground station decoding and inversion module 303 is used to decode and reconstruct the encoded streams of each region to obtain target spectral data for functional trait inversion. For detailed explanations, please refer to the relevant descriptions in the above method embodiments; they will not be repeated here.

[0075] Figure 4 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 4 As shown, the electronic device may include a processor 410, a communication interface 420, a memory 430, and a communication bus 440, wherein the processor 410, the communication interface 420, and the memory 430 communicate with each other through the communication bus 440. The processor 410 can call logical instructions in the memory 430 to execute a UAV spectral data compression method. The method includes: acquiring spectral image data of a target area and determining a set of multi-indicator functional traits to be inverted in the target area, wherein the spectral image data is acquired by a hyperspectral imager installed on the UAV; calculating the joint importance score of each spectral band to all functional traits based on the set of multi-indicator functional traits using a spectral sensitivity analysis model; dividing each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; applying feature-preserving encoding based on deep learning to the core region, transform encoding to the transition region, and feature extraction encoding to the redundant region; and decoding and reconstructing the encoded streams of each region to obtain target spectral data for functional trait inversion.

[0076] Furthermore, the logical instructions in the aforementioned memory 430 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0077] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the UAV spectral data compression method provided by the above methods. The method includes: acquiring spectral image data of a target area and determining a set of multi-indicator functional traits to be inverted in the target area, wherein the spectral image data is acquired by a hyperspectral imager installed on a UAV; calculating the joint importance score of each spectral band to all functional traits based on the set of multi-indicator functional traits using a spectral sensitivity analysis model; dividing each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; applying feature-preserving encoding based on deep learning to the core region, transform encoding to the transition region, and feature extraction encoding to the redundant region; and decoding and reconstructing the encoded streams of each region to obtain target spectral data for functional trait inversion.

[0078] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the UAV spectral data compression method provided by the above methods. This method includes: acquiring spectral image data of a target region and determining a set of multi-indicator functional traits to be inverted in the target region, wherein the spectral image data is acquired by a hyperspectral imager mounted on a UAV; calculating the joint importance score of each spectral band to all functional traits based on the set of multi-indicator functional traits using a spectral sensitivity analysis model; dividing each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; employing deep learning-based feature-preserving encoding for the core region, transform encoding for the transition region, and feature extraction encoding for the redundant region; and decoding and reconstructing the encoded streams of each region to obtain target spectral data for functional trait inversion.

[0079] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0080] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for compressing spectral data from unmanned aerial vehicles (UAVs), characterized in that, include: Collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area. The spectral image data is collected by a hyperspectral imager set on a UAV. Based on the aforementioned set of multi-index functional traits, the joint importance score of each spectral band to all functional traits is calculated using a spectral sensitivity analysis model. Based on the joint importance score, each spectral band is divided into a core region, a transition region, and a redundant region; The core region employs deep learning-based feature-preserving encoding, the transition region employs transform encoding, and the redundant region employs feature extraction encoding. The encoded streams of each region are decoded and reconstructed to obtain the target spectral data for functional trait inversion.

2. The method according to claim 1, characterized in that, The calculation of the joint importance score of each spectral band for all functional traits based on the multi-index functional trait set using a spectral sensitivity analysis model includes: A spectral sensitivity analysis model is pre-built and trained. The spectral sensitivity analysis model includes a multi-task inversion neural network. The input of the multi-task inversion neural network is a spectral vector, and the output is the predicted value of each functional trait. For each spectral band, the average contribution of the input spectral vector of each spectral band to the predicted value of each functional trait is calculated by the integral gradient method based on the multi-task inversion neural network, and the average contribution is used as the joint importance score.

3. The method according to claim 1 or 2, characterized in that, The process of dividing each spectral band into a core region, a transition region, and a redundant region based on the joint importance score includes: The division of each spectral band is based on a preset statistical threshold, and the formula is as follows: ; in, b For bands, As the core area, As a transition zone, This is a redundant area. The mean of the joint importance scores for all bands. The standard deviation of the joint importance score for all bands.

4. The method according to claim 3, characterized in that, The deep learning-based feature-preserving encoding of the core region includes: Construct a deep learning-based encoder and decoder, wherein the encoder is used to compress and encode the spectral data of the core region, and the decoder is used to decode and reconstruct the encoded bitstream; The encoder and the decoder are jointly trained with the overall loss function as the optimization objective. The overall loss function includes reconstruction loss, task fidelity loss and bit rate estimation loss. The task fidelity loss is determined based on the difference between the first inversion result of the pre-trained multi-task inversion network on the original core region spectral data and the second inversion result on the decoded and reconstructed data.

5. The method according to claim 4, characterized in that, The bitrate estimation loss is obtained by estimating the probability distribution entropy of the quantized latent features, as shown in the formula: ； in, Loss estimation for bitrate For parameters The fitted marginal probability distribution model is jointly trained with the encoder. For quantized hidden features, For the spectral data of the core area, For encoder.

6. The method according to claim 3, characterized in that, The transformation coding applied to the transition region and the feature extraction coding applied to the redundant region include: The spectral data of the transition region is compressed using transform coding with a preset compression ratio; Principal component analysis was used to encode the spectral data of the redundant region, followed by dimensionality reduction and quantization.

7. The method according to claim 1, characterized in that, The decoding and reconstruction of the encoded streams of each region to obtain target spectral data for functional trait inversion includes: Based on the band index information, the reconstructed data of each region are reorganized in the original band order to obtain complete target spectral data; The target spectral data is input into a multi-task inversion network to generate a spatial distribution map of the multi-index functional traits.

8. A UAV spectral data compression device, characterized in that, include: The data acquisition and transmission module is used to collect spectral image data of the target area and determine the set of multi-index functional traits to be inverted in the target area. The spectral image data is collected by a hyperspectral imager set on the UAV. An airborne edge processing module is used to calculate the joint importance score of each spectral band to all functional traits based on the multi-index functional trait set using a spectral sensitivity analysis model. The airborne edge processing module is also used to divide each spectral band into a core region, a transition region, and a redundant region according to the joint importance score; The airborne edge processing module is also used to apply deep learning-based feature preservation coding to the core area, transform coding to the transition area, and feature extraction coding to the redundant area. The ground station decoding and inversion module is used to decode and reconstruct the encoded streams of each region to obtain target spectral data for functional trait inversion.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the UAV spectral data compression method as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the UAV spectral data compression method as described in any one of claims 1 to 7.

Images