A method and system for fine surface albedo acquisition for hyperspectral unmanned aerial vehicle remote sensing

Abstract

The application discloses a fine ground albedo acquisition method and system for hyper-spectral unmanned aerial vehicle remote sensing, wherein a visible imaging lens fixed on an unmanned aerial vehicle load is used to shoot a ground feature image corresponding to each image element, and each ground feature image is preprocessed, adaptively cropped and semantically segmented; according to the semantic segmentation result and in combination with the ground albedo of each ground type, weighted calculation of the ground feature image is carried out to obtain the fine ground albedo of the ground feature image. In this way, the ground albedo corresponding to each image element observed by the unmanned aerial vehicle can be quickly and accurately acquired.

Description

Technical Field

[0001] This invention belongs to the field of environmental remote sensing technology, specifically relating to a method and system for obtaining fine surface albedo for hyperspectral unmanned aerial vehicle remote sensing. Background Technology

[0002] Satellite remote sensing has a large spatiotemporal resolution, while vehicle-mounted monitoring instruments cannot enter the factory area for monitoring. UAV spectral remote sensing can overcome technical problems such as low spatiotemporal resolution, monitoring blind spots, and difficulty in identifying pollution sources.

[0003] A significant technical challenge in using UAV spectral remote sensing for atmospheric monitoring is the rapid changes in land surface conditions. Changes in land surface type directly affect surface reflectance, a crucial parameter in remote sensing data processing. Different land surface types (such as water bodies, forests, grasslands, and urban buildings) possess different reflectance characteristics. If the land surface type changes rapidly during UAV flight, the resulting change in reflectance will significantly alter the spectral data received by the remote sensing sensor.

[0004] Land surface albedo is a key parameter in calculating the atmospheric quality factor (AMF) in radiative transfer models, and accurate land surface albedo is particularly important for processing spectral data. If land surface albedo is not properly corrected, the radiation calculations in the radiative transfer model will be affected, leading to errors in key parameters such as the radiation output and the atmospheric quality factor (AMF). These errors directly affect the accuracy of environmental monitoring data such as pollutant concentrations and gas distribution. Summary of the Invention

[0005] In view of the above, the purpose of this invention is to provide a method and system for obtaining fine surface albedo for hyperspectral unmanned aerial vehicle (UAV) remote sensing, so as to quickly and accurately obtain the surface albedo corresponding to each pixel observed by the UAV.

[0006] To achieve the above-mentioned objectives, an embodiment provides a method for obtaining fine surface albedo using hyperspectral unmanned aerial vehicle (UAV) remote sensing, comprising the following steps:

[0007] The surface feature image corresponding to each pixel is captured by a visible imaging lens fixed on the drone payload, and each surface feature image is preprocessed, adaptively cropped, and semantically segmented.

[0008] Based on the semantic segmentation results and combined with the surface albedo of each surface type, a weighted calculation of the surface feature image is performed to obtain the fine surface albedo of the surface feature image.

[0009] Preferably, the visible imaging lens is fixed on the same axis as the telescope module on the UAV payload using a coaxial fixing method, so that the center of the photo taken by the visible imaging lens coincides with the center of the telescope module, while ensuring that the telescope elevation angle and the visible imaging lens rotate synchronously, ensuring that each surface feature image captured by the visible imaging lens corresponds one-to-one with the pixel of each elevation angle of the telescope.

[0010] Preferably, each surface feature image undergoes preprocessing, including:

[0011] Geometric correction of surface feature images is performed based on random sampling consistency algorithm and automatic matching algorithm of common source points to eliminate the influence of geometric deformation on surface albedo in surface feature images.

[0012] Preferably, adaptive cropping is performed on each surface feature image, including:

[0013] The field of view θ of a visible imaging lens camera and the field of view θ of the telescope module telescope The ratio is used to calculate the image width W of the visible imaging lens. camera and the image width W of the telescope telescope The ratio, that is:

[0014]

[0015] Adaptive cropping of surface feature images is performed based on aspect ratio to ensure that each image captured by the visible imaging lens corresponds perfectly with each pixel of the telescope.

[0016] Preferably, the field of view θ of the telescope module telescope It is generally a fixed value, the magnitude of which depends on the optical fiber and lens used, and the field of view θ of the visible imaging lens. camera Calculated in the following way:

[0017]

[0018] Where h is the altitude of the UAV above the ground, α is the tilt angle between the telescope and the horizontal direction, and W pixel and H pixel W represents the pixel width and height of the captured surface feature image, respectively. image and H image θ represents the width and height of the camera sensor, respectively. horizonrtal and θ vertical θ represents the field of view of the visible imaging lens in the horizontal and vertical directions. Since the change in field of view caused by the change in the focal length of the visible imaging lens is linear, θ is chosen as the focal length. horizonrtal The field of view of a visible imaging lens, denoted as θ. camera .

[0019] Preferably, semantic segmentation is performed on each surface feature image, including:

[0020] A deep convolutional neural network model is used to segment the adaptively cropped surface feature image. The model output is a segmentation mask M. i (x,y), where M i (x,y) represents the land surface type i at pixel (x,y).

[0021] Preferably, the deep convolutional neural network model adopts the SegNet model.

[0022] Preferably, a weighted calculation of the surface feature image is performed based on the semantic segmentation results and the surface albedo of each surface type, including:

[0023] For each cell, use A i Represents the surface albedo of surface type i, based on surface albedo A i The weighted calculation of surface feature images is represented as follows:

[0024]

[0025] Among them, S pixel S represents the actual ground area of ​​each pixel, n represents the total number of land surface types, Albedo represents the weighted fine albedo, and S represents the weighted fine albedo. i The area of ​​land surface type i is represented by the following formula:

[0026]

[0027] Where W and H represent the width and height of the surface feature image, respectively.

[0028] Preferably, the method further includes: obtaining fine surface albedo at different center wavelengths through a conversion coefficient, including:

[0029]

[0030] Albedo'=f×Albedo

[0031] Where f is the conversion coefficient, k is an exponential variation coefficient related to land cover type or material, which determines the rate of change of albedo with wavelength, reflecting the reflectivity of surface material at different wavelengths; λ1 and λ2 are two center wavelengths, Albedo' represents the fine surface albedo at center wavelength λ2, and Albedo represents the fine surface albedo at center wavelength λ1.

[0032] To achieve the aforementioned objectives, the embodiments also provide a refined surface albedo acquisition system for hyperspectral unmanned aerial vehicle (UAV) remote sensing, comprising a UAV payload with a visible imaging lens and a telescope module, and a computing module.

[0033] The UAV payload captures images of the ground features corresponding to each pixel using its visible imaging lens.

[0034] The calculation module preprocesses, adaptively crops, and semantically segments each captured surface feature image. Based on the semantic segmentation results and combined with the surface albedo of each surface type, it performs a weighted calculation of the surface feature images to obtain the fine surface albedo of the surface feature images.

[0035] Compared with the prior art, the beneficial effects of the present invention include at least the following:

[0036] This invention employs a combination of hardware and software. For the surface feature image corresponding to each pixel acquired by the hardware visible imaging lens, a specific software algorithm is used to preprocess, adaptively crop, and semantically segment the surface feature image. Based on the semantic segmentation results and combined with the surface albedo of different surface types, a weighted calculation of the surface feature image is performed to obtain a fine surface albedo. This allows for a precise characterization of the surface albedo of each observed pixel within the entire target area, resulting in more accurate and reliable observation results compared to the traditional single surface albedo method. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a method for obtaining fine surface albedo for hyperspectral UAV remote sensing provided in the embodiments;

[0039] Figure 2 This is a schematic diagram of surface feature images captured by a drone observation and visible imaging lens provided in the embodiment;

[0040] Figure 3 This is a schematic diagram of adaptive clipping provided in the embodiment;

[0041] Figure 4 This invention provides a method for obtaining the surface albedo result of a single surface feature image for obtaining the corresponding pixel. Detailed Implementation

[0042] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not limit the scope of protection of this invention.

[0043] As shown in the figure, an embodiment provides a method for obtaining fine surface albedo for hyperspectral UAV remote sensing, including the following steps:

[0044] S1 captures a surface feature image corresponding to each pixel using a visible imaging lens fixed on the UAV payload, and performs preprocessing, adaptive cropping, and semantic segmentation on each surface feature image.

[0045] In this embodiment, the visible imaging lens and the telescope module on the UAV payload are fixed on the same axis using a coaxial fixing method. This ensures that the center of the image captured by the visible imaging lens coincides with the center of the telescope module, while also ensuring that the telescope's elevation angle and the visible imaging lens rotate synchronously. This guarantees that each surface feature image captured by the visible imaging lens corresponds one-to-one with a pixel at each elevation angle of the telescope. To ensure that the UAV observation results achieve meter-level resolution, the UAV flies at an altitude of 100-120 meters; therefore, the telescope's field of view needs to be controlled within 5°.

[0046] During drone flight, drone vibrations and changes in telescope elevation angle occur. To ensure the clarity of the surface feature images captured by the visible imaging lens, the visible imaging lens automatically zooms. Therefore, the field of view of the visible imaging lens changes. The visible imaging lens can be calculated based on the captured surface feature images. The calculation method is as follows:

[0047]

[0048] The drone flies at an altitude of h above the ground, the telescope's tilt angle is α (the angle with the horizontal), and the pixel width and height of the captured surface feature image are W, respectively. pixel and H pixel The width and height of the camera sensor are W respectively. image and H image θ horizonrtal and θ vertical This represents the field of view of the visible imaging lens in the horizontal and vertical directions. Since the change in field of view caused by the change in the focal length of the visible imaging lens is linear, θ is chosen as the focal length. horizonrtal The field of view of a visible imaging lens, denoted as θ in the following text, is camera .

[0049] After obtaining the surface feature images, each surface feature image is first preprocessed, including: for each surface feature image acquired by the visible imaging lens, geometric correction is performed on the photo based on RANSAC (Random Sample Consensus Algorithm) and the automatic matching algorithm of source points to eliminate the influence of geometric deformation on the surface albedo.

[0050] Subsequently, each preprocessed surface feature image undergoes adaptive cropping to ensure a perfect match with the pixel size, specifically including: Figure 2 As shown, the field of view θ using a visible imaging lens camera and the field of view θ of the telescope module telescope The ratio of the image width of the visible imaging lens to the image width of the telescope is used to calculate the ratio, where the field of view θ of the telescope module is... telescope It is a fixed value, and its size depends on the optical fiber and lens used, that is:

[0051]

[0052] Adaptive cropping of surface feature images is performed based on aspect ratio to ensure that each image captured by the visible imaging lens corresponds perfectly with each pixel of the telescope.

[0053] Finally, semantic segmentation of land surface types based on machine learning is performed on the adaptively cropped land surface feature images. To achieve precise semantic segmentation of land surface types, a deep convolutional neural network (SegNet) model is introduced to segment the adaptively cropped land surface feature images into different land surface types i (i = 1, 2, 3...n), such as grassland, bare soil, vegetation, cement road surface, and asphalt road surface. The model output is a segmentation mask M. i (x,y), where M i (x,y) represents the land surface type i at pixel (x,y). The area S of each land surface type i is... i It can be calculated using the following formula:

[0054]

[0055] Among them, S pixel is the actual ground area of ​​each pixel, and W and H are the width and height of the surface feature image, respectively.

[0056] S2, based on the semantic segmentation results and combined with the surface albedo of each surface type, performs weighted calculation of the surface feature image to obtain the fine surface albedo of the surface feature image.

[0057] In the embodiment, for each pixel, A is used. iThe surface albedo represents the surface albedo of surface type i (surface albedo varies with wavelength; during calculation, the surface albedo value corresponding to the inversion band required for the target gas in the hyperspectral inversion process is selected). Table 1 shows common surface types and their surface albedo. The refined surface albedo is calculated using weighted surface albedo values ​​for different surface types.

[0058]

[0059] Table 1. Common land surface types and their surface albedo

[0060]

[0061]

[0062] Surface albedo varies with wavelength. The center wavelength of the desired band is used in the calculation. The conversion factor between surface albedo and wavelength is as follows:

[0063]

[0064] Where f is the conversion coefficient, k is an exponential variation coefficient related to land cover type or material, which determines the rate of change of albedo with wavelength. The exponential variation coefficient k is usually obtained by fitting experimental data and reflects the reflectivity of surface materials at different wavelengths; λ1 and λ2 are two center wavelengths. Common k values ​​for land cover types are: green vegetation k value range: 0.02–0.05; water body k value range: 0.001–0.005; soil k value range: 0.01–0.03.

[0065] Therefore, the fine surface albedo at different center wavelengths can be obtained through the conversion factor as follows:

[0066] Albedo'=f×Albedo (7)

[0067] Where Albedo' represents the fine surface albedo at the center wavelength λ2, and Albedo represents the fine surface albedo at the center wavelength λ1.

[0068] Based on the above method, the embodiment also provides an experimental example, specifically selecting an observation site, namely an environmental energy company, which is a centralized waste treatment area, and taking UAV hyperspectral remote sensing technology as an example, selecting the process of UAV hyperspectral remote sensing observation to illustrate the complete process of the method of the present invention.

[0069] In the experiment, the UAV flew at an altitude of 110 meters, and the telescope's elevation angles (denoted as 0° between the telescope and the horizontal) were set to: 8°, 10°, 12°, 14°, 17°, 22°, 32°, and 52°. This uneven elevation angle setting ensured the uniformity of ground pixels. The UAV had eight horizontal hovering points, and the telescope rotated once at each hovering point according to the elevation angle sequence. Therefore, there were a total of 64 pixels in the entire observation area.

[0070] Subsequently, adaptive cropping and fine-grained surface albedo calculation are performed on each surface feature image acquired by the visible imaging lens. The detailed process is as follows: Figure 3 As shown.

[0071] The main land surface types observed include grassland, bare soil, cement surface, broad-leaved trees, shrubs, and asphalt pavement. Combining Table 1 and formulas (5), (6), and (7), the surface albedo of the above land surface types in the ultraviolet band can be obtained (the surface albedo at the center wavelength λ = 354 nm of the NO2 inversion band 338–370 nm is calculated here), as shown in Table 2.

[0072] Table 2. Surface albedo of major surface types in the observation area at λ = 354 nm

[0073]

[0074] By combining formulas (1-7) and Table 2, the detailed surface albedo values ​​for the observation area can be obtained, such as... Figure 4 As shown, the surface types in the observation area are varied, and a refined surface albedo is crucial to the results, being more accurate than a uniform surface.

[0075] The embodiment also provides a fine surface albedo acquisition system for hyperspectral UAV remote sensing, including a UAV payload with a visible imaging lens and a telescope module, and a computing module. The UAV payload captures a surface feature image corresponding to each pixel through its visible imaging lens. The computing module preprocesses, adaptively crops, and semantically segments each captured surface feature image. Based on the semantic segmentation results and combined with the surface albedo of each surface type, it performs a weighted calculation of the surface feature images to obtain the fine surface albedo of the surface feature images.

[0076] In the embodiment system, the specific implementation methods of the UAV payload and computing module are detailed in the above embodiment method, and will not be repeated here.

[0077] The methods and systems provided in the above embodiments, as a novel scheme for calculating the surface albedo of a single pixel, can finely characterize the surface albedo of each observed pixel within the entire target area for UAV hyperspectral observations. Compared with traditional single surface albedo, this makes the observation results more accurate and reliable.

[0078] The specific embodiments described above illustrate the technical solution and beneficial effects of the present invention in detail. It should be understood that the above description is only the most preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, additions, and equivalent substitutions made within the scope of the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for obtaining fine surface albedo for hyperspectral unmanned aerial vehicle remote sensing, characterized in that, Includes the following steps: The visible imaging lens, fixed on the UAV payload, captures a ground feature image corresponding to each pixel. Each ground feature image is then preprocessed, adaptively cropped, and semantically segmented. The visible imaging lens is coaxially fixed to the telescope module on the UAV payload, ensuring that the center of the image captured by the visible imaging lens coincides with the center of the telescope module. Simultaneously, the telescope elevation angle and the visible imaging lens rotate synchronously, ensuring that each ground feature image captured by the visible imaging lens corresponds one-to-one with a pixel at each elevation angle of the telescope. Adaptive cropping is performed on each surface feature image, including: Field of view using a visible imaging lens and the field of view of the telescope module The ratio is used to calculate the image width of the visible imaging lens. and the image width of the telescope The ratio, that is: Adaptive cropping of surface feature images is performed based on aspect ratio to ensure that each image captured by the visible imaging lens corresponds perfectly with each pixel of the telescope, thus maximizing the field of view of the telescope module. It is a fixed value, and its magnitude depends on the optical fiber and optical lens used; the field of view of the visible imaging lens is... Calculated in the following way: in, The altitude of the drone above the ground. The tilt angle formed by the telescope and the horizontal direction. and These represent the pixel width and height of the captured surface feature image, respectively. and These are the width and height of the camera sensor, respectively. and This represents the field of view of the visible imaging lens in the horizontal and vertical directions. Since the change in field of view caused by the change in the focal length of the visible imaging lens is linear, we choose... The field of view of a visible imaging lens, denoted as . ; Based on the semantic segmentation results and combined with the surface albedo of each surface type, a weighted calculation of the surface feature image is performed to obtain the fine surface albedo of the surface feature image, including: For each cell, use Indicates land surface type The surface albedo, based on the surface albedo The weighted calculation of surface feature images is represented as follows: in, This represents the actual ground area of ​​each pixel. n Indicates the total number of land surface types. Albedo This represents the weighted, fine-grained surface albedo. Indicates land surface type The area is calculated using the following formula: in, and These represent the width and height of the surface feature image, respectively. It also includes: obtaining fine surface albedo at different center wavelengths through conversion factors, including: in, For conversion factors, This is an exponential variation coefficient related to land cover type or material. k The rate at which albedo changes with wavelength reflects the reflectivity of surface materials at different wavelengths. and For two center wavelengths, Indicates center wavelength Fine surface albedo at the location, Indicates center wavelength Fine surface albedo at the location.

2. The method for obtaining fine surface albedo for hyperspectral UAV remote sensing according to claim 1, characterized in that, Preprocessing is performed on each surface feature image, including: Geometric correction of surface feature images is performed based on random sampling consistency algorithm and automatic matching algorithm of common source points to eliminate the influence of geometric deformation on surface albedo in surface feature images.

3. The method for obtaining fine surface albedo for hyperspectral UAV remote sensing according to claim 1, characterized in that, Semantic segmentation is performed on each surface feature image, including: A deep convolutional neural network model is used to segment the adaptively cropped surface feature image, and the model output is a segmentation mask. ,in Represents a pixel Surface type .

4. The method for obtaining fine surface albedo for hyperspectral UAV remote sensing according to claim 3, characterized in that, The deep convolutional neural network model used is the SegNet model.

5. A refined surface albedo acquisition system for hyperspectral unmanned aerial vehicle remote sensing, characterized in that, The method for implementing any one of claims 1-4 includes a drone payload with a visible imaging lens and a telescope module, and a computing module. The UAV payload captures images of the ground features corresponding to each pixel using its visible imaging lens. The calculation module preprocesses, adaptively crops, and semantically segments each captured surface feature image. Based on the semantic segmentation results and combined with the surface albedo of each surface type, it performs a weighted calculation of the surface feature images to obtain the fine surface albedo of the surface feature images.

Images