Low-altitude intelligent sensor data processing method, device, equipment and medium

Abstract

The application relates to a low-altitude intelligent sensor data processing method, device, equipment and medium, the method comprising the following steps: an airborne edge end extracts at least one hardware feature in a hardware device for binding, initiates a dynamic authorization application to a cloud server, receives a short-term lease and a data encryption key encapsulated by a key encryption key issued by the cloud, verifies the validity of the short-term lease in an isolated memory, obtains the data encryption key in plaintext after verification, and decrypts and loads a preset ciphertext inversion model by using the data encryption key; the airborne edge end calculates a vegetation index physical prior feature according to ground reflectivity data, fuses the vegetation index physical prior feature with a multispectral feature map processed by 1*1 convolution, performs inference inversion in an end-side intelligent acceleration engine by using the fused feature, and obtains multidimensional monitoring indexes of a target area. The application significantly improves the algorithm power reuse rate of the airborne edge end, and reduces power consumption and calculation cost.

Description

Technical Field

[0001] This application relates to the field of unmanned aerial vehicle (UAV) technology, and in particular to a low-altitude intelligent sensor data processing method, corresponding device, electronic equipment, and computer-readable storage medium. Background Technology

[0002] In recent years, with the rapid development of unmanned aerial vehicle (UAV) platform technology, low-altitude remote sensing has been increasingly widely used in precision agriculture, forestry resource surveys, water quality monitoring, and ecological environmental protection. As the core optical sensor for acquiring the physicochemical parameters of surface targets, the quality of data acquired by multispectral cameras and the timeliness of post-processing directly determine the operational value of issuing agricultural prescription maps or making emergency environmental protection decisions.

[0003] However, existing airborne multispectral sensors and their supporting systems still have the following technical shortcomings in practical engineering applications:

[0004] Firstly, the operational mode is rigid, resulting in poor data timeliness. Traditional systems often employ a "capture, store, and calculate" separation operational mode. Multispectral cameras are only responsible for acquiring and storing raw digital images in the air. After the operation, the data must be manually copied to a ground workstation for post-processing operations such as orthophoto stitching, radiometric calibration, and reflectance calculation. The entire processing cycle typically takes several hours or even days, making it difficult to meet the extremely time-sensitive application scenarios such as emergency disaster assessment and real-time crop stress diagnosis, and failing to achieve the practical requirement of "calculate and decide immediately upon takeoff."

[0005] Secondly, edge computing power is not fully activated, and the synergy between hardware and software is low. Currently, the computing boards and sensing modules of airborne cameras are physically separated, and data is exchanged via external Ethernet. This not only limits bandwidth but also poses a high risk of packet loss and transmission latency. Meanwhile, traditional computational algorithms are mostly developed based on general-purpose CPU architectures and have not been deeply adapted to the underlying tensor computation cores of heterogeneous acceleration chips (such as GPUs and NPUs), resulting in the ineffective utilization of airborne AI acceleration capabilities. Furthermore, when large-format, high-resolution images are inferred at the edge, they are prone to overflow due to insufficient GPU memory resources, limiting the real-time processing capabilities at the edge.

[0006] Third, the security of core data assets and algorithm models is weak. In existing deployment schemes, inversion models are mostly stored directly in plaintext on onboard devices, lacking effective encryption protection and anti-reverse engineering measures, making core algorithms vulnerable to theft or illegal copying. Furthermore, the lack of a unified cloud-based management mechanism after equipment sale makes it impossible to dynamically authorize and manage model usage permissions on demand, failing to meet compliance requirements for digital copyright protection and commercial monetization.

[0007] In summary, in order to address the problems commonly found in existing low-altitude remote sensing systems, such as poor processing timeliness due to the separation of "image capture, storage, and computation", large-format inference memory overflow caused by low edge computing power and hardware synergy, and weak algorithm security protection due to the plaintext storage of inversion models and the lack of dynamic cloud management, the applicant has made corresponding explorations to solve these problems. Summary of the Invention

[0008] The purpose of this application is to solve the above-mentioned problems by providing a low-altitude intelligent sensor data processing method, corresponding device, electronic device and computer-readable storage medium.

[0009] To achieve the various objectives of this application, the following technical solution is adopted:

[0010] A low-altitude smart sensor data processing method proposed for one of the purposes of this application includes:

[0011] The airborne edge device acquires multi-band raw images of the target area and the corresponding raw digital response signals from a multispectral array camera.

[0012] The airborne edge terminal performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal to generate radiance data.

[0013] The airborne edge terminal acquires solar down-direction normal irradiance data synchronously collected by the airborne down-direction optical sensor, acquires aircraft attitude data fed back by the airborne inertial navigation unit, and calculates the solar incidence zenith angle in real time based on the attitude data and the solar altitude angle, and obtains sky diffuse irradiance according to a preset radiative transfer model; the irradiance data is then subjected to illumination compensation and atmospheric correction using the solar down-direction normal irradiance, the solar incidence zenith angle, and the sky diffuse irradiance, and the surface reflectance data is output.

[0014] The airborne edge device extracts at least one hardware feature from the hardware device and binds it, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key, and uses the data encryption key to decrypt and load a preset ciphertext inversion model.

[0015] The airborne edge device calculates the physical prior features of vegetation index based on the surface reflectance data, fuses the physical prior features of vegetation index with the multispectral feature map after 1×1 convolution, and uses the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators of the target area.

[0016] Optionally, the airborne edge terminal performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal to generate radiance data, including:

[0017] The real-time internal temperature and current exposure time of the low-altitude intelligent sensor are obtained. The dark current background matrix parameters at the pixel level are determined based on the real-time internal temperature and the current exposure time. The dark current background matrix parameters are subtracted from the original digital response signal to obtain the dark current corrected response signal.

[0018] The lens vignetting polynomial correction coefficients are obtained. The vignetting polynomial correction coefficients are sixth-order polynomials with the radial distance from the pixel coordinates to the optical principal point as the independent variable. The sixth-order polynomials are used to perform lens vignetting compensation on the response signal after dark current correction to obtain the vignetting compensated response signal.

[0019] Obtain the factory absolute calibration coefficients for each band, and use the factory absolute calibration coefficients to map the vignetting compensated response signal into radiance data.

[0020] Optionally, the step of performing illumination compensation and atmospheric correction on the radiance data using the solar downward normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance, and outputting the surface reflectance data, includes:

[0021] Obtain the solar downward normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance;

[0022] Calculate the cosine value corresponding to the solar incident zenith angle, calculate the first product between the solar downward normal irradiance and the cosine value, and obtain the surface direct irradiance component.

[0023] Calculate the first sum between the direct surface irradiance component and the diffuse sky irradiance to obtain the total incident irradiance on the surface;

[0024] Calculate the first ratio between the radiance data and the total incident irradiance on the Earth's surface, and multiply the first ratio by π to construct a surface reflectance calculation model and obtain surface reflectance data.

[0025] Optionally, the airborne edge device extracts at least one hardware feature from the hardware device for binding, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by a key encryption key from the cloud, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key, and uses the data encryption key to decrypt and load a preset ciphertext inversion model, including the following steps:

[0026] Extract at least one hardware feature from the hardware device and bind it as a device identity credential when the airborne edge terminal initiates a dynamic authorization request to the cloud server;

[0027] Send an authorization request carrying the hardware features to the cloud server, and receive a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, wherein the short-term lease includes a valid usage period;

[0028] The short-term lease and the data encryption key encapsulated by the key encryption key are stored in isolated memory, and the validity of the short-term lease is verified in the isolated memory;

[0029] After successful verification, the data encryption key encapsulated by the key encryption key is decapsulated in the isolated memory using the key encryption key to obtain the plaintext data encryption key;

[0030] The preset ciphertext inversion model is decrypted in the isolated memory using the data encryption key of the plaintext, and the decrypted inversion model is loaded into a streaming deserialization. The decryption and loading processes are both completed in the isolated memory to perform zero-disk loading of plaintext model data.

[0031] Optionally, the airborne edge computing unit calculates physical prior features of vegetation indices based on the surface reflectance data, fuses these physical prior features with a multispectral feature map processed by 1×1 convolution, and uses the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multidimensional monitoring indicators of the target area. This includes the following steps:

[0032] The surface reflectance data is acquired, and the surface reflectance data includes reflectance information of each channel of the multispectral array camera;

[0033] The physical prior features of vegetation index are calculated based on the surface reflectance data. These physical prior features of vegetation index are used to characterize the physiological and biochemical parameter status of vegetation in the target area.

[0034] The multispectral feature map acquired by the multispectral array camera is obtained, and the multispectral feature map is subjected to cross-channel feature integration and nonlinear activation through 1×1 convolution to obtain the dimension-reduced depth feature map.

[0035] The physical prior features of the vegetation index are fused with the dimensionality-reduced depth feature map along the channel dimension to obtain the fused features.

[0036] The fusion feature input terminal side intelligent acceleration engine performs inference inversion to obtain multi-dimensional monitoring indicators of the target area.

[0037] Optionally, the hardware features include one or more of the following: MAC address, CPU serial number, or trusted platform module fingerprint features.

[0038] Optionally, the physical prior features of the vegetation index include at least one of the normalized vegetation index, the red-edge normalized index, and the leaf nitrogen index.

[0039] A low-altitude smart sensor data processing device provided for another purpose of this application includes:

[0040] The sensing and acquisition module is configured to acquire multi-band raw images of the target area and the corresponding raw digital response signals from the multispectral array camera at the airborne edge.

[0041] The physical correction module is configured to perform dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal at the airborne edge end, thereby generating radiance data.

[0042] The security authentication module is configured to acquire solar down-direction irradiance data synchronously collected by the airborne down-direction optical sensor at the airborne edge end, acquire aircraft attitude data fed back by the airborne inertial navigation unit, and calculate the solar incidence zenith angle in real time based on the attitude data and the solar altitude angle, and obtain the sky diffuse irradiance according to the preset radiative transfer model; perform illumination compensation and atmospheric correction on the irradiance data using the solar down-direction normal irradiance, the solar incidence zenith angle, and the sky diffuse irradiance, and output the surface reflectance data;

[0043] The inversion inference module is configured to extract at least one hardware feature from the hardware device at the airborne edge and bind it, initiate a dynamic authorization request to the cloud server, receive a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, verify the validity of the short-term lease in isolated memory, and after successful verification, decapsulate and obtain the plaintext data encryption key, and use the data encryption key to decrypt and load the preset ciphertext inversion model.

[0044] The business output module is configured to calculate the physical prior features of vegetation index based on the surface reflectance data at the airborne edge, fuse the physical prior features of vegetation index with the multispectral feature map after 1×1 convolution, and use the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators of the target area.

[0045] An electronic device provided for another purpose of this application includes a central processing unit and a memory, the central processing unit being configured to invoke and run a computer program stored in the memory to perform the steps of the low-altitude smart sensor data processing method of this application.

[0046] A computer-readable storage medium is provided for another purpose of this application, which stores, in the form of computer-readable instructions, a computer program implemented according to the low-altitude smart sensor data processing method, which, when called by a computer, executes the steps included in the corresponding method.

[0047] Compared to existing technologies, this application addresses the common problems in existing low-altitude remote sensing systems, such as poor processing timeliness due to the separation of "image capture, storage, and computation," large-format inference memory overflow caused by low edge computing power and hardware synergy, and weak algorithm security protection due to plaintext storage of inversion models and lack of dynamic cloud management. This application includes, but is not limited to, the following beneficial effects:

[0048] Firstly, this application places the entire data processing flow from raw digital response signals to operational monitoring indicators at the airborne edge, completing radiometric calibration, reflectivity calculation, and AI inference in real time during flight. This eliminates the need for manual data copying to ground workstations for post-processing after the operation is completed, reducing the traditional hourly or even dayly processing cycle to the second level. This solves the problem of poor data timeliness caused by the separation of "photographing, storing, and calculating" in existing technologies, achieving "calculation and decision-making immediately upon flight," and meeting the needs of application scenarios with extremely high timeliness requirements, such as emergency disaster assessment and real-time crop stress diagnosis.

[0049] Secondly, this application achieves a direct physical connection between the multispectral sensing array and the AI ​​acceleration chip through a high-speed board-level bus, eliminating the bandwidth limitations and packet loss risks of traditional Ethernet transmission. At the same time, through the adaptive block division mechanism in the edge-side intelligent acceleration engine, the input data is dynamically divided according to the chip's Shape parameters. Combined with the 1×1 convolution acceleration of the tensor calculation core and INT8 / FP16 quantization inference, it effectively avoids memory overflow of high-resolution multispectral images during edge inference, improving the computing power utilization efficiency and processing stability of the airborne edge.

[0050] Thirdly, this application employs a dynamic dark current subtraction and a sixth-order halo polynomial compensation model in the radiometric calibration stage. In the reflectance calculation stage, it introduces a three-dimensional compensation mechanism combining DLS measured illumination, real-time pose calculation, and a radiative transfer model, ensuring that the real-time generated surface reflectance data possesses quantitative accuracy comparable to that of ground post-processing. In the AI ​​inversion stage, physical prior features of vegetation indices such as NDVI, NDRE, and LCI are calculated and fused with depth features extracted by 1×1 convolution in the channel dimension, forming a dual-drive mode of "physical prior + depth features." This guides the neural network to have a clear business orientation, improving both inversion accuracy and model interpretability.

[0051] Fourth, this application employs a DEK / KEK dual-key encryption mechanism to protect the inversion model. Combined with device hardware feature binding and dynamically distributed short-term leases from the cloud, model decryption and streaming deserialization loading are completed in isolated memory, enabling zero-disk execution of plaintext model data. This mechanism ensures that the model is always stored in encrypted form at the edge, only briefly decrypted and executed in authorized isolated memory, effectively preventing the core algorithm from being reverse-engineered. Simultaneously, the automatic circuit breaker mechanism suspends inference capabilities upon expiration of the short-term lease, enabling on-demand licensing and dynamic management after device sale, providing compliant technical support for digital copyright protection and commercial monetization.

[0052] Fifth, this application adopts a split network model architecture, extracts common features at the bottom layer through a unified backbone network, and splits multiple inference pipeline branches for different business indicators in the backend to perform concurrent inversion. In one inference, multi-dimensional monitoring indicators such as leaf area index, vegetation health, and nitrogen content are output simultaneously, avoiding the computational redundancy caused by deploying independent models for each indicator, significantly improving the computing power reuse rate of the airborne edge, and reducing power consumption and computational overhead. Attached Figure Description

[0053] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0054] Figure 1 This is a flowchart illustrating the low-altitude intelligent sensor data processing method in an embodiment of this application.

[0055] Figure 2 This is a schematic diagram of the external three-dimensional structure of the low-altitude intelligent sensor in the front side of the embodiment of this application.

[0056] Figure 3 This is a schematic diagram of the external three-dimensional structure of the low-altitude intelligent sensor from the side and rear of the embodiment of this application;

[0057] Figure 4This is a cross-sectional view of the internal casing of the low-altitude intelligent sensor in the embodiments of this application and a schematic diagram of the multi-layer motherboard stack;

[0058] Figure 5 This is a front view of the multispectral lens array at the front end of the low-altitude smart sensor in an embodiment of this application;

[0059] Figure 6 This is a schematic block diagram of the low-altitude intelligent sensor data processing device in the embodiments of this application;

[0060] Figure 7 This is a schematic diagram of the structure of the computer device in the embodiments of this application. Detailed Implementation

[0061] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain this application, and should not be construed as limiting this application.

[0062] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It should be understood that when we say an element is “connected” or “coupled” to another element, it can be directly connected or coupled to the other element, or there may be intermediate elements. Furthermore, “connected” or “coupled” as used herein can include wireless connections or wireless coupling. The term “and / or” as used herein includes all or any units and all combinations of one or more associated listed items.

[0063] Those skilled in the art will understand that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0064] Those skilled in the art will understand that the terms "client," "terminal," and "terminal device" as used herein include both devices that receive wireless signals, devices that only possess wireless signal receiver capabilities without transmission capabilities, and devices with receiving and transmitting hardware, devices that have receiving and transmitting hardware capable of bidirectional communication over a bidirectional communication link. Such devices may include: cellular or other communication devices such as personal computers or tablets, having single-line displays, multi-line displays, or cellular or other communication devices without multi-line displays; PCS (Personal Communications Service) that can combine voice, data processing, fax, and / or data communication capabilities; PDAs (Personal Digital Assistants) that may include radio frequency receivers, pagers, internet / intranet access, web browsers, notebooks, calendars, and / or GPS (Global Positioning System) receivers; and conventional laptops and / or handheld computers or other devices that have and / or include radio frequency receivers. As used herein, "client," "terminal," and "terminal device" can be portable, transportable, installed in a means of transportation (air, sea, and / or land), or suitable and / or configured to operate locally and / or in a distributed manner, operating in any other location on Earth and / or in space. "Client," "terminal," and "terminal device" as used herein can also be a communication terminal, an internet access terminal, or a music / video playback terminal, such as a PDA, a MID (Mobile Internet Device), and / or a mobile phone with music / video playback capabilities, or a smart TV, set-top box, etc.

[0065] The hardware referred to by the names "server," "client," and "service node" in this application is essentially an electronic device with the equivalent capabilities of a personal computer. It is a hardware device with the necessary components revealed by the von Neumann architecture, such as a central processing unit (including an arithmetic logic unit and a control unit), memory, input devices, and output devices. The computer program is stored in its memory, and the central processing unit loads the program stored in the secondary storage into the main memory to run it, execute the instructions in the program, and interact with the input and output devices to complete specific functions.

[0066] It should be noted that the concept of "server" used in this application can also be extended to the case of server clusters. Based on the network deployment principles understood by those skilled in the art, the servers should be logically divided. Physically, these servers can be independent of each other but accessible through interfaces, or they can be integrated into a single physical computer or a computer cluster. Those skilled in the art should understand this flexibility and should not use it to constrain the implementation of the network deployment method in this application.

[0067] Unless otherwise expressly specified, one or more technical features of this application may be deployed on a server and accessed by a client through remote invocation of the online service interface provided by the server, or they may be directly deployed and run on a client for access.

[0068] Unless otherwise specified, the neural network models referenced or potentially referenced in this application may be deployed on a remote server and invoked remotely on the client, or deployed on a client with the capability to invoke directly. In some embodiments, when running on the client, the corresponding intelligence may be acquired through transfer learning in order to reduce the requirements on the client's hardware resources and avoid excessive consumption of the client's hardware resources.

[0069] Unless otherwise specified, all data involved in this application may be stored remotely on a server or on a local terminal device, as long as it is suitable for use by the technical solution of this application.

[0070] Those skilled in the art will understand that although the various methods in this application are described based on the same concept and thus present commonality among them, they can be performed independently unless otherwise specified. Similarly, the various embodiments disclosed in this application are all based on the same inventive concept; therefore, concepts expressed in the same way, as well as concepts that are appropriately changed for convenience but are expressed differently, should be understood equivalently.

[0071] Unless otherwise expressly stated, the various embodiments disclosed in this application can be combined in a cross-cutting manner to flexibly construct new embodiments, as long as such combination does not depart from the inventive spirit of this application and can meet the needs of the prior art or solve a certain deficiency in the prior art. Those skilled in the art should be aware of such modifications.

[0072] Please see Figure 1 In one embodiment of the low-altitude smart sensor data processing method of this application, the method includes:

[0073] Step S10: The airborne edge device acquires the multi-band raw images of the target area and the corresponding raw digital response signals collected by the multispectral array camera.

[0074] The airborne edge device acquires multi-band raw images of the target area and the corresponding raw digital response signals from a multispectral array camera.

[0075] Specifically, please refer to Figures 2 to 5 The physical design of low-altitude intelligent sensors fully considers the high degree of integration between optical precision and edge computing performance, such as... Figure 2 and Figure 5 As shown, its front-end sensing panel employs a precise hexagonal array arrangement, placing one RGB visible light channel at the geometric center, dedicated to capturing ultra-high resolution surface texture features and spatial contour information. Around this central channel are six independent multispectral narrowband channels, their center wavelengths precisely defined as 450nm, 555nm, 660nm, 720nm, 750nm, and 840nm, respectively. The multispectral array camera, triggered by the flight control system, performs high-frequency synchronous exposure acquisition, obtaining multi-band raw images of the target area and corresponding raw digital response signals. This compact "1+6" ring array layout not only significantly reduces parallax errors during low-altitude flight by minimizing the physical distance between optical principal points, but also provides a stable physical foundation for back-end multispectral feature fusion and sub-pixel-level image registration. Inside the casing, a board-level high-speed bus enables closed-loop interconnection between the sensing array and the computing chip, allowing the raw digital response signals to be directly transmitted to the edge-side intelligent acceleration engine via the board-level high-speed bus. Figure 2 and Figure 3 As shown, the cylindrical structure on top of the casing serves as the physical mounting and communication interface for docking with the drone platform. The Ambient Light Compensation Module (DLS) is deployed as an independent component on the top of the drone's back and is connected to the system via a cable to ensure that its light-sensing window always faces vertically towards the sky in various flight attitudes of the drone, avoiding obstruction by the fuselage and rotor.

[0076] Step S20: The airborne edge terminal performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal to generate radiance data.

[0077] After acquiring the multi-band raw images of the target area and the corresponding raw digital response signals collected by the multispectral array camera, the airborne edge terminal performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on the factory absolute calibration coefficient on the raw digital response signals to generate radiance data.

[0078] In some embodiments, the steps of performing dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal by the airborne edge terminal to generate radiance data include:

[0079] Step S201: Obtain the real-time internal temperature and current exposure time of the low-altitude intelligent sensor, determine the pixel-level dark current background matrix parameters based on the real-time internal temperature and the current exposure time, and subtract the dark current background matrix parameters from the original digital response signal to obtain the dark current corrected response signal.

[0080] Step S202: Obtain the lens vignetting polynomial correction coefficient. The vignetting polynomial correction coefficient is a sixth-order polynomial with the radial distance from the pixel coordinate to the optical principal point as the independent variable. The sixth-order polynomial is used to perform lens vignetting compensation on the response signal after dark current correction to obtain the vignetting compensated response signal.

[0081] Step S203: Obtain the factory absolute calibration coefficients for each band, and use the factory absolute calibration coefficients to map the vignetting compensated response signal into radiance data.

[0082] Specifically, in the data processing flow, the preprocessing unit first performs rigorous radiometric calibration on the acquired raw digital response signal. A dynamic dark current subtraction mechanism is established at the system's underlying layer. The system acquires the real-time internal temperature of the low-altitude intelligent sensor. and current exposure time The pixel-level dark current background matrix parameters are determined based on the real-time internal temperature and the current exposure time. The dark current background matrix parameters are subtracted from the original digital response signal. The response signal after dark current correction was obtained.

[0083] Subsequently, to address the off-axis image field energy attenuation caused by the lens edges, the system introduces a high-order polynomial flat-field correction model for vignetting compensation. The system obtains lens vignetting polynomial correction coefficients, which are sixth-order polynomials with the radial distance from the pixel coordinates to the optical principal point as the independent variable. These sixth-order polynomials are used to perform lens vignetting compensation on the dark current-corrected response signal, resulting in a vignetting-compensated response signal. Finally, the system obtains the factory absolute calibration coefficients for each band and uses these coefficients to map the vignetting-compensated response signal into radiance data. The specific radiometric calibration calculation model is as follows:

[0084]

[0085] in, Represents the band at pixel coordinates (x, y). Radiance data; Represents the band at pixel coordinates (x, y). The original digital response signal (Digital Number); Indicates the real-time internal temperature of the sensor and current exposure time Related dark current background matrix parameters; This represents the system gain coefficient of the sensor; Indicates the current exposure time; This represents the polynomial correction coefficient for lens vignetting.

[0086] Lens vignetting polynomial correction coefficient Vignetting compensation is performed using a 6th-order polynomial; lens vignetting polynomial correction coefficients. Represented as:

[0087]

[0088] In the formula, to This indicates the preset lens vignetting and distortion compensation parameters; The radial Euclidean distance from the pixel coordinates (x, y) to the optical principal point of the image is expressed by the following formula: ,in, , Represents the pixel coordinates of a cell; , This indicates the coordinates of the lens's principal optical point.

[0089] Through the above-mentioned radiometric calibration calculation model, the system realizes the real-time conversion from raw photoelectric signals to physically meaningful radiance data, laying a foundation for accuracy in subsequent quantitative calculations.

[0090] Step S30: The airborne edge terminal acquires the solar down-direction normal irradiance data synchronously collected by the airborne down-direction optical sensor, acquires the aircraft attitude data fed back by the airborne inertial navigation unit, and calculates the solar incident zenith angle in real time based on the attitude data and the solar altitude angle, and obtains the sky diffuse irradiance according to the preset radiative transfer model; the irradiance data is then subjected to illumination compensation and atmospheric correction using the solar down-direction normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance, and the surface reflectance data is output.

[0091] The airborne edge device performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the raw digital response signal to generate radiance data. Then, the airborne edge device acquires solar down-direction normal irradiance data synchronously collected by the airborne down-direction optical sensor, acquires aircraft attitude data fed back by the airborne inertial navigation unit, and calculates the solar zenith angle in real time based on the attitude data and the solar altitude angle. It also obtains the sky diffuse irradiance according to a preset radiative transfer model. The radiance data is then subjected to illumination compensation and atmospheric correction using the solar down-direction normal irradiance, the solar zenith angle, and the sky diffuse irradiance, outputting surface reflectance data. The radiative transfer model refers to an atmospheric radiative transfer model. These models calculate the signal received by the sensor by simulating the physical processes of scattering and absorption of solar radiation as it passes through the atmosphere.

[0092] In some embodiments, the step of performing illumination compensation and atmospheric correction on the radiance data using the solar downward normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance, and outputting surface reflectance data, includes:

[0093] Step S301: Obtain the solar downward normal irradiance, solar incident zenith angle, and sky diffuse irradiance;

[0094] Step S302: Calculate the cosine value corresponding to the solar incident zenith angle, calculate the first product between the solar downward normal irradiance and the cosine value, and obtain the surface direct irradiance component.

[0095] Step S303: Calculate the first sum between the direct surface irradiance component and the diffuse sky irradiance to obtain the total incident irradiance on the surface;

[0096] Step S304: Calculate the first ratio between the radiance data and the total incident irradiance of the earth's surface, and multiply the first ratio by π to construct a surface reflectance calculation model and obtain the surface reflectance data.

[0097] Specifically, the system then uses DLS sensor data and onboard attitude to perform real-time atmospheric correction. The core of this process lies in using the downward-facing normal irradiance of the sun. Sun's angle of incidence at the zenith and sky diffuse irradiance Radiance Illumination compensation and atmospheric correction are performed. This includes the downward-facing normal irradiance. The solar zenith angle θ is synchronously acquired by the airborne downlink optical sensor; it is calculated in real time based on the UAV attitude angles (Pitch, Roll, Yaw) fed back by the airborne inertial navigation unit and the solar altitude angle; the sky diffuse irradiance is... The surface reflectance is estimated based on the radiative transfer model, where the real-time solution model for surface reflectance is expressed as:

[0098]

[0099] Based on the zenith angle of the sun's incidence Calculate the cosine value Calculate the downward normal irradiance of the sun. and The first product between them yields the surface direct irradiance component. ; Calculate the direct surface irradiance component and the diffuse sky irradiance. The first sum between the two values ​​yields the total incident irradiance on the Earth's surface. ; Calculate radiance data Total irradiance incident on the Earth's surface The first ratio between the two values ​​is taken, and then multiplied by pi (π) to obtain the surface reflectance data. .

[0100] This real-time surface reflectance calculation model ensures that the generated reflectance matrix has extremely high consistency under different light intensities and angles, eliminating spectral distortion caused by cloud cover, changes in solar azimuth, and aircraft shaking, thus meeting the requirements for quantitative inversion.

[0101] Step S40: The airborne edge terminal extracts at least one hardware feature from the hardware device and binds it, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, verifies the validity of the short-term lease in isolated memory, and after verification, decapsulates and obtains the plaintext data encryption key, and uses the data encryption key to decrypt and load the preset ciphertext inversion model.

[0102] The airborne edge device acquires solar down-direction irradiance data synchronously collected by the airborne down-direction optical sensor, acquires aircraft attitude data fed back by the airborne inertial navigation unit, and calculates the solar zenith angle in real time based on the attitude data and the solar altitude angle. It then obtains the sky diffuse irradiance according to a preset radiative transfer model. After performing illumination compensation and atmospheric correction on the irradiance data using the solar down-direction irradiance, the solar zenith angle, and the sky diffuse irradiance, and outputting the surface reflectance data, the airborne edge device extracts at least one hardware feature from the hardware device for binding, initiates a dynamic authorization request to the cloud server, receives a short-term lease from the cloud and a data encryption key encapsulated by a key encryption key, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key. The data encryption key is then used to decrypt and load a preset ciphertext inversion model. The hardware features include one or more of the following: MAC address, CPU serial number, or Trusted Platform Module (TPM) fingerprint features.

[0103] In some embodiments, the airborne edge device extracts at least one hardware feature from the hardware device for binding, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by a key encryption key from the cloud, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key, and uses the data encryption key to decrypt and load a preset ciphertext inversion model, including:

[0104] Step S401: Extract at least one hardware feature from the hardware device and bind it as the device identity credential when the airborne edge terminal initiates a dynamic authorization request to the cloud server;

[0105] Step S402: Send an authorization request carrying the hardware features to the cloud server, and receive a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud. The short-term lease includes a valid usage period.

[0106] Step S403: Store the short-term lease and the data encryption key encapsulated by the key encryption key in isolated memory, and verify the validity of the short-term lease in the isolated memory;

[0107] Step S404: After successful verification, the data encryption key encapsulated by the key encryption key is decapsulated in the isolated memory using the key encryption key to obtain the plaintext data encryption key;

[0108] Step S405: Decrypt the preset ciphertext inversion model in the isolated memory using the data encryption key of the plaintext, and load the decrypted inversion model into a streaming deserialization. The decryption and loading processes are both completed in the isolated memory to perform zero-disk loading of plaintext model data.

[0109] Specifically, the system integrates a cloud-edge collaborative authentication mechanism. All inversion models are encapsulated using symmetric encryption based on a two-layer key (DEK / KEK) before leaving the factory and stored in the onboard device as ciphertext engine files. After device startup, at least one hardware feature (such as MAC address, CPU serial number, or TPM fingerprint) is extracted from the hardware device and bound to it as the device identity credential when the onboard edge terminal initiates a dynamic authorization request to the cloud server, sending an authorization request carrying the hardware feature to the cloud server. After verifying the legality of the hardware feature, the cloud server issues a short-term lease containing a valid usage period and a data encryption key (DEK) encapsulated in a key encryption key (KEK). The onboard edge terminal stores the short-term lease and the KEK-encapsulated DEK in isolated memory, verifying the validity of the short-term lease in the isolated memory. After successful verification, the KEK-encapsulated DEK is decapsulated using the KEK in the isolated memory to obtain the plaintext DEK. Subsequently, the pre-set ciphertext inversion model is decrypted using the plaintext DEK in the isolated memory, and the decrypted inversion model is then loaded via streaming deserialization. The entire decryption and loading process is completed in the isolated memory, ensuring that the plaintext model does not end up on external storage media. This achieves zero-disk loading of plaintext model data and realizes a secure business closed loop with dynamically controllable authorization and mandatory circuit breaker termination upon expiration.

[0110] In the final presentation stage of the business loop, the high-precision multispectral imagery and multi-indicator monitoring results data processed at the edge will be uploaded in real time to a designated cloud server via an air-to-ground wireless data link. This server is equipped with a monitoring data visualization platform, allowing end users to log in anytime, anywhere via a webpage or mobile device to view plot-level heat maps, growth assessments, and decision reports for the target area. This completely breaks the spatial and temporal limitations of traditional remote sensing operations, achieving seamless integration from front-end perception to back-end display.

[0111] Step S50: The airborne edge terminal calculates the physical prior features of vegetation index based on the surface reflectance data, fuses the physical prior features of vegetation index with the multispectral feature map after 1×1 convolution, and uses the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators of the target area.

[0112] The airborne edge device extracts at least one hardware feature from the hardware device and binds it, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by a key encryption key from the cloud, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key. Using the data encryption key, it decrypts and loads a pre-set ciphertext inversion model. The airborne edge device then calculates the physical prior features of the vegetation index based on the surface reflectance data, fuses these physical prior features with a multispectral feature map processed by 1×1 convolution, and uses the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators for the target area. The physical prior features of the vegetation index include at least one of the normalized vegetation index, red-edge normalized index, and leaf nitrogen index.

[0113] In some embodiments, the steps of calculating physical prior features of vegetation index based on the surface reflectance data by the airborne edge terminal, fusing the physical prior features of vegetation index with a multispectral feature map processed by 1×1 convolution, and performing inference inversion in the edge-side intelligent acceleration engine using the fused features to obtain multidimensional monitoring indicators of the target area include:

[0114] Step S501: Obtain the surface reflectance data, which includes reflectance information of each channel of the multispectral array camera;

[0115] Step S502: Calculate the physical prior features of vegetation index based on the surface reflectance data. The physical prior features of vegetation index are used to characterize the physiological and biochemical parameter status of vegetation in the target area.

[0116] Step S503: Obtain the multispectral feature map acquired by the multispectral array camera, and perform cross-channel feature integration and nonlinear activation on the multispectral feature map through 1×1 convolution to obtain the dimension-reduced depth feature map.

[0117] Step S504: The physical prior features of the vegetation index are fused with the reduced depth feature map in the channel dimension to obtain the fused features;

[0118] Step S505: The fused feature input terminal side intelligent acceleration engine performs inference inversion to obtain multi-dimensional monitoring indicators of the target area.

[0119] Specifically, high-precision surface reflectance data after physical correction is pushed to an intelligent scheduling inversion engine that supports multi-model compatible deployment. The surface reflectance data includes reflectance information from each channel of a multispectral array camera. To further improve the inversion accuracy for different crops such as wheat, corn, and rice, and their different growth stages, the system performs physical prior feature extraction before inference. The system calculates vegetation index physical prior features in real time in volatile memory based on the surface reflectance data. These vegetation index physical prior features characterize the physiological and biochemical parameter status of vegetation in the target area, specifically calculating vegetation indices (PIs) sensitive to agronomic indicators. Specifically, for crop cover evaluation, the system schedules the normalized vegetation index calculation formula:

[0120]

[0121] For detecting subtle fluctuations in chlorophyll content, the system schedules the red-edge normalization index calculation formula:

[0122]

[0123] For real-time assessment of crop nutrient stress and nitrogen content, the system schedules the leaf nitrogen index calculation formula as follows:

[0124]

[0125] in, , , These represent reflectance data for the 840nm, 720nm, and 660nm wavelength bands, respectively.

[0126] Simultaneously, the system acquires multispectral feature maps from a multispectral array camera. These feature maps are then subjected to 1×1 convolution with depth-weighted and nonlinear activation across spectral channels to obtain a dimensionality-reduced depth feature map. The aforementioned vegetation index physical prior features and the dimensionality-reduced depth feature map are subsequently fused along the channel dimension using a feature concatenation operation to obtain fused features. This dual-drive mode of "physical prior indices + depth-reduced features" guides the neural network to have a clear business orientation from the initial stage of feature map acquisition. For example, when retrieving wheat chlorophyll content, the system dynamically adjusts the weight contribution of the 720nm (red-edge band) band through 1×1 convolution, and, in conjunction with the attached NDRE exponential component, enables the neural network to more sensitively capture physiological characteristic changes during the vegetation greening period.

[0127] The fused features are input into the edge-side intelligent acceleration engine for inference inversion, yielding multi-dimensional monitoring indicators for the target area. Addressing the challenge of limited edge-side memory resources, the inversion engine employs a split-cascade network model architecture to support efficient concurrent monitoring of multiple indicators. This architecture allows the system to function as a general algorithmic foundation, extracting common features at a shallow level using a unified shared backbone network, and outputting differentiated results by switching between different inference pipeline branches at deeper levels. For example, in a monitoring task targeting the corn jointing stage, after extracting shared backbone features, the system calculates the leaf area index (LAI) map through task branch A, and simultaneously calculates the vegetation health evaluation map through task branch B. The final inversion mapping relationship of the business indicator Y follows a nonlinear composite function. This significantly improved the computing power reuse rate, among which, This represents the final operational monitoring indicators output. For example, in a corn jointing stage monitoring task, different branches can output different indicators such as leaf area index (LAI) and vegetation health. This represents the mapping function for the split cascaded network. It signifies the complex nonlinear transformation performed by the entire split network. The architecture shares an underlying backbone network, but task branches are deeply separated to process different tasks synchronously. This represents the depth feature map after 1x1 convolution dimensionality reduction. This represents physical prior features of vegetation indices, such as NDVI, NDRE, and LCI. These are concatenated with deep features at the channel dimension to provide the network with explicit physical prior knowledge, giving the network a clear business orientation.

[0128] To address the memory overflow risk associated with ultra-high resolution multispectral matrices, the system's internal SDK integrates a dynamic adaptive tile partitioning mechanism. Based on the acceleration chip's shape parameter limits detected at runtime, the engine performs adaptive tile partitioning on the input large-format reflectance map, using overlapping regions. These tile blocks are batch-push into the tensor core for parallel inference, with each tile edge retaining a certain proportion of overlapping pixels for subsequent pixel-level stitching using a weighted smoothing algorithm. This stitching logic follows a weighted fusion formula based on spatial distance. ,in, This represents a spatial domain weighted fusion function. It is used to calculate the final confidence weight of a pixel within the overlapping region, and this weight value changes smoothly with the distance of the pixel from the edge of the overlapping region. This represents the distance (in pixels) of a pixel from the edge of the slice. This indicates the preset width of the overlapping area (in pixels), which determines the number of overlapping pixels between blocks; Used to map distance proportions to the phase change of a cosine function.

[0129] This logic assigns a higher confidence weight to the overlapping central region, effectively eliminating inference gaps at slice boundaries and ensuring the global continuity of the final business monitoring results on a spatial scale.

[0130] As can be seen from the above embodiments, compared with the prior art, this application addresses the problems commonly found in existing low-altitude remote sensing systems, such as poor processing timeliness due to the separation of "image capture, storage, and computation," large-format inference memory overflow caused by low edge computing power and hardware synergy, and weak algorithm security protection due to the plaintext storage of inversion models and the lack of dynamic cloud management. This application includes, but is not limited to, the following beneficial effects:

[0131] Firstly, this application places the entire data processing flow from raw digital response signals to operational monitoring indicators at the airborne edge, completing radiometric calibration, reflectivity calculation, and AI inference in real time during flight. This eliminates the need for manual data copying to ground workstations for post-processing after the operation is completed, reducing the traditional hourly or even dayly processing cycle to the second level. This solves the problem of poor data timeliness caused by the separation of "photographing, storing, and calculating" in existing technologies, achieving "calculation and decision-making immediately upon flight," and meeting the needs of application scenarios with extremely high timeliness requirements, such as emergency disaster assessment and real-time crop stress diagnosis.

[0132] Secondly, this application achieves a direct physical connection between the multispectral sensing array and the AI ​​acceleration chip through a high-speed board-level bus, eliminating the bandwidth limitations and packet loss risks of traditional Ethernet transmission. At the same time, through the adaptive block division mechanism in the edge-side intelligent acceleration engine, the input data is dynamically divided according to the chip's Shape parameters. Combined with the 1×1 convolution acceleration of the tensor calculation core and INT8 / FP16 quantization inference, it effectively avoids memory overflow of high-resolution multispectral images during edge inference, improving the computing power utilization efficiency and processing stability of the airborne edge.

[0133] Thirdly, this application employs a dynamic dark current subtraction and a sixth-order halo polynomial compensation model in the radiometric calibration stage. In the reflectance calculation stage, it introduces a three-dimensional compensation mechanism combining DLS measured illumination, real-time pose calculation, and a radiative transfer model, ensuring that the real-time generated surface reflectance data possesses quantitative accuracy comparable to that of ground post-processing. In the AI ​​inversion stage, physical prior features of vegetation indices such as NDVI, NDRE, and LCI are calculated and fused with depth features extracted by 1×1 convolution in the channel dimension, forming a dual-drive mode of "physical prior + depth features." This guides the neural network to have a clear business orientation, improving both inversion accuracy and model interpretability.

[0134] Fourth, this application employs a DEK / KEK dual-key encryption mechanism to protect the inversion model. Combined with device hardware feature binding and dynamically distributed short-term leases from the cloud, model decryption and streaming deserialization loading are completed in isolated memory, enabling zero-disk execution of plaintext model data. This mechanism ensures that the model is always stored in encrypted form at the edge, only briefly decrypted and executed in authorized isolated memory, effectively preventing the core algorithm from being reverse-engineered. Simultaneously, the automatic circuit breaker mechanism suspends inference capabilities upon expiration of the short-term lease, enabling on-demand licensing and dynamic management after device sale, providing compliant technical support for digital copyright protection and commercial monetization.

[0135] Fifth, this application adopts a split network model architecture, extracts common features at the bottom layer through a unified backbone network, and splits multiple inference pipeline branches for different business indicators in the backend to perform concurrent inversion. In one inference, multi-dimensional monitoring indicators such as leaf area index, vegetation health, and nitrogen content are output simultaneously, avoiding the computational redundancy caused by deploying independent models for each indicator, significantly improving the computing power reuse rate of the airborne edge, and reducing power consumption and computational overhead.

[0136] Please see Figure 6A low-altitude intelligent sensor data processing device provided for one of the purposes of this application includes a sensing and acquisition module 1100, a physical correction module 1200, a security authentication module 1300, an inversion reasoning module 1400, and a service output module 1500. The sensing and acquisition module 1100 is configured to acquire multi-band raw images of the target area and corresponding raw digital response signals from a multispectral array camera at the airborne edge end; the physical correction module 1200 is configured to perform dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the raw digital response signals at the airborne edge end to generate radiance data; the security authentication module 1300 is configured to acquire solar down-direction irradiance data synchronously acquired by the airborne down-direction optical sensor at the airborne edge end, acquire aircraft attitude data fed back by the airborne inertial navigation unit, and calculate the solar zenith angle in real time based on the attitude data and the solar altitude angle, and obtain the sky diffuse irradiance according to a preset radiative transfer model; and use the solar down-direction irradiance, the solar zenith angle, and the sky diffuse irradiance to evaluate the radiance data. Brightness data undergoes illumination compensation and atmospheric correction to output surface reflectance data. The inversion inference module 1400 is configured to extract at least one hardware feature from the hardware device at the airborne edge, bind it, initiate a dynamic authorization request to the cloud server, receive a short-term lease from the cloud and a data encryption key encapsulated by a key encryption key, verify the validity of the short-term lease in isolated memory, and after successful verification, decapsulate and obtain the plaintext data encryption key. The data encryption key is then used to decrypt and load a pre-set encrypted inversion model. The service output module 1500 is configured to calculate the vegetation index physical prior features based on the surface reflectance data at the airborne edge, fuse the vegetation index physical prior features with a multispectral feature map processed by 1×1 convolution, and use the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators for the target area.

[0137] Based on any embodiment of this application, please refer to Figure 7 Another embodiment of this application also provides an electronic device, which can be implemented by a computer device, such as... Figure 7The diagram shows the internal structure of a computer device. The computer device includes a processor, a computer-readable storage medium, a memory, and a network interface connected via a system bus. The computer-readable storage medium stores an operating system, a database, and computer-readable instructions. The database may store control information sequences. When the computer-readable instructions are executed by the processor, the processor can implement a low-altitude intelligent sensor data processing method. The processor of the computer device provides computing and control capabilities, supporting the operation of the entire computer device. The memory of the computer device may store computer-readable instructions. When the computer-readable instructions are executed by the processor, the processor can execute the low-altitude intelligent sensor data processing method of this application. The network interface of the computer device is used for communication with a terminal. Those skilled in the art will understand that… Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0138] In this embodiment, the processor is used to execute... Figure 6 The specific functions of each module are described, and the memory stores the program code and various data required to execute these modules. The network interface is used for data transmission between the user terminal and the server. In this embodiment, the memory stores the program code and data required to execute all modules in the low-altitude intelligent sensor data processing device of this application, and the server can call the server's program code and data to execute the functions of all modules.

[0139] This application also provides a storage medium storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the steps of the low-altitude smart sensor data processing method described in any embodiment of this application.

[0140] This application also provides a computer program product, including a computer program / instructions that, when executed by one or more processors, implement the steps of the low-altitude smart sensor data processing method described in any embodiment of this application.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments of this application can be implemented by a computer program instructing related hardware. This computer program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0142] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for processing low-altitude intelligent sensor data, characterized in that, include: The airborne edge device acquires multi-band raw images of the target area and the corresponding raw digital response signals from a multispectral array camera. The airborne edge terminal performs dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal to generate radiance data. The airborne edge terminal acquires solar down-direction normal irradiance data synchronously collected by the airborne down-direction optical sensor, acquires aircraft attitude data fed back by the airborne inertial navigation unit, and calculates the solar incidence zenith angle in real time based on the attitude data and the solar altitude angle, and obtains sky diffuse irradiance according to a preset radiative transfer model; the irradiance data is then subjected to illumination compensation and atmospheric correction using the solar down-direction normal irradiance, the solar incidence zenith angle, and the sky diffuse irradiance, and the surface reflectance data is output. The airborne edge device extracts at least one hardware feature from the hardware device and binds it, initiates a dynamic authorization request to the cloud server, receives a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, verifies the validity of the short-term lease in isolated memory, and after successful verification, decapsulates and obtains the plaintext data encryption key, and uses the data encryption key to decrypt and load a preset ciphertext inversion model. The airborne edge device calculates the physical prior features of vegetation index based on the surface reflectance data, fuses the physical prior features of vegetation index with the multispectral feature map after 1×1 convolution, and uses the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators of the target area.

2. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The steps of performing dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal at the airborne edge terminal to generate radiance data include: The real-time internal temperature and current exposure time of the low-altitude intelligent sensor are obtained. The dark current background matrix parameters at the pixel level are determined based on the real-time internal temperature and the current exposure time. The dark current background matrix parameters are subtracted from the original digital response signal to obtain the dark current corrected response signal. The lens vignetting polynomial correction coefficients are obtained. The vignetting polynomial correction coefficients are sixth-order polynomials with the radial distance from the pixel coordinates to the optical principal point as the independent variable. The sixth-order polynomials are used to perform lens vignetting compensation on the response signal after dark current correction to obtain the vignetting compensated response signal. Obtain the factory absolute calibration coefficients for each band, and use the factory absolute calibration coefficients to map the vignetting compensated response signal into radiance data.

3. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The steps of performing illumination compensation and atmospheric correction on the radiance data using the solar downward normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance, and outputting surface reflectance data, include: Obtain the solar downward normal irradiance, the solar incident zenith angle, and the sky diffuse irradiance; Calculate the cosine value corresponding to the solar incident zenith angle, calculate the first product between the solar downward normal irradiance and the cosine value, and obtain the surface direct irradiance component. Calculate the first sum between the direct surface irradiance component and the diffuse sky irradiance to obtain the total incident irradiance on the surface; Calculate the first ratio between the radiance data and the total incident irradiance on the Earth's surface, and multiply the first ratio by π to construct a surface reflectance calculation model and obtain surface reflectance data.

4. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The steps include: an airborne edge device extracts at least one hardware feature from a hardware device for binding; initiates a dynamic authorization request to a cloud server; receives a short-term lease and a data encryption key encapsulated by a key encryption key from the cloud; verifies the validity of the short-term lease in isolated memory; after successful verification, decapsulates and obtains the plaintext data encryption key; and uses the data encryption key to decrypt and load a pre-set ciphertext inversion model. Extract at least one hardware feature from the hardware device and bind it as a device identity credential when the airborne edge terminal initiates a dynamic authorization request to the cloud server; Send an authorization request carrying the hardware features to the cloud server, and receive a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, wherein the short-term lease includes a valid usage period; The short-term lease and the data encryption key encapsulated by the key encryption key are stored in isolated memory, and the validity of the short-term lease is verified in the isolated memory; After successful verification, the data encryption key encapsulated by the key encryption key is decapsulated in the isolated memory using the key encryption key to obtain the plaintext data encryption key; The preset ciphertext inversion model is decrypted in the isolated memory using the data encryption key of the plaintext, and the decrypted inversion model is loaded into a streaming deserialization. The decryption and loading processes are both completed in the isolated memory to perform zero-disk loading of plaintext model data.

5. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The steps of calculating the physical prior features of vegetation index based on the surface reflectance data, fusing the physical prior features of vegetation index with the multispectral feature map processed by 1×1 convolution, and using the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain the multidimensional monitoring indicators of the target area include: The surface reflectance data is acquired, and the surface reflectance data includes reflectance information of each channel of the multispectral array camera; The physical prior features of vegetation index are calculated based on the surface reflectance data. These physical prior features of vegetation index are used to characterize the physiological and biochemical parameter status of vegetation in the target area. The multispectral feature map acquired by the multispectral array camera is obtained, and the multispectral feature map is subjected to cross-channel feature integration and nonlinear activation through 1×1 convolution to obtain the dimension-reduced depth feature map. The physical prior features of the vegetation index are fused with the dimensionality-reduced depth feature map along the channel dimension to obtain the fused features. The fusion feature input terminal side intelligent acceleration engine performs inference inversion to obtain multi-dimensional monitoring indicators of the target area.

6. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The hardware features include one or more of the following: MAC address, CPU serial number, or trusted platform module fingerprint features.

7. The low-altitude intelligent sensor data processing method according to claim 1, characterized in that, The physical prior features of the vegetation index include at least one of the normalized vegetation index, the red-edge normalized index, and the leaf nitrogen index.

8. A low-altitude intelligent sensor data processing device, characterized in that, include: The sensing and acquisition module is configured to acquire multi-band raw images of the target area and the corresponding raw digital response signals from the multispectral array camera at the airborne edge. The physical correction module is configured to perform dark current dynamic subtraction, relative radiometric correction based on lens vignetting polynomial compensation, and absolute radiometric correction based on factory absolute calibration coefficients on the original digital response signal at the airborne edge end, thereby generating radiance data. The security authentication module is configured to acquire solar down-direction irradiance data synchronously collected by the airborne down-direction optical sensor at the airborne edge end, acquire aircraft attitude data fed back by the airborne inertial navigation unit, and calculate the solar incidence zenith angle in real time based on the attitude data and the solar altitude angle, and obtain the sky diffuse irradiance according to the preset radiative transfer model; perform illumination compensation and atmospheric correction on the irradiance data using the solar down-direction normal irradiance, the solar incidence zenith angle, and the sky diffuse irradiance, and output the surface reflectance data; The inversion inference module is configured to extract at least one hardware feature from the hardware device at the airborne edge and bind it, initiate a dynamic authorization request to the cloud server, receive a short-term lease and a data encryption key encapsulated by the key encryption key issued by the cloud, verify the validity of the short-term lease in isolated memory, and after successful verification, decapsulate and obtain the plaintext data encryption key, and use the data encryption key to decrypt and load the preset ciphertext inversion model. The business output module is configured to calculate the physical prior features of vegetation index based on the surface reflectance data at the airborne edge, fuse the physical prior features of vegetation index with the multispectral feature map after 1×1 convolution, and use the fused features to perform inference inversion in the edge-side intelligent acceleration engine to obtain multi-dimensional monitoring indicators of the target area.

9. An electronic device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images