Image real-time radiometric correction method of pushbroom hyperspectral camera

Abstract

The application discloses a kind of push-broom hyperspectral camera's image real-time radiation correction method, it is related to image processing method technical field.The method includes the following steps: based on the real-time radiation correction system built and defines acquisition parameter, using hyperspectral camera collects the hyperspectral data of four standard reflectance plates;High spectral image and spectrometer information are aligned by timestamp, and the image preprocessing of push-broom hyperspectral camera is completed by creating ROI area method;The information collected by spectrometer is corrected by the correction of cosine corrector azimuth angle, and a hyperspectral correction model is established to predict reflectivity.The method is accurate in data acquisition and can achieve real-time radiation correction of hyperspectral image, and the correction accuracy is high.

Description

Technical Field

[0001] This invention relates to the field of image processing methods, and in particular to a real-time radiometric correction method for images from a pushbroom hyperspectral camera. Background Technology

[0002] Current UAV hyperspectral imaging systems suffer from two major drawbacks: poor platform dynamic adaptability and inaccurate radiometric references. Traditional radiometric calibration relies on satellite remote sensing data, which has a revisit period of over 16 days and is severely affected by atmospheric interference, resulting in band calibration errors exceeding 12%. While ground-based fixed spectrometers possess high spectral resolution, they cannot adapt to the moving UAV platform, leading to the standard reflectivity plate being obscured by the UAV's shadow during pushbroom imaging (measured shadow coverage exceeds 40%). In existing technologies, even with network time protocol synchronization between the spectrometer and camera, there is still an error of more than 3 seconds, and a real-time solar angle feedback mechanism is lacking—for every 10-degree change in solar altitude angle, the radiative flux fluctuates by up to 35%. Furthermore, CMOS sensors exhibit weak responses in certain bands, and the dark current temperature drift effect during motion (dark current increases by 5-10% for every 1-degree Celsius increase in temperature) further degrades radiometric consistency.

[0003] Pushbroom hyperspectral image preprocessing suffers from core problems of spatiotemporal misalignment accumulation and insufficient near-infrared signal-to-noise ratio. Existing methods estimate spatial coordinates by integrating the UAV's GPS velocity, but velocity fluctuations lead to positioning errors exceeding 0.8 pixels (equivalent to 4 cm ground feature offset at a flight altitude of 100 meters). Regarding noise suppression, the near-infrared band suffers from a signal-to-noise ratio degradation to 38 dB (over 60 dB in the visible light band) due to silicon-based CMOS quantum efficiency being below 30%. Traditional static region selection methods exhibit single-frame data fluctuations of up to ±18% in scenarios with fluctuating illumination (irradiance variations exceeding 15 W / m²), and the false negative rate exceeds 15% when using a double standard deviation criterion for outlier filtering, resulting in a root mean square error exceeding 8% for radiometric correction in low-reflectivity regions.

[0004] Existing radiometric correction models suffer from fundamental flaws, including uncompensated geometric errors and incomplete calibration coverage. The azimuth deviation of the spectrometer's cosine receiver is not corrected; when the deviation from the solar azimuth exceeds 30 degrees, the downlink spectral measurements deviate from the true value by 18%. Current angle compensation schemes only consider the solar altitude angle, ignoring dynamic changes in azimuth (a 15-degree change per hour results in a residual error exceeding 7%). The band matching stage uses nearest-neighbor interpolation, leading to reconstruction errors exceeding 12% in the 400nm and 1000nm edge bands. The calibration model relies on a single 30% reflectivity standard plate; in low-reflectivity scenarios (7%), due to the detector's nonlinear response (coefficient of determination below 0.92), the reflectivity inversion value is systematically underestimated (only 11.3% for a 15% target). Summary of the Invention

[0005] The technical problem to be solved by the present invention is how to provide a real-time radiometric correction method for a pushbroom hyperspectral camera that can accurately acquire data, achieve real-time radiometric correction of hyperspectral images, and has high correction accuracy.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a real-time radiometric correction method for images from a pushbroom hyperspectral camera, comprising the following steps:

[0007] S1: Based on the established real-time radiometric correction system and defined acquisition parameters, hyperspectral data from four standard reflectivity plates are collected using a hyperspectral camera;

[0008] S2: Align hyperspectral images with spectrometer information using timestamps and create ROI regions to complete image preprocessing for pushbroom hyperspectral cameras;

[0009] S3: By correcting the azimuth angle of the cosine corrector and adjusting the information collected by the spectrometer, a hyperspectral correction model is established to predict reflectance.

[0010] A further technical solution is as follows: the real-time radiation correction system includes a drone platform, on the upper side of which a three-axis gimbal and a computer processing module are installed. A spectrometer is installed on the three-axis gimbal, which is used to collect downlink solar spectral irradiance in real time and store the data in the computer processing module in real time; a hyperspectral camera is connected to the lower part of the drone platform through a camera gimbal.

[0011] A further technical solution involves: the hyperspectral camera employing a pushbroom imaging method, with each frame corresponding to a scan line on the ground; the computer processing module aligns with the timestamp of the spectrometer via a network, and the hyperspectral camera time is UTC time; during the data acquisition phase, experiments are conducted throughout the day to adapt to changes in light intensity, and the real-time radiometric correction system simultaneously performs two tasks: first, the hyperspectral camera adjusts its exposure time according to lighting conditions to photograph four standard reflectivity plates; second, the spectrometer records the solar spectral irradiance with an independently set integration time, while simultaneously recording the solar altitude angle, the UAV's heading angle, and the solar azimuth angle and their corresponding times; and dark current data from the hyperspectral camera and spectrometer are collected before and after each flight to eliminate sensor noise.

[0012] The beneficial effects of adopting the above technical solution are as follows: The method described in this application constructs a real-time radiometric correction system for UAV-spectrometer in hardware, achieving second-level synchronization using satellite time, and dynamically eliminating shadows with a four-reflectivity standard plate; the preprocessing stage uses timestamp-driven geographic coordinates for direct mapping to eliminate accumulated errors, and combined with dynamic regional statistical filtering, improves the near-infrared signal-to-noise ratio to 75.8 dB; the correction model innovatively integrates azimuth cosine compensation, cubic spline band matching, and four-reference joint calibration, ultimately achieving the inversion of reflectivity within irradiance variations in 47 sets of all-weather experiments. Data acquisition is accurate and enables real-time radiometric correction of hyperspectral images with high correction precision. Attached Figure Description

[0013] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0014] Figure 1 This is the main flowchart of the method described in the embodiments of the present invention;

[0015] Figure 2 This is a technical roadmap of the method described in the embodiments of the present invention;

[0016] Figure 3 This is a detailed flowchart of the method described in the embodiments of the present invention;

[0017] Figure 4 This is a schematic diagram of the structure of the real-time radiation correction system in an embodiment of the present invention;

[0018] Figure 5 This is a schematic diagram of the data acquisition method in the embodiment of the present invention;

[0019] Figure 6 This is a dot plot of hyperspectral DN values ​​in the method described in the embodiments of the present invention;

[0020] Figure 7 This is a dot plot of the reflectance of a standard emissivity plate in different wavelength bands in the method described in this embodiment of the invention;

[0021] Figure 8 This is a graph showing the reflectivity of a standard emissivity plate at different times in the method described in this embodiment of the invention.

[0022] Figure 9a This is a diagram showing the correction results of the traditional empirical line method in the implementation of this invention;

[0023] Figure 9b This is a diagram showing the correction results of the real-time radiation correction method implemented in this invention;

[0024] The components include: 1. Unmanned aerial vehicle platform; 2. Three-axis gimbal; 3. Computer processing module; 4. Spectrometer; 5. Camera gimbal; 6. Hyperspectral camera. Detailed Implementation

[0025] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0026] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0027] like Figure 1 As shown in the figure, an embodiment of the present invention discloses a real-time radiometric correction method for images from a pushbroom hyperspectral camera, comprising the following steps:

[0028] S1: Based on the established real-time radiometric correction system and defined acquisition parameters, hyperspectral data from four standard reflectivity plates are collected using a hyperspectral camera;

[0029] S2: Align hyperspectral images with spectrometer information using timestamps and create ROI regions to complete image preprocessing for pushbroom hyperspectral cameras;

[0030] S3: By correcting the azimuth angle of the cosine corrector and adjusting the information collected by the spectrometer, a hyperspectral correction model is established to predict reflectance.

[0031] like Figures 2-3 As shown below, the above method will be explained in detail with reference to specific content:

[0032] like Figure 4 As shown, the real-time radiation correction system includes a drone platform 1. A three-axis gimbal 2 and a computer processing module 3 are mounted on the upper side of the drone platform 1. A spectrometer 4 is installed on the three-axis gimbal 2. The spectrometer 4 is used to collect downlink solar spectral irradiance in real time and store the data in the computer processing module 3 in real time. A hyperspectral camera 6 is connected to the lower part of the drone platform 1 through a camera gimbal 5.

[0033] The real-time radiometric correction system uses a UAV as its core platform. A three-axis gimbal 2 is mounted on top of the UAV, and a high-precision spectrometer 4 (spectral resolution ≤5nm) with a wavelength range of 400-1000nm is fixed on the gimbal 2. This spectrometer is used to collect downlink solar spectral irradiance in real time, ensuring the quality of the collected data is not affected by strong fluctuations. The data is then stored in real time in a computer processing module, providing an accurate reference for subsequent radiometric correction and ensuring the reliability of data storage. A hyperspectral camera 6 (spatial resolution ≤5cm, spectral channels ≥400) in the same wavelength band is simultaneously mounted on the bottom of the UAV. The hyperspectral camera 6 uses a pushbroom imaging method, with each frame corresponding to a scan line on the ground. This imaging method is particularly suitable for mobile data acquisition on a UAV platform.

[0034] To ensure data synchronization, the computer processing module was networked to align with the spectrometer's timestamp, while the camera time was set to UTC. This resulted in a time error of ≤1 second between the spectrometer and camera acquisition. This configuration enabled synchronous acquisition of spectral and image data, laying the foundation for subsequent data matching and processing. Furthermore, four uniformly distributed reflectance standard plates (7%, 15%, 30%, and 50%, with dimensions ≥1m) were used in the experiment. 2 As a benchmark for radiometric calibration, it provides reference targets with different reflectivity levels to verify the accuracy of radiometric correction and ensure the radiometric consistency of hyperspectral camera imaging.

[0035] During the data acquisition phase, experiments were conducted throughout the day to adapt to changes in light intensity. Therefore, the real-time radiometric correction system simultaneously performed two tasks: first, the hyperspectral camera adjusted its exposure time according to lighting conditions to photograph four standard reflectivity plates; second, the spectrometer recorded solar spectral irradiance using an independently set integration time (adjusted according to changes in light intensity), while also recording the solar altitude angle, UAV heading angle, and solar azimuth angle, along with their corresponding times. Dark current data from the hyperspectral camera and spectrometer were collected before and after each flight to eliminate sensor noise.

[0036] like Figure 5 For data acquisition, a pushbroom hyperspectral camera and an overhead spectrometer are integrated into a UAV platform, with a flight altitude of 3 meters to ensure that the imaging of the ground standard reflectivity panel covers the entire field of view. Radiation data at 47 time points within a single day (covering different solar altitude angles and atmospheric conditions) are continuously collected. The shadow interference from the reflectivity panel is eliminated by the movement of the UAV to obtain baseline data without shadow effects. The original DN value of the hyperspectral camera, the radiance data measured by the spectrometer, the solar incidence angle, and atmospheric parameters are recorded simultaneously at each time point.

[0037] Image preprocessing methods for pushbroom hyperspectral cameras:

[0038] This application addresses the raw data acquired by hyperspectral cameras and spectrometers. For hyperspectral cameras, in hyperspectral pushbroom imaging, each frame of data contains a precise timestamp (generated by high-precision...). GPS (Synchronous recording), and the final stitching process strictly relies on spatiotemporal information (timestamp + location data). Specifically, a single frame of data is represented as... (Number of spatial rows × number of columns × number of bands), and associated with the corresponding timestamps. t and GPS coordinate( , The purpose of the timestamp is to directly associate the frame's acquisition time with the platform's location, eliminating the need to estimate spatial coordinates through velocity. The core logic of stitching is to map the timestamp to geospatial coordinates, first by... GPS The starting geographic coordinates provided for each frame ( Determine their spatial arrangement and strictly align them according to timestamp order, that is... ,in( )= The concatenated time dimension directly inherits from the original timestamp sequence. Each space row H=k Corresponding to a unique timestamp Therefore, it can be accessed through Query the time information for any pixel row. Finally, stack the corrected frames in chronological order to form a stitched image. Output timestamp mapping table This process ensures strict alignment of the hyperspectral data in the spatiotemporal dimensions while preserving precise temporal information for each frame, providing a reliable spatiotemporal reference for subsequent analysis.

[0039] During the drone's flight, the hyperspectral camera and the overhead spectrometer achieve second-level synchronous data acquisition via a hardware trigger signal. Their time synchronization relationship can be expressed as (synchronization error). 1s):

[0040] ;

[0041] in and These represent the timestamps of the k-th hyperspectral image and the m-th spectrometer sampling, respectively.

[0042] To address the radiometric measurement bias in the near-infrared band caused by the low signal-to-noise ratio (SNR) of the CMOS sensor in pushbroom hyperspectral cameras, this application proposes a benchmark data optimization method based on dynamic ROI selection. This method selects hyperspectral image frames within a continuous time window (≤3 seconds) where the measured radiance fluctuation range of the spectrometer is ≤3% (corresponding to essentially constant irradiance), ensuring temporal consistency of the data (corresponding to a solar irradiance change ΔE ≤ 10 W / m²).2 Select a set of hyperspectral image frames within a continuous time window T (T≤3 seconds). :

[0043] ;

[0044] in The radiance data are measured by a spectrometer at specific time points. The k-th frame hyperspectral image data cube (dimension: 1×W×C).

[0045] Statistical filtering was performed on the DN values ​​of all pixels (approximately 6000) within the ROI: outliers caused by sensor noise (such as outliers outside ±3σ) were removed, and the band average DN value of the remaining pixels was calculated as the standard response value for that band under the current lighting conditions. For each frame... A rectangular region of interest (ROI) of 100×60 pixels is selected in the center of the image, and its DN value matrix is ​​as follows. Outliers are removed using the 3σ criterion:

[0046] ;

[0047] in Let be the ROI pixel matrix (dimensions: H×W×C) of the k-th frame hyperspectral image. Let be the pixel DN value of band c at spatial location (i,j) in frame k. The mean value of all pixels in band c. denoted as the standard deviation of pixel DN values ​​for band c.

[0048] By averaging pixels across a large area of ​​the region of interest (ROI), statistical principles are applied to reduce single-pixel random noise to its original value. (N is the number of effective pixels), significantly improving the signal-to-noise ratio in the near-infrared band. Experiments show that when N=6000, the SNR in the near-infrared band (900-1000nm) can be improved by approximately 77 times (theoretical value: 10log). 10 (6000)≈37.8dB), effectively overcoming the sensitivity deficiency of CMOS sensors in the long-wavelength region. The above average DN value is bound to the radiance data measured by the synchronous spectrometer and environmental parameters (solar angle, atmospheric optical thickness) to form a training sample set for the dynamic radiometric correction model, ensuring the spatial representativeness and noise robustness of the model input data.

[0049] Dark current correction and exposure normalization were performed to eliminate the influence of sensor noise and different exposure conditions on radiometric calibration. The specific implementation process is as follows: Under the same ambient temperature, dark current correction was performed by acquiring full-dark images from the hyperspectral camera and dark background signals from the spectrometer: For the hyperspectral camera, at least 10 full-dark images were acquired with the lens cap closed, and the average dark current value for each band was calculated. and from the original image Deducting from = For the spectrometer, the dark background signal is acquired by closing the incident light path within the same integration time. and from the original radiance Deducting from = Simultaneously, the integration time of the spectrometer is recorded. and the exposure time of the hyperspectral camera and gain value Exposure normalization is then performed on the corrected data:

[0050] Hyperspectral images are normalized using the following formula:

[0051] ;

[0052] Spectrometer data are normalized using the following formula:

[0053] ;

[0054] This preprocessing method aims to eliminate the influence of different exposure parameters on the data. Finally, the linear correlation of the normalized data under different illumination conditions is examined to verify the correction effect and ensure that the data can be used for subsequent radiometric calibration modeling. This preprocessing method effectively solves the problem of inconsistent radiometric response caused by dark current noise in CMOS sensors and variations in exposure parameters, providing a reliable data foundation for high-precision radiometric correction.

[0055] Real-time radiometric correction method for images from pushbroom hyperspectral cameras:

[0056] First, the downlink solar spectrum acquired by the spectrometer is corrected. Because the angle between the spectrometer's azimuth and the sun's azimuth changes, the downlink solar spectrum values ​​acquired by the spectrometer will exhibit a cosine variation. According to the formula... Calculate the relative azimuth angle of the spectrometer, where, Indicates the azimuth angle of the sun. This indicates the azimuth angle of the spectrometer's orientation. Based on the spectrometer's relative azimuth angle and solar altitude angle, according to the formula... ( The cosine error of the solar spectrum after dark current and exposure normalization is corrected, whereby... This represents the solar spectrum after cosine error correction. Indicates the solar altitude angle. ( ) represents the cosine function obtained by fitting the solar altitude angle and the relative azimuth angle of the spectrometer.

[0057] Another method for band matching between hyperspectral camera and spectrometer data is discussed, specifically addressing the band alignment problem between hyperspectral image data after dark current correction and exposure normalization and downlink solar spectral data. The hyperspectral camera has n discrete bands, whose wavelength vector is denoted as... =[ , ,......, The corresponding digital quantization value (DN value) vector is: The spectrometer has m discrete bands (m>n), and its wavelength vector is denoted as... =[ , ,......, The corresponding DN value vector is To achieve band matching between the two, this method first reconstructs the spectrometer data in a continuous wavelength space using an interpolation algorithm, establishing a continuous spectral function. :

[0058] ;

[0059] Specifically, linear interpolation or cubic spline interpolation methods are used, among which, The linear interpolation formula is:

[0060] ;

[0061] Cubic spline interpolation constructs a smooth interpolation function using piecewise cubic polynomials. The cubic spline interpolation formula is:

[0062] ;

[0063] In the MATLAB implementation, the `interp1` function is called with the `linear` and `spline` options to perform the two interpolation methods described above. The output is the spectrometer DN value corresponding to the center wavelength of each hyperspectral band, forming an n×2 matching matrix. , ( This method ensures the accuracy of subsequent radiometric correction or reflectivity calculation through rigorous band alignment processing, making it particularly suitable for collaborative observation scenarios involving UAV-borne hyperspectral imaging systems and ground-based spectrometers.

[0064] like Figure 6 As shown in the figure, experiments have verified a significant linear relationship between the detector's response value (DN value) and the reference spectrometer's measurement value (DN value) over a wide spectral range. This excellent quantum response linearity ensures the reliability of the final hyperspectral image data, providing a crucial guarantee for accurate quantitative inversion of ground cover information.

[0065] For each band The response slope was obtained by performing a least-squares linear fit between the hyperspectral DN values ​​(x-axis) corresponding to the four reflectivity plates and the spectrometer DN values ​​(y-axis). (n=1, 2, 3, 4 correspond to different reflectivity plates), its mathematical expression is:

[0066]

[0067] Where M is the number of sampling points. For each band The final spectral response coefficient is obtained by arithmetically averaging the correction factors of the four reflectivity plates. :

[0068]

[0069] Will Construct a diagonalized correction matrix for the original hyperspectral data. Perform band-by-band correction: = The final formula for calculating reflectance is:

[0070]

[0071] in, Figure 7 This is a dot plot of the reflectance of the standard emissivity plates in different bands in the method described in this embodiment of the invention. The dashed lines represent the actual reflectance values ​​of the four standard reflectivity plates. Figure 8 This is a reflectance curve of the standard emissivity plates at different times in the method described in this embodiment of the invention. The dashed line represents the actual reflectance values ​​of the four standard reflectivity plates. This method effectively compensates for the differences in detector quantum efficiency through joint calibration of multiple reflectivity benchmarks and band-level response modeling, achieving real-time radiometric correction of hyperspectral images. The results are as follows: Figures 9a-9b As shown, Figure 9a The correction results are from the traditional empirical line method. Figure 9b This is the correction result of the real-time radiation correction method implemented in this invention.

[0072] The real-time radiometric correction system provided in this application achieves accurate data acquisition. It uses a drone to fly and photograph a standard reflectivity plate, which basically eliminates the interference of ground shadows on the measurement. It uses a high-precision clock to synchronize the camera and spectrometer, which solves the problem of data misalignment caused by inconsistent time in existing equipment. Furthermore, it uses experimental data from the entire day for verification experiments, achieving real-time radiometric correction of hyperspectral images for the first time.

[0073] The image preprocessing method for pushbroom hyperspectral cameras proposed in this application not only stitches together the images using GPS location tags in each frame of the hyperspectral image, but also proposes a method for aligning time based on satellite timestamps. The method proposed in this application for extracting the DN values ​​of the original hyperspectral camera using ROI regions effectively improves the signal-to-noise ratio of the hyperspectral images. This preprocessing method effectively solves the problem of inconsistent radiometric response caused by dark current noise in CMOS sensors and variations in exposure parameters, providing a reliable data foundation for high-precision radiometric correction.

[0074] This application proposes a real-time radiometric correction method for pushbroom hyperspectral cameras. By using four standard reflectivity plates for joint calibration, it effectively reduces spectral response errors. Furthermore, a cosine correction method is proposed to address measurement distortion caused by azimuth changes when the equipment is in different flight zones. Ultimately, it enables joint calibration of multiple reflectivity references and band-level response modeling, effectively compensating for differences in detector quantum efficiency and achieving real-time radiometric correction of hyperspectral images.

Claims

1. A real-time radiometric correction method for images from a pushbroom hyperspectral camera, characterized in that... Includes the following steps: S1: Based on the established real-time radiometric correction system and defined acquisition parameters, hyperspectral data from four standard reflectivity plates are collected using a hyperspectral camera; S2: Align hyperspectral images with spectrometer information using timestamps and create ROI regions to complete image preprocessing for pushbroom hyperspectral cameras; S3: By correcting the azimuth angle of the cosine corrector and adjusting the information collected by the spectrometer, a hyperspectral correction model is established to predict reflectance; In step S2, the radiation measurement deviation is corrected using a reference data optimization method, including the following steps: To ensure temporal consistency of the data, hyperspectral image frames within a continuous time window of ≤3% radiance fluctuation range measured by the spectrometer were selected. A set of hyperspectral image frames within a continuous time window T ≤ 3 seconds was chosen. : ； in The radiance data are measured by a spectrometer at specific time points. The k-th frame hyperspectral image data cube has dimensions of 1×W×C; Statistical filtering is performed on the DN values ​​of all pixels within the ROI: outliers caused by sensor noise are removed, and the average DN value of the remaining pixels is calculated as the standard response value of that band under the current lighting conditions. For each frame A rectangular region of interest (ROI) of 100×60 pixels is selected in the center of the image, and its DN value matrix is ​​as follows. Outliers are removed using the 3σ criterion: ； in Let be the ROI pixel matrix of the k-th frame of the hyperspectral image. Let be the pixel DN value of band c at spatial location (i,j) in frame k. Let be the mean value of all pixels in band c. denoted as the standard deviation of pixel DN values ​​for band c.

2. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that: The real-time radiation correction system includes a drone platform (1), on the upper side of which is a three-axis gimbal (2) and a computer processing module (3). A spectrometer (4) is installed on the three-axis gimbal (2). The spectrometer (4) is used to collect downlink solar spectral irradiance in real time and store the data in the computer processing module (3) in real time. A hyperspectral camera (6) is connected to the lower part of the drone platform (1) through a camera gimbal (5).

3. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 2, characterized in that: The hyperspectral camera (6) adopts a push-broom imaging method, with each frame of the image corresponding to a scan line on the ground; the computer processing module (3) is connected to the network to align the timestamp of the spectrometer (4), and the time of the hyperspectral camera (6) is UTC time; During the data acquisition phase, experiments were conducted throughout the day to adapt to changes in light intensity. The real-time radiation correction system simultaneously performed two tasks: First, the hyperspectral camera (6) adjusted the exposure time according to the lighting conditions to photograph four standard reflectivity plates; second, the spectrometer (4) recorded the solar spectral irradiance with an independently set integration time, and simultaneously recorded the solar altitude angle, the UAV heading angle and the solar azimuth angle and their corresponding times. Dark current data of the hyperspectral camera and spectrometer were collected before and after each flight to eliminate sensor noise.

4. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 2, characterized in that: The flight altitude of the UAV platform (1) in the real-time radiation correction system is set to 3 meters to ensure that the imaging of the ground standard reflectivity plate covers the entire field of view; radiation data at a preset number of time points are continuously collected within a single day, and the shadow interference of the reflectivity plate is eliminated by the movement of the UAV to obtain the reference data without shadow influence. The system synchronously records the original DN value of the hyperspectral camera, the radiance data measured by the spectrometer, the solar incidence angle, and atmospheric parameters at each time point.

5. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that, In S2: The single-frame data acquired by the hyperspectral camera (6) is represented as follows: and associate with the corresponding timestamp t and GPS coordinate( , The timestamp is used to directly associate the frame acquisition time with the location of the UAV platform (1); Mapping timestamps to geospatial coordinates begins first through... GPS The starting geographic coordinates provided for each frame ( Determine their spatial arrangement and align them according to timestamp order, i.e. ,in( )= The concatenated time dimension directly inherits from the original timestamp sequence. Each space row H=k Corresponding to a unique timestamp Therefore, it can be accessed through Query the time information of any pixel row; finally, stack the corrected frames in chronological order to form a stitched image. Output timestamp mapping table ; During the flight of the UAV platform (1), the hyperspectral camera (6) and the spectrometer (4) achieve second-level synchronous acquisition through hardware trigger signals. Their time synchronization relationship is expressed as follows: ； in and _spec(m) represents the timestamp of the k-th hyperspectral image and the m-th spectrometer sampling, respectively, and the synchronization error. 1s.

6. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that, In step S2, dark current correction and exposure normalization are performed to eliminate the influence of sensor noise and different exposure conditions on radiometric calibration. Specifically, this includes the following steps: Under the same ambient temperature, dark current correction was performed by acquiring full-dark images from the hyperspectral camera (6) and dark background signals from the spectrometer (4). For the hyperspectral camera (6), acquire no fewer than 10 full-dark images with the lens cap closed, and calculate the average dark current for each band. and from the original image Deducting from = ; For the spectrometer (4), the dark background signal is acquired by closing the incident light path during the same integration time. and from the original radiance Deducting from = ; Simultaneously, the integration time of the spectrometer (4) is recorded. and the exposure time of the hyperspectral camera and gain value Exposure normalization is performed on the corrected data: Hyperspectral images are processed using the formula: Perform normalization; Spectrometer data is obtained through the following formula: Perform normalization; To eliminate the influence of different exposure parameters on the data.

7. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that, In S3: First, the downward solar spectrum collected by the spectrometer (4) is corrected according to the formula. Calculate the relative azimuth angle of the spectrometer, where, Indicates the azimuth angle of the sun. Indicates the azimuth angle of the spectrometer's orientation; Based on the relative azimuth angle and solar altitude angle of the spectrometer (4), according to the formula ( The cosine error of the solar spectrum after dark current and exposure normalization is corrected. in This represents the solar spectrum after cosine error correction. Indicates the solar altitude angle. ( ) represents the cosine function obtained by fitting the solar altitude angle and the relative azimuth angle of the spectrometer.

8. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that, In step S3, the accuracy of radiation correction or reflectivity calculation is improved by using a band matching method, including the following steps: The hyperspectral camera (6) has n discrete bands, and its wavelength vector is denoted as . =[ , ,......, The corresponding digital quantization value vector is... ; The spectrometer (4) has m discrete bands, where m>n, and its wavelength vector is denoted as... =[ , ,......, The corresponding DN value vector is ; To achieve band matching between the two, the spectrometer data is first reconstructed in a continuous wavelength space using an interpolation algorithm to establish a continuous spectral function. : ； Linear interpolation or cubic spline interpolation methods are used, where, The linear interpolation formula is: ； Cubic spline interpolation constructs a smooth interpolation function using piecewise cubic polynomials. The cubic spline interpolation formula is: ； The output is the spectrometer DN value corresponding to the center wavelength of each band in the hyperspectral spectrum, forming an n×2 matching matrix. , ( )).

9. The real-time radiometric correction method for images from a pushbroom hyperspectral camera as described in claim 1, characterized in that, In S3: For each band The response slope was obtained by performing least-squares linear fitting between the hyperspectral DN values ​​corresponding to the four reflectivity plates and the DN values ​​of the spectrometer. For n=1, 2, 3, 4, corresponding to different reflectivity plates, the mathematical expression is: ； Where M is the number of sampling points, for each band The final spectral response coefficient is obtained by arithmetically averaging the correction factors of the four reflectivity plates. : ； Will Construct a diagonalized correction matrix for the original hyperspectral data. Perform band-by-band correction: = ； The final formula for calculating reflectance is: 。

Images