A UAV remote sensing monitoring method and equipment for measuring vegetation cover in desert steppe.

Abstract

This invention discloses a UAV remote sensing monitoring method and device for measuring vegetation cover in desert steppe, belonging to the field of environmental monitoring technology. The core elements of its technical solution include: a UAV, a multispectral camera, a novel vegetation index, and a UAV hangar. The multispectral camera is mounted on the UAV to collect multispectral remote sensing data of desert steppe vegetation, and based on this data, a novel vegetation index—NDVI—is constructed. RGB This invention significantly improves the accuracy of inversion of sparse vegetation cover in desert steppes. The aforementioned drone hangar is an intelligent drone parking base station integrating functions such as automatic take-off and landing, parking, energy replenishment, data relay, environmental protection, and remote control. This invention enables continuous and automated monitoring of desert steppe areas, overcoming the problems of insufficient resolution and low model accuracy of existing satellite remote sensing in desert steppe areas, as well as the low efficiency of manual surveys, providing high-precision technical support for desert steppe ecosystem monitoring.

Description

Technical Field

[0001] This invention belongs to the field of environmental monitoring technology, specifically relating to a UAV remote sensing monitoring method and equipment for measuring vegetation coverage in desert grasslands. Technical Background

[0002] Desert steppe is a typical arid grassland ecosystem transitioning from grassland to desert. It is mainly distributed in areas with an annual precipitation of approximately 200 mm and is characterized by infertile soil, sparse plant species, low vegetation cover and productivity, and poor ecosystem stability. It is extremely sensitive to climate fluctuations and human activities (such as overgrazing and land reclamation), making it one of the driest and most vulnerable types of grassland ecosystems. Due to its extremely low ecological resilience, once damaged, the natural recovery process is slow and difficult, making it highly susceptible to frequent and severe natural disasters such as sandstorms, droughts, desertification, and land degradation, posing a serious threat to regional ecological security.

[0003] However, current ecological monitoring of desert steppes still has significant shortcomings: on the one hand, the vegetation stability of desert steppes is poor and the interannual fluctuation is large, making it difficult to obtain basic ecological data; on the other hand, vegetation productivity simulation and monitoring technologies suitable for large-scale, high-precision, multi-resolution spatial information integration of desert steppes are not yet mature and cannot meet the accuracy requirements of actual monitoring. The direct consequence is that relevant technical departments do not have a detailed grasp of the basic data on vegetation cover and productivity of desert steppes, and their understanding of the factors influencing their changes is not clear enough. This makes it difficult for related research results to effectively support the prediction of the development trend of desert steppe ecosystems under climate change and socio-economic development scenarios, and also makes it difficult to provide a scientific basis for ecological risk prevention and control. Therefore, it is urgent to break through high-precision monitoring technologies for the ecological characteristics of desert steppes.

[0004] With the rapid development and widespread application of satellite remote sensing technology, the scientific community has developed a variety of remote sensing-driven regional and global-scale vegetation productivity models, and derived corresponding data products. These models and products have been widely validated in the dynamic monitoring and assessment of vegetation productivity at medium and large regional scales, especially in forests, farmland, and grassland ecosystems with good vegetation growth conditions (such as meadow steppe), demonstrating outstanding applicability and reliability.

[0005] Representative vegetation productivity models include light energy use efficiency models, machine learning models, VPM models, EC-LUE models, MODIS-PSN models, GLO-PEM models, TG models, and BEPS models, some of which have generated and published vegetation productivity datasets on a global scale. The core technical logic of these models lies in using remotely sensed leaf area index (LAI) and vegetation indices (such as Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI)) as key linking points between ecosystem process models and remote sensing data. This promotes the expansion of plant leaf-scale process models to canopy, ecosystem, and landscape scales, achieving the integrated application of physiological and ecological process models and remote sensing technology. Theoretically, the ability of vegetation indices such as NDVI and EVI to monitor vegetation signals directly determines the simulation accuracy of these regional-scale vegetation productivity models, especially in areas with smaller study areas and similar climatic conditions (such as desert-steppe regions), where the performance of vegetation indices in vegetation productivity simulation is even more crucial.

[0006] To mitigate the interference of soil reflection and scattering on NDVI in sparsely vegetated areas, researchers have proposed the Soil Adjusted Vegetation Index (SAVI) and its improved versions (such as MSAVI, TSAVI, and OSAVI). These indices attempt to reduce the interference of soil background on vegetation signals by introducing a soil correction factor L. However, in practical applications, the soil correction factor L is usually set to a fixed value (default L=0.5) or determined only based on fixed environmental conditions in a specific experimental area. This limits its applicability to the index development area and makes it difficult to extend to vegetation monitoring in larger areas or complex environments. Furthermore, the spatial resolution of existing vegetation productivity data products is generally low, typically concentrated between 500m and 5.6km (or 0.05°). While this resolution may meet monitoring needs in areas with uniform vegetation cover and vigorous growth, its applicability is severely limited in sparsely vegetated areas with significant spatial heterogeneity, such as desert steppes.

[0007] Although the aforementioned existing technologies have achieved significant results in vegetation productivity simulation and vegetation dynamic monitoring studies at medium and large regional scales, the following technical problems still urgently need to be solved in the measurement of desert steppe vegetation cover and productivity monitoring:

[0008] (1) Existing vegetation productivity products have low spatial resolution (500m-5.6km), which cannot effectively capture detailed information on vegetation changes in desert steppe. Due to the characteristics of desert steppe, such as sparse vegetation, low cover, uneven spatial distribution, and significant surface heterogeneity, low-resolution data products cannot accurately reflect the local vegetation status, resulting in distorted vegetation productivity estimation results. On the one hand, low spatial resolution products cannot support accurate monitoring of grassland degradation at the small-scale, and it is also difficult to assess the actual effects of ecological engineering projects such as local vegetation restoration or degraded grassland restoration; on the other hand, they cannot accurately reflect the seasonal growth changes of desert steppe vegetation and the dynamic fluctuations in productivity caused by short-term explosive growth after rainfall, thus failing to truly reflect the growth status of vegetation.

[0009] (2) Existing models have poor applicability in desert steppe areas, with significant differences in simulation results and frequent data gaps (i.e., missing data). Taking the desert steppe of Siziwang Banner, Inner Mongolia in July 2010 as an example, eight vegetation productivity products released globally (such as MODIS-GPP, CNN-GPP, PML-GPP, and NIRv-GPP) show drastically different spatial distributions of vegetation productivity, and some products exhibit significant "gaps" due to missing data, failing to meet the needs of actual research and application. The fundamental reason is that existing models mostly assume high vegetation cover or are designed for areas with good vegetation cover, without fully considering the unique environmental characteristics of sparse vegetation in desert steppes. Desert steppes have a single plant species and low cover, and their remote sensing signals are easily interfered with by strong reflection from the soil background. This causes the values ​​of key input parameters (such as NDVI) in the model to be significantly low, thus misjudging sparse vegetation areas as "no vegetation areas," ultimately leading to failure in vegetation cover and productivity estimation, or significant deviations in results, or even directly forming areas with data gaps.

[0010] (3) Existing soil-adjusted vegetation indices (SAVI, MSAVI, etc.) use a fixed soil correction factor L (default L=0.5) or correction methods developed only for specific regions, making them difficult to flexibly apply to differences in vegetation cover and soil type variations in different areas of desert steppe. Furthermore, these indices cannot address the inherent characteristics of desert steppe vegetation, such as seasonal dynamics (e.g., alternation of growth and decay) and explosive growth after rainfall, leading to a significant decrease in their monitoring accuracy in desert steppe areas and failing to effectively reduce soil background interference. Therefore, these indices cannot meet the needs for accurate measurement of desert steppe vegetation cover.

[0011] (4) Lack of automated and high-precision monitoring methods applicable to desert steppe: Traditional desert steppe vegetation cover measurement relies on manual surveys, which are time-consuming and labor-intensive, and difficult to achieve large-scale and continuous monitoring; existing satellite remote sensing technology suffers from the aforementioned problems of low resolution and significant differences in results between various models, while UAV remote sensing technology, although possessing the advantage of high resolution, has not yet formed a standardized monitoring method for desert steppe (such as flight parameter settings, data processing procedures, and especially the construction of new vegetation indices). At the same time, the lack of supporting automated monitoring equipment (such as long-term unattended UAV take-off and landing and data transmission systems) makes it difficult to achieve efficient, continuous, and high-precision monitoring of desert steppe vegetation cover.

[0012] In view of this, the present invention aims to provide a UAV remote sensing monitoring method and equipment architecture for measuring vegetation cover in desert steppes, so as to effectively solve the above-mentioned technical problems. Summary of the Invention

[0013] The purpose of this invention is to provide a UAV remote sensing monitoring method and equipment for measuring vegetation cover in desert steppes. This invention achieves this by constructing a novel vegetation index—NDVI. RGB This technology effectively suppresses the interference of soil background on remote sensing monitoring of desert steppe vegetation coverage, significantly improves the inversion accuracy of sparse vegetation coverage in desert steppe, and, combined with UAV remote sensing monitoring equipment with automatic charging, communication and remote control functions, realizes long-term, unattended automated monitoring of vast desert steppe areas. It solves the problems of insufficient resolution and low model accuracy of existing satellite remote sensing products in desert steppe areas, as well as the low efficiency and high cost of manual surveys.

[0014] The above technical objectives are achieved through the following technical solution: a drone remote sensing monitoring device for measuring vegetation coverage in desert steppe, comprising a drone, a multispectral camera, and a drone hangar; the multispectral camera is mounted on the drone and is used to collect multispectral remote sensing data of desert steppe; while the drone hangar is a drone intelligent parking base station that integrates functions such as automatic take-off and landing, parking, energy replenishment, data relay, environmental protection, and remote control of drones.

[0015] The present invention is further configured such that: the drone hangar includes an external structure, a power supply and charging module, a communication and data module, a drone docking and take-off and landing module, and an environmental monitoring and safety module;

[0016] The external structure is composed of a protective shell made of high-strength composite materials or metal, which has high temperature resistance, UV protection, and wind and sand protection properties, and has a built-in active temperature control component. The temperature control component is a small fan, a TEC cooling chip, or a phase change material, used to ensure that the drone battery is not damaged under extreme temperature differences. A solar panel is arranged on the top of the external structure to power the drone hangar and the drone battery.

[0017] The power and charging module includes a fast charging interface to support automatic docking and charging of drone batteries; it has an internal battery compartment with multiple spare batteries pre-installed and is equipped with a robotic arm or automatic battery swapping mechanism; in addition, the power and charging module is also equipped with an intelligent power management unit for real-time monitoring of battery status and dynamic allocation of charging priorities.

[0018] The communication and data module includes 4G / 5G or satellite communication components for remote control and data backhaul; in addition, the communication and data module is also equipped with a GPS positioning component to ensure accurate positioning and support the networking deployment of multiple drone hangars in the future.

[0019] The UAV docking and take-off / landing module is equipped with a sliding rail or lifting platform, and combined with a communication and data module, it supports the automatic opening and closing of the hangar cover to guide the UAV to take off and land safely and dock precisely. The UAV docking and take-off / landing module also has a dustproof sealing function and is equipped with infrared, ultrasonic or RTK guidance and positioning components.

[0020] The environmental monitoring and safety module includes temperature sensors, humidity sensors, wind speed sensors, and dust concentration sensors to acquire environmental data and assess the flight safety of the drone. Simultaneously, the module is equipped with an electronic lock and remote alarm components to protect the drone hangar, the drone itself, and the multispectral camera. Furthermore, the module includes fans, filters, or vibration devices to periodically clean dust from the air inlets and solar panels.

[0021] The present invention is further configured such that: the UAV hangars are evenly distributed in the desert grassland test area, with a quantity of 5, and the arrangement location is determined by selecting representative sample plots based on the pixel resolution characteristics of remote sensing images, supporting the verification and correction of remote sensing data with different resolutions of 30m, 250m, 500m and 1000m.

[0022] The present invention is further configured such that the equipment also includes a ground spectrometer, a meteorological sensor and a vegetation phenology camera, all of which are set up near the hangar in the test area to collect surface reflectance, micro-meteorological data and vegetation phenology information to assist in monitoring the vegetation growth process and estimating vegetation coverage.

[0023] The present invention is further configured to: establish a multi-level verification and integrated application chain from ground observation to UAV network observation to satellite remote sensing observation, realize accurate monitoring of vegetation coverage and growth status in large-scale desert steppe, and support quantitative correction of remote sensing data with different spatial resolutions of 30m, 250m, 500m and 1000m, thereby promoting the integrated application of multi-source data from air, space and ground in desert steppe areas.

[0024] This invention also provides a UAV remote sensing monitoring method for measuring vegetation cover in desert steppes, comprising the following:

[0025] S1: Deploy unmanned aerial vehicle (UAV) remote sensing monitoring equipment in the desert steppe experimental area;

[0026] S2: Use a drone equipped with a multispectral camera to acquire multispectral images of the desert steppe test area at a preset flight altitude;

[0027] S3: Preprocess the acquired multispectral images;

[0028] S4: Divide the preprocessed multispectral image into several sub-blocks, and use an autoencoder plus KMeans clustering algorithm to perform unsupervised classification of vegetation and soil in each sub-block, thereby obtaining the number of vegetation pixels and soil pixels in each sub-block.

[0029] S5: Calculate the vegetation cover and multispectral reflectance data for each sub-block based on the number of vegetation pixels and soil pixels in each sub-block;

[0030] S6: Based on the vegetation cover and multispectral reflectance data of each sub-block, construct a new improved vegetation index;

[0031] S7: By utilizing the improved linear relationship between vegetation index and vegetation coverage, high-precision estimation of vegetation coverage in desert steppe can be achieved.

[0032] The present invention is further configured such that: in step S2, the drone is a DJI MK300 model, and the multispectral camera it carries is a YUSENSE MS600Pro. The multispectral camera covers blue light, green light, red light, red edge 1, red edge 2 and near-infrared typical vegetation remote sensing detection bands, with center wavelengths of 450 nm, 555 nm, 660 nm, 720 nm, 760 nm and 840 nm, respectively.

[0033] The present invention is further configured such that, in step S2, the UAV has a preset flight altitude of 15m, a lateral overlap rate of 70%, a forward overlap rate of 80%, and a flight speed of 1.5 m / s.

[0034] The present invention is further configured such that: in step S3, the preprocessing of the multispectral image includes data export and track stitching, radiometric correction, and geometric correction and registration; wherein,

[0035] During the data export and flight path stitching process, the original images are exported from the multispectral camera on the UAV, and multiple original images acquired from the same flight mission are stitched together to generate a complete seamless stitched image of the test area.

[0036] During radiometric correction, a whiteboard attached to the camera is used to correct reflectivity in order to eliminate the effects of changes in light intensity and to remove noise signals generated by the sensor itself.

[0037] During the geometric correction and registration process, distortion, viewpoint shift and terrain effects are corrected to achieve strict spatial alignment of images in different bands, avoiding misalignment in subsequent calculations. At the same time, RTK or ground control points are used to geolocate the images, eliminating deformation caused by terrain undulations or camera tilt, thereby giving the images map attributes.

[0038] The present invention is further configured such that: in step S4, the preprocessed multispectral image is divided into 10 small blocks along the row and column directions, resulting in a total of 100 sub-blocks; and unsupervised classification of vegetation and soil is performed on each sub-block to generate a classification map of vegetation and soil for each sub-block. The process includes:

[0039] Effective pixels are filtered out from the preprocessed multispectral images, and all pixels with band values ​​less than zero are removed.

[0040] Median filling and standardization are performed on effective pixels. Greenness features are introduced and standardized. The standardized six-dimensional spectral features and greenness features are combined as seven-dimensional input variables for the autoencoder plus KMeans clustering algorithm.

[0041] Latent space features are obtained by performing nonlinear dimensionality reduction and deep feature extraction on seven-dimensional input variables using an autoencoder.

[0042] The latent space features and weighted greenness features are input into the KMeans clustering algorithm. The vegetation and soil categories are determined based on the median greenness of each cluster, and a classification map of vegetation and soil for each sub-block is generated.

[0043] The present invention is further configured such that, in step S5, the process for calculating the vegetation cover of each sub-block is as follows:

[0044] Based on the classification results of vegetation and soil pixels in each sub-block, vegetation cover is defined as the proportion of green vegetation pixels in each sub-block to the total number of pixels in that sub-block. The calculation formula is as follows:

[0045]

[0046] Among them, FVC t V represents the vegetation cover of subblock t. t N represents the number of pixels within sub-block t that are identified as green vegetation. t This represents the total number of pixels within the sub-block, which is the sum of vegetation pixels and soil pixels.

[0047] The present invention is further configured such that: in step S5, the process of calculating the multispectral reflectance of each sub-block is as follows: the preprocessed multispectral image is subjected to the same block processing, and the average value of effective pixels with pixel values ​​greater than zero is statistically calculated for each band in each sub-block, which is used as the average reflectance of each band in each sub-block.

[0048] The present invention is further configured such that, in step S6, the process of constructing the novel improved vegetation index is as follows: The reflectance of the blue and green light bands is introduced into the calculation framework of the vegetation index NDVI, and combined with the near-infrared and red light bands used in the NDVI calculation, thereby forming the novel improved vegetation index; its calculation formula is:

[0049]

[0050] R, G, B, and NIR represent the reflectance of red, green, blue, and near-infrared bands, respectively. This improved vegetation index is used to achieve high-precision measurement of vegetation coverage in desert steppes.

[0051] In summary, the present invention has the following beneficial effects:

[0052] 1. This invention significantly improves the measurement accuracy of vegetation cover in desert steppes and effectively reduces the interference of soil and atmosphere on vegetation remote sensing monitoring. Addressing the shortcomings of existing vegetation indices (such as NDVI and SAVI) in desert steppe applications—being affected by strong soil background reflection and being difficult to adapt to sparsely vegetated environments due to fixed correction parameters—this invention proposes a novel improved vegetation index—NDVI. RGB This index innovatively introduces green and blue light reflectance into the traditional NDVI calculation architecture. By enhancing the difference between the "high reflectance band of vegetation (near-infrared + green light)" and the "strong absorption band of vegetation (red + blue light)," it can effectively reduce the impact of soil background reflection and atmospheric scattering on the monitoring results of desert steppe vegetation. Experimental data show that NDVI... RGB The coefficient of determination (R²) for fitting vegetation cover reached 0.95, significantly higher than the 0.93 of traditional NDVI. In areas with vegetation cover below 50%, NDVI... RGB In addition to demonstrating stronger anti-interference capabilities, it also avoids the problem of existing indices misjudging sparse vegetation as "no vegetation areas," thus making it more suitable for the ecological characteristics of desert steppe with sparse vegetation and strong surface heterogeneity, providing a more reliable basis for high-precision quantification of desert steppe vegetation coverage.

[0053] 2. This invention enables large-scale, long-term automated monitoring of desert steppes, significantly improving monitoring efficiency and overcoming the drawbacks of traditional manual surveys, which are time-consuming and labor-intensive, as well as the limitations of existing drone monitoring, which relies on manual operation and cannot continuously cover large areas. The drone remote sensing monitoring equipment in this invention features "automatic storage, intelligent power supply, remote communication, and environmental protection." It achieves energy self-sufficiency through a solar-powered top cover, shortens drone standby time with a robotic arm / automatic battery swapping mechanism, and supports remote control and data transmission via 4G / 5G / satellite links. Furthermore, the equipment is equipped with active temperature control, a windproof and sand-resistant shell, and a self-cleaning device, enabling it to adapt to the harsh environment of extreme temperature differences and high dust levels in desert steppes. This monitoring equipment can achieve periodic automatic take-off and landing monitoring, completing continuous data collection over large areas of desert steppes such as Siziwang Banner without manual intervention. This greatly improves the efficiency of traditional manual surveys while avoiding errors caused by manual operation, meeting the needs of "large-scale, high-frequency, and long-term" monitoring of desert steppe vegetation growth.

[0054] 3. The novel vegetation index provided by this invention possesses strong transplantability and robustness, and can adapt to different desert steppe ecological scenarios. Addressing the issue that existing soil-adjusted vegetation indices (such as MSAVI and TSAVI) are only applicable to specific experimental areas and their accuracy drops sharply when expanded to other regions, this invention selects a typical verification area (sandy / loamy soil with significant differences in vegetation cover) in Siziwang Banner, Inner Mongolia, to test the transplantability and robustness of the novel vegetation index, confirming the effectiveness of NDVI. RGB High robustness and portability. In the validation region, NDVI RGB The R² of the fit with vegetation cover remained at 0.92, significantly better than the 0.89 of traditional NDVI. Meanwhile, NDVI... RGB It can stably capture vegetation signals under different vegetation cover levels (17.9%~75.5%) and different soil types, and is not limited by differences in local ecological conditions. This characteristic makes NDVI... RGB It is expected to be extended to other desert and grassland areas in northern my country, without the need to readjust model parameters for different regions, thus lowering the technical application threshold and expanding the applicability of the monitoring method.

[0055] 4. This invention constructs a multi-scale data verification chain, aiming to improve the application accuracy of satellite remote sensing data and support refined ecological assessment. By establishing a multi-level verification system of "ground-based fixed sensor observation (spectrometer, meteorological sensor) — UAV centimeter-level high-resolution data — satellite 30~1000m resolution data," this invention can effectively correct the vegetation index (NDVI, EVI) and vegetation cover estimation model of satellite imagery, making up for the shortcomings of large-scale satellite data in "sparse vegetation identification and local degradation capture" in desert grasslands. For example, using UAV multispectral data and new improved vegetation indices, small-scale grassland degradation patches can be accurately identified, which can solve the current contradiction of "insufficient accuracy of large-scale data and lack of small-scale data," providing technical support for the refined assessment of the benefits of the "Three-North" wind and sand source control and vegetation restoration projects.

[0056] 5. The synergistic application of the method and equipment provided by this invention will meet the needs of remote sensing monitoring of desert steppe vegetation, providing technical and data support for ecological security and livestock development. By closely focusing on the characteristics of desert steppe—"high ecological vulnerability, poor self-repair capacity, and important ecological security status"—this invention provides relevant solutions: On the one hand, high-precision vegetation cover data can quantify the degree of grassland degradation and dynamically monitor changes in vegetation productivity, thus providing basic data and key technologies for assessing the impact of climate change and human activities on desert steppe, overcoming the shortcomings of "insufficient basic data and unclear influencing factors" in existing research; on the other hand, automated monitoring equipment using unmanned aerial vehicles (UAVs) can provide real-time feedback on grassland growth, providing timely information for the sustainable development of livestock farming. Ultimately, with precise ecological monitoring and data support, this invention will contribute to the construction of an ecological security barrier in northern my country, providing a scientific basis for desert steppe dust storm prevention and control, land desertification management, and ecological risk prediction. Attached Figure Description

[0057] Figure 1 This is a schematic diagram illustrating the working principle of the UAV remote sensing monitoring device in this embodiment of the invention;

[0058] Figure 2 This is a schematic diagram showing the layout of the drone hangar in an embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of a drone hangar in an embodiment of the present invention;

[0060] Figure 4 This is a schematic diagram and numbering diagram of multispectral image block processing of the test area UAV in this embodiment of the invention;

[0061] Figure 5 This is a schematic diagram of the multispectral color composite image of the test area and its block superposition results in an embodiment of the present invention;

[0062] Figure 6This is a schematic diagram illustrating the application effect of the unsupervised classification method based on autoencoder and Kmeans clustering algorithm in the test area in this embodiment of the invention;

[0063] Figure 7 This is a schematic diagram illustrating the variation characteristics of vegetation coverage in each sub-block in an embodiment of the present invention;

[0064] Figure 8 This is a schematic diagram illustrating the variation characteristics of the average reflectance of each sub-block in an embodiment of the present invention;

[0065] Figure 9 This is a schematic diagram showing the curves of reflectance variation with wavelength under different vegetation coverage in the embodiments of the present invention, as well as the relative rate of change of reflectance with vegetation coverage in each band.

[0066] Figure 10 This is a schematic diagram illustrating the correlation between commonly used vegetation indices and vegetation coverage in embodiments of the present invention;

[0067] Figure 11 This is a novel improved vegetation index NDVI in the embodiments of the present invention. RGB A schematic diagram illustrating the correlation between vegetation cover and vegetation coverage;

[0068] Figure 12 This is a schematic diagram of the multispectral color composite image of the verification area and its block superposition result in an embodiment of the present invention;

[0069] Figure 13 This is a schematic diagram illustrating the application effect of the unsupervised classification method based on autoencoder and Kmeans clustering algorithm in the validation area in an embodiment of the present invention;

[0070] Figure 14 This is a schematic diagram showing the curves of reflectance variation with wavelength under different vegetation coverage in the verification area in this embodiment of the invention, as well as the relative rate of change of reflectance with vegetation coverage in each band.

[0071] Figure 15 In this embodiment of the invention, the traditional NDVI and the novel vegetation index NDVI are used in the verification area. RGB A schematic diagram illustrating the correlation between vegetation cover and vegetation coverage. Detailed Implementation

[0072] The following will be combined with the appendix Figures 1-15 The present invention will now be described in more detail.

[0073] Example 1: A UAV remote sensing monitoring method for measuring vegetation cover in desert steppe, comprising the following steps:

[0074] S1, the desert steppe test area and the configuration of the UAVs and multispectral cameras used:

[0075] The experimental area selected in this embodiment is located in the central part of Siziwang Banner, Ulanqab City, Inner Mongolia Autonomous Region, with geographical coordinates between 41°43'–41°47' N and 111°53'–111°57' E, at an altitude of approximately 1500 m. Situated in the central part of the Inner Mongolia Plateau, it represents a typical desert-steppe ecosystem in northern China. The region is approximately 128 km from Hohhot City and 30 km from Wulanhua Town. The climate of this area is a mid-latitude semi-arid continental monsoon climate, with an average annual temperature of approximately 3-5℃. The average temperature in January can drop below -15℃, while the average temperature in July can reach above 20℃. Annual precipitation is approximately 250 mm, with a highly uneven seasonal distribution, mainly concentrated from July to September. The region experiences abundant sunshine and strong evaporation, with evaporation far exceeding precipitation.

[0076] The drone used in this embodiment is a DJI MK300 model, equipped with a YUSENSEMS600Pro multispectral camera, which covers typical vegetation remote sensing bands such as blue light, green light, red light, red edge 1, red edge 2 and near infrared, with center wavelengths of 450 nm, 555 nm, 660 nm, 720 nm, 760 nm and 840 nm, respectively, which can provide high-quality multispectral data support for the accurate measurement and monitoring of vegetation coverage in desert steppe.

[0077] S2, UAV multispectral image acquisition in the desert steppe test area:

[0078] This embodiment revolves around the core technical line of "flight platform → parameter setting → data acquisition → data storage → subsequent use". Combining the spatial resolution characteristics of the YUSENSE MS600Pro multispectral camera, and taking into account the actual situation of low vegetation cover and small size of desert grassland, the flight altitude is set to 15 m, the lateral overlap rate is 70%, the forward overlap rate is 80%, and the flight speed is set to 1.5 m / s to ensure that the acquired multispectral remote sensing images can clearly reflect the vegetation distribution characteristics of the test area while ensuring coverage integrity.

[0079] S3. Preprocessing of UAV multispectral images:

[0080] In this embodiment, the preprocessing of UAV multispectral data includes data processing, radiometric correction, geometric and spectral registration, image stitching and orthorectification, cropping and denoising, and band synthesis. These preprocessing steps help improve the quality and consistency of the image data, providing reliable foundational data for subsequent analyses (such as vegetation cover measurement, image classification, and image segmentation). The specific process is as follows:

[0081] Data export and track stitching

[0082] The original images are exported from the multispectral camera on the drone, and multiple images acquired from the same flight mission are stitched together to generate a complete seamless image of the test area.

[0083] Radiation correction

[0084] Whiteboard calibration: Use the whiteboard (or standard reflector) that comes with the camera to perform reflectivity calibration, eliminating the effects of changes in light intensity.

[0085] Dark current correction: Eliminates noise signals generated by the sensor itself.

[0086] Geometric correction and registration

[0087] Image geometric correction: correcting distortion, viewpoint shift, and terrain effects.

[0088] Multispectral channel registration: Ensures that images of different bands are strictly aligned in space to avoid misalignment in subsequent calculations.

[0089] Geographic coordinate correction: Geographically locate the image using RTK or ground control points (GCP).

[0090] Orthorectification: Eliminates distortion caused by terrain undulations or camera tilt, giving the image map-like properties.

[0091] S4. Sub-block segmentation of UAV multispectral imagery and unsupervised classification of vegetation and soil:

[0092] In this embodiment, to obtain more data sequences and construct a relationship model between multispectral vegetation index and vegetation coverage, the preprocessed UAV multispectral image is divided into 10 small blocks along the row and column directions, resulting in a total of 100 sub-blocks. Figure 4 This diagram illustrates the block processing and numbering of multispectral images from UAVs in the test area. Figure 5 This is a multispectral color composite image (RGB=321 bands) of the test area and its block overlay results.

[0093] S5. Calculate vegetation cover and multispectral reflectance data based on the sub-block classification results:

[0094] (1) Vegetation coverage measurement: This process is divided into two steps, namely the classification of vegetation pixels and soil pixels, and the calculation of vegetation coverage.

[0095] Vegetation and Soil Pixel Classification: To construct a novel vegetation index model for calculating vegetation cover, this embodiment proposes a vegetation and soil classification method for UAV multispectral imagery based on an autoencoder and KMeans clustering approach. First, the preprocessed six-band imagery is filtered for effective pixels, removing all pixels with band values ​​of zero (i.e., background pixels not captured). Next, median imputation and standardization are performed on the effective pixels in each band, and greenness information is introduced for further standardization. These are then combined and used as input data. Finally, the standardized six-dimensional spectral and greenness features are combined into a seven-dimensional input, and an autoencoder is used to perform nonlinear dimensionality reduction and depth feature extraction under unsupervised conditions. In the clustering stage, latent space features and weighted greenness features are jointly input into the KMeans clustering algorithm (K=2), and vegetation and soil categories are automatically determined based on the median greenness of each cluster. Finally, a classification map of the entire image is generated, where pixel values ​​of 0, 1, and 2 represent invalid pixels, soil, and green vegetation, respectively. Visual representation and file generation in ENVI software are also completed. This process achieves robust separation of vegetation and bare soil without the need for annotation, providing reliable basic data for the construction of new multispectral vegetation indices and vegetation cover inversion. Figure 6 The application effect of the unsupervised classification method based on "autoencoder + Kmeans" clustering in the experimental area is demonstrated. Yellow and green pixels represent soil and vegetation, respectively, and a 10×10 block grid (red line) is superimposed for subsequent analysis.

[0096] Vegetation Cover Calculation: Based on the classification results of the "autoencoder + KMeans" clustering algorithm, Fractional Vegetation Cover (FVC) is defined as the proportion of green vegetation pixels to the total number of pixels in each grid (or sub-block) spatial unit. By calculating the FVC of all sub-blocks sequentially, vegetation cover data for all sub-blocks in the entire image of the experimental area can be obtained, thus providing key foundational data for subsequently constructing a model of the relationship between vegetation cover and spectral reflectance. The formula for calculating FVC is:

[0097]

[0098] Among them, V t N represents the number of pixels within sub-block t that are identified as green vegetation. t This represents the total number of pixels within the sub-block, which is the sum of vegetation pixels and soil pixels. Figure 7 This is a schematic diagram illustrating the variation characteristics of vegetation coverage in each sub-block in an embodiment of the present invention.

[0099] (2) Calculation of multispectral reflectance:

[0100] In this embodiment, the calculation method for multispectral reflectance is basically the same as the aforementioned calculation process for vegetation cover. Specifically, the UAV images of the six bands in the test area are processed in the same block manner, and the average reflectance is calculated band by band within each block. For each block, the effective pixels with pixel values ​​greater than zero are counted and their average value is calculated to obtain the average reflectance of the corresponding band; then the calculation results of each block are summarized to generate a result table. Figure 8 The reflectance fluctuation characteristics of each sub-block in six bands.

[0101] (3) Sensitivity analysis of multispectral reflectance to vegetation cover:

[0102] To analyze the sensitivity of reflectance in the blue, green, red, red-edge 1, red-edge 2, and near-infrared bands to changes in vegetation cover, the following were used: Figure 6 The sub-blocks were divided into eight intervals according to vegetation coverage: <20%, 20%–30%, 30%–40%, 40%–50%, 60%–70%, 70%–80%, 80%–90%, and 90%–100%. The average vegetation coverage and average reflectance of the six bands were then calculated for each interval, and the variation of reflectance with wavelength under different vegetation coverage was analyzed. Figure 9 (a) shows the reflectance variation curves with wavelength under different vegetation cover levels in this embodiment of the invention. Simultaneously, the reflectance change rate during the process of increasing vegetation cover was also calculated. Figure 9 b), its calculation formula is:

[0103]

[0104] Among them, R ratio Represents the rate of change of reflectance in each band, with high and low representing high and low vegetation cover conditions, respectively, and R and VC representing reflectance and vegetation cover in each band, respectively. Figure 9 (b) is a schematic diagram showing the relative rate of change of reflectance of each band with vegetation cover.

[0105] from Figure 9 (a) It can be seen that the reflectance of each band exhibits a significant pattern under different vegetation cover conditions. First, the reflectance gradually increases with increasing wavelength, showing a significant jump at approximately 700 nm, reaching a relatively high value in the near-infrared band, which is consistent with typical vegetation spectral characteristics. Second, with increasing cover, the reflectance of each band generally shows a decreasing trend. This reflects the strong absorption of light energy by vegetation in the visible light region and the weakening of soil background contribution under high cover conditions. Among different bands, the red band ( Figure 9The third data point from the left in region a is most sensitive to changes in vegetation cover, exhibiting the largest decrease in reflectance with increasing cover, demonstrating high response characteristics. Therefore, the red band plays a crucial role in vegetation cover retrieval and the construction of novel vegetation indices. When vegetation cover exceeds 80%, the difference between the curves in the red-edge and near-infrared bands gradually narrows, indicating that under high cover conditions, the reflectance difference tends to saturate. These characteristics are also confirmed in the analysis of the relative rate of change of reflectance with vegetation cover in each band (…). Figure 9 b). From Figure 9 (b) It can be seen that the reflectance change rate is the largest with increasing vegetation cover, indicating that it is the most sensitive to the change in vegetation cover and is the main indicator band for changes in vegetation cover. In addition, the rate of change in the green light band is also relatively large under medium and low vegetation cover conditions, indicating that green light can reflect the differences in vegetation cover under medium and low vegetation cover conditions. In contrast, the reflectance change rate in the near-infrared band is smaller and tends to stabilize under high vegetation cover conditions, indicating that when the vegetation cover is high, the influence of near-infrared reflectance on vegetation cover gradually weakens, and there is a certain saturation effect.

[0106] S6. Construct an improved vegetation index based on vegetation cover and multispectral reflectance data for each sub-block:

[0107] Before constructing a new improved vegetation index, this embodiment analyzed the vegetation indices currently widely used in vegetation remote sensing research and evaluated their performance in predicting desert steppe vegetation cover. Figure 10 The results showed that NDVI had the strongest linear correlation with vegetation cover, with an R² value of 0.93, demonstrating its significant predictive ability. Other vegetation indices showed lower predictive ability than NDVI. Therefore, the core objective of this embodiment is to propose a novel vegetation index based on the traditional NDVI calculation framework to improve the accuracy of desert steppe vegetation cover measurement. The formula for calculating NDVI is as follows:

[0108]

[0109] In the formula, R and NIR represent the reflectivity of red light and near-infrared bands, respectively.

[0110] Therefore, combining the sensitivity analysis results of multispectral reflectance on vegetation cover in S5, the core objective of this invention is to comprehensively and rationally utilize red, green, blue, and near-infrared band information during the construction of a new vegetation index, in order to reduce the interference of soil background on monitoring results, and thus propose a new and improved vegetation index suitable for desert steppe. The core idea of ​​this embodiment is: based on the traditional NDVI framework (which has the highest monitoring accuracy, see...) Figure 10This invention introduces green and blue light bands to balance the impact of soil reflection and atmospheric scattering on remote sensing monitoring of desert steppe vegetation cover. The specific invention process, physical significance, and corresponding calculation formulas are as follows:

[0111] Adding the green light band (G): In the visible light spectrum, green light exhibits relatively high reflectivity to vegetation (hence the green color of leaves), and its reflectivity is correlated with vegetation cover. Introducing green light, along with the near-infrared band, into the molecule of the NDVI calculation formula enhances the sensitivity to "leaf reflectivity characteristics." This adjustment further highlights the difference between "high-reflectivity vegetation bands (near-infrared reflectivity + green light reflectivity)" and "strong-absorption vegetation bands (red light reflectivity + blue light reflectivity)," thereby improving the index's responsiveness to changes in vegetation cover.

[0112] Introducing the blue light band (B): Blue light also exhibits a significant absorption effect on vegetation (similar to red light), but it is easily affected by atmospheric scattering. Introducing blue light into the construction of the novel vegetation index not only counteracts the interference effects of atmospheric and soil background but also enhances the strong absorption of blue light by vegetation. Therefore, in low-coverage or desert-steppe conditions, this helps improve the accuracy of the index in monitoring changes in vegetation cover. Based on this, the improved formula for calculating the novel vegetation index is:

[0113]

[0114] In the formula, R, G, B, and NIR represent the reflectance of red, green, blue, and near-infrared light bands, respectively. The numerator of the formula highlights the difference between "high reflectance and strong absorption" of vegetation, meaning that the value increases accordingly with the increase of vegetation coverage; the denominator provides a "spectral comparison benchmark" to reduce the interference of single-band anomalies or noise on the results and ensure the accuracy of the monitoring results.

[0115] S7. Achieving high-precision inversion of desert steppe vegetation cover based on the linear relationship between a novel improved vegetation index and vegetation cover:

[0116] from Figure 11 It can be seen that the new improved vegetation index NDVI RGB A significant linear relationship was observed between the new vegetation index and vegetation cover, with a coefficient of determination of 0.95. Compared to the correlation between the traditional NDVI and vegetation cover (0.93), the fitting accuracy was significantly improved. This indicates that by introducing green and blue light reflectance, the new vegetation index demonstrates a significant improvement in predicting vegetation cover, effectively mitigating the interference of soil background in low-coverage areas and enhancing the sensitivity of the vegetation index to changes in vegetation cover.

[0117] S8. Transplantability test of the novel improved vegetation index:

[0118] To verify the applicability (transplantability) of the novel vegetation index in other areas of desert steppe, this embodiment selected a typical desert steppe sample area in Siziwang Banner for verification. This area has flat terrain, low vegetation cover, and soil types mainly consisting of sandy and loamy soils. Its ecological environment characteristics and vegetation cover level differ significantly from the original experimental area (i.e., the development area of ​​this embodiment). Figure 12 and Figure 13 In the verification area, the reflectance variation curves with wavelength under different vegetation cover, and the relative change rate of reflectance in each band with vegetation cover ( Figure 14 The results are all basically consistent with the results of the development area of ​​this invention. This can be summarized as follows:

[0119] Overall trend: As the wavelength increases from 450nm to 870nm, the reflectance gradually increases, exhibiting typical vegetation spectral characteristics.

[0120] In the visible light region, reflectance gradually decreases with increasing vegetation cover in the blue, green, and red light bands. The variation is most pronounced in red light, with a significant decrease in reflectance at higher vegetation cover, reflecting the strong absorption effect of chlorophyll on red light.

[0121] Near-infrared region: Reflectance increases significantly with increasing vegetation cover. In high-coverage plots (e.g., VC≈75.5%), near-infrared reflectance is close to 0.3, while in low-coverage plots (VC≈17.9%) it decreases to 0.25. This indicates that the strong reflectance effect caused by leaf cell structure is more significant under high cover.

[0122] Depend on Figure 15 (a) As can be seen, in the validation area, the traditional NDVI shows a strong linear relationship with vegetation cover, but it is still significantly affected by soil background, which limits the fitting accuracy, resulting in a system value of 0.89. The novel vegetation index NDVI proposed in this embodiment… RGB The correlation with vegetation cover remained high in the validation area, with a coefficient of determination of 0.92. Figure 15 (b) This is significantly better than the 0.89 of traditional NDVI. This indicates that by introducing the green and blue light bands, NDVI... RGB It can effectively reduce the interference of soil background, so that it remains stable in the vegetation cover estimation of the validation area, has good model transferability, and is expected to be promoted and applied in the desert steppe areas of northern my country.

[0123] Example 2: A UAV remote sensing monitoring device for measuring vegetation cover in desert steppe

[0124] The total area of ​​Siziwang Banner is approximately 25,500 km². 2 Of which, grassland area reaches 21,000 km². 2Due to the vast area of ​​grasslands, traditional manual surveys are not only time-consuming and labor-intensive, but also difficult to achieve large-scale, continuous monitoring. To address this issue, this embodiment proposes a UAV remote sensing monitoring device for measuring vegetation cover in desert steppes. This device establishes a multi-level verification and integrated application chain of "ground observation—UAV network observation—satellite remote sensing observation" to achieve precise UAV remote sensing monitoring of large-scale desert steppe vegetation cover and growth processes. It also supports quantitative correction of remote sensing data at different spatial resolutions of 30m, 250m, 500m, and 1000m, promoting the integrated application of multi-source data from air, space, and ground in desert steppe regions. Specifically, the UAV network observation involves evenly distributing five UAV hangars within the test area. Each hangar houses UAVs, multispectral cameras, batteries, and a remote control system. Remote control enables the opening and closing of the hangars, intelligent take-off and landing of the UAVs, and multispectral imaging of the target ground. The drone hangar is an intelligent drone parking base station that integrates functions such as automatic take-off and landing, parking, energy replenishment, data relay, environmental protection, and remote control. The hangar deployment scheme involves evenly distributing five hangars within the desert-steppe experimental area. The locations are determined by selecting representative plots based on the pixel resolution characteristics of the remote sensing images, supporting the verification and correction of remote sensing data at different resolutions of 30m, 250m, 500m, and 1000m. Figure 2 This is a schematic diagram of the layout of the drone hangar in an embodiment of the present invention.

[0125] (1) Core functions and modules of the hangar:

[0126] Figure 3 This is a schematic diagram of a drone hangar in an embodiment of the present invention;

[0127] External structure:

[0128] Protective shell: Made of high-strength composite materials or metal shell, with high temperature resistance, UV protection and wind and sand protection properties.

[0129] Temperature control system: Built-in active temperature control (small fan, TEC cooling plate or phase change material) to ensure that the drone battery is not damaged under extreme temperature differences.

[0130] Solar roof: Solar panels are installed on top of the enclosure to power the enclosure itself and to charge the drone hangar and drone batteries.

[0131] Power supply and charging module:

[0132] Fast charging port: Supports automatic docking and charging of drone batteries.

[0133] Battery compartment storage: Multiple spare batteries can be pre-installed, supporting robotic arms / automatic battery swapping mechanisms to achieve "battery swapping instead of charging", shortening standby time.

[0134] Intelligent power management: Real-time monitoring of battery status and dynamic allocation of charging priorities.

[0135] Communication and Data Module:

[0136] Long-distance communication: Utilizing 4G / 5G / satellite links to achieve remote control and data backhaul.

[0137] Geographic positioning: The chassis is equipped with GPS for precise positioning and subsequent deployment of multi-hangar networking.

[0138] Drone docking and take-off / landing module:

[0139] Automatic opening and closing cover: The drone can take off and land automatically, and the cover has a sand and dust seal.

[0140] Guidance and positioning system: guides the drone to a precise docking location via infrared / ultrasound / RTK.

[0141] Take-off and landing platform: A sliding rail / lift platform can be used to ensure the smooth take-off and landing of the drone.

[0142] Environmental monitoring and safety module:

[0143] Sensor integration: monitoring of temperature, humidity, wind speed, and dust concentration to assess flight safety.

[0144] Anti-theft and anti-vandalism: Equipped with electronic locks and remote alarm functions to ensure equipment security.

[0145] Self-cleaning device: Equipped with a fan, filter or vibration device to regularly clean the dust from the air inlet and solar panels.

[0146] (2) Intelligent monitoring of vegetation cover in desert grassland based on UAV remote sensing monitoring equipment:

[0147] Figure 1 This is a schematic diagram illustrating the working principle of the UAV remote sensing monitoring device in this embodiment of the invention. Its working content can be summarized as follows:

[0148] Drone hangars: Multiple drone hangars are set up within the test area to store and charge drones, ensuring their continuous mission performance in outdoor environments. Under remote control, drones intelligently take off from the hangars, enabling remote sensing calculation and monitoring of vegetation cover in the desert grassland.

[0149] Remote drone command: Operators can control the take-off and landing of drones through remote devices (such as a joystick) and view real-time flight processes and monitoring data to ensure the successful completion of missions.

[0150] Data transmission and processing: Image data captured by the drone is transmitted back to the data, communication, and command center via wireless communication. Through this center, the drone can be remotely commanded and its data analyzed and processed.

[0151] Data storage and management: The collected data is stored in a database for long-term tracking and analysis of changes in the growth status of desert steppe vegetation.

[0152] Satellite observation: By acquiring overall regional data through satellites and combining it with UAV observation data, macro-scale monitoring and positioning support for desert grassland vegetation can be achieved.

[0153] Desert steppe vegetation dynamics monitoring: Through the capture and collection of multispectral data by drones, the vegetation coverage and growth status of desert steppe can be remotely monitored, analyzed and evaluated.

[0154] This specific embodiment is merely an explanation of the present invention and is not intended to limit the present invention. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but as long as they are within the scope of the claims of the present invention, they are protected by patent law.

Claims

1. A UAV remote sensing monitoring device for measuring vegetation cover in desert steppes, characterized in that: It includes drones, multispectral cameras, and drone hangars; the multispectral cameras are mounted on drones and are used to collect multispectral remote sensing data of desert grasslands; while the drone hangars are intelligent drone parking base stations that support automatic take-off and landing, parking, energy replenishment, data relay, environmental protection, and remote control functions for drones. The monitoring method of the monitoring equipment includes the following: S1: Deploy UAV remote sensing monitoring equipment in the desert steppe test area; S2: Use a drone equipped with a multispectral camera to acquire multispectral images of the desert steppe test area at a preset flight altitude; S3: Preprocess the acquired multispectral images; S4: Divide the preprocessed multispectral image into several sub-blocks, and use an autoencoder plus KMeans clustering algorithm to perform unsupervised classification of vegetation and soil in each sub-block, thereby obtaining the number of vegetation pixels and soil pixels in each sub-block. S5: Calculate the vegetation cover and multispectral reflectance data for each sub-block based on the number of vegetation pixels and soil pixels in each sub-block; S6: Based on the vegetation cover and multispectral reflectance data of each sub-block, construct a new improved vegetation index; S7: Utilize the improved linear relationship between vegetation index and vegetation coverage to achieve high-precision estimation of vegetation coverage in desert steppe; In step S6, the process of constructing a new improved vegetation index is as follows: on the calculation framework of the vegetation index NDVI, the reflectance of the blue light band and the green light band is introduced and combined with the near-infrared and red light bands used in the NDVI calculation to form a new improved vegetation index. The calculation formula is as follows: R, G, B, and NIR represent the reflectance of red, green, blue, and near-infrared bands, respectively. This improved vegetation index is used to achieve high-precision measurement of vegetation coverage in desert steppes.

2. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: The drone hangar includes an external structure, a power supply and charging module, a communication and data module, a drone docking and take-off and landing module, and an environmental monitoring and safety module. The external structure is composed of a protective shell made of high-strength composite materials or metal, which has high temperature resistance, UV protection, and wind and sand protection properties, and has a built-in active temperature control component. The temperature control component is a small fan, a TEC cooling chip, or a phase change material, used to ensure that the drone battery is not damaged under extreme temperature differences. A solar panel is arranged on the top of the external structure to power the drone hangar and the drone battery. The power and charging module includes a fast charging interface to support automatic docking and charging of drone batteries; it has an internal battery compartment with multiple spare batteries pre-installed and is equipped with a robotic arm or automatic battery swapping mechanism; in addition, the power and charging module is also equipped with an intelligent power management unit for real-time monitoring of battery status and dynamic allocation of charging priorities. The communication and data module includes 4G / 5G or satellite communication components for remote control and data backhaul; in addition, the communication and data module is also equipped with a GPS positioning component to ensure accurate positioning and support the networking deployment of multiple drone hangars in the future. The UAV docking and take-off / landing module is equipped with a sliding rail or lifting platform, and combined with a communication and data module, it supports the automatic opening and closing of the hangar cover to guide the UAV to take off and land safely and dock precisely. The UAV docking and take-off / landing module also has a sand and dust-proof sealing function and is equipped with infrared, ultrasonic or RTK guidance and positioning components. The environmental monitoring and safety module includes temperature sensors, humidity sensors, wind speed sensors, and dust concentration sensors to acquire environmental data and assess the flight safety of the drone. Simultaneously, the module is equipped with an electronic lock and remote alarm components to protect the drone hangar, the drone itself, and the multispectral camera. Furthermore, the module includes fans, filters, or vibration devices to periodically clean dust from the air inlets and solar panels.

3. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 2, characterized in that: The UAV hangars are evenly distributed within the desert grassland test area, with a total of 5 hangars. The locations are determined by selecting representative sample plots based on the pixel resolution characteristics of the remote sensing images, supporting the verification and correction of remote sensing data at different resolutions of 30m, 250m, 500m, and 1000m.

4. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 2, characterized in that: The equipment also includes a ground spectrometer, meteorological sensors, and vegetation phenology cameras, all located near the hangar in the test area. These are used to collect surface reflectance, micro-meteorological data, and vegetation phenology information to assist in monitoring the vegetation growth process and estimating vegetation coverage.

5. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 2, characterized in that: By establishing a multi-level verification and integrated application chain from ground observation to UAV network observation to satellite remote sensing observation, we can achieve accurate monitoring of vegetation coverage and growth status in large-scale desert steppes, and support quantitative correction of remote sensing data with different spatial resolutions of 30m, 250m, 500m and 1000m, thus promoting the integrated application of multi-source data from air, space and ground in desert steppe areas.

6. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S2, the drone is a DJI MK300 model, and the multispectral camera it carries is a YUSENSEMS600Pro. This multispectral camera covers blue light, green light, red light, red edge 1, red edge 2 and near-infrared typical vegetation remote sensing detection bands, with center wavelengths of 450 nm, 555 nm, 660 nm, 720 nm, 760 nm and 840 nm, respectively.

7. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S2, the UAV is set to fly at a height of 15m, with a lateral overlap rate of 70%, a lateral overlap rate of 80%, and a flight speed of 1.5 m / s.

8. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S3, the preprocessing of the multispectral image includes data export and track stitching, radiometric correction, and geometric correction and registration; among which, During the data export and flight path stitching process, the original images are exported from the multispectral camera on the UAV, and multiple original images acquired from the same flight mission are stitched together to generate a complete seamless stitched image of the test area. During radiometric correction, a whiteboard attached to the camera is used to correct reflectivity in order to eliminate the effects of changes in light intensity and to remove noise signals generated by the sensor itself. During the geometric correction and registration process, distortion, viewpoint shift and terrain effects are corrected to achieve strict spatial alignment of images in different bands, avoiding misalignment in subsequent calculations. At the same time, RTK or ground control points are used to geolocate the images, eliminating deformation caused by terrain undulations or camera tilt, thereby giving the images map attributes.

9. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S4, the preprocessed multispectral image is divided into 10 blocks along both the row and column directions, resulting in a total of 100 sub-blocks. Unsupervised classification of vegetation and soil is then performed on each sub-block to generate a classification map of vegetation and soil for each sub-block. The process includes: Effective pixels are filtered out from the preprocessed multispectral images, and all pixels with band values ​​less than zero are removed. Median filling and standardization are performed on effective pixels. Greenness features are introduced and standardized. The standardized six-dimensional spectral features and greenness features are combined as seven-dimensional input variables for the autoencoder plus KMeans clustering algorithm. Latent space features are obtained by performing nonlinear dimensionality reduction and deep feature extraction on seven-dimensional input variables using an autoencoder. The latent space features and weighted greenness features are input into the KMeans clustering algorithm. The vegetation and soil categories are determined based on the median greenness of each cluster, and a classification map of vegetation and soil for each sub-block is generated.

10. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S5, the process for calculating the vegetation cover of each sub-block is as follows: Based on the classification results of vegetation and soil pixels in each sub-block, vegetation cover is defined as the proportion of green vegetation pixels in each sub-block to the total number of pixels in that sub-block. The calculation formula is as follows: Among them, FVC t V represents the vegetation cover of subblock t. t N represents the number of pixels within sub-block t that are identified as green vegetation. t This represents the total number of pixels within the sub-block, which is the sum of vegetation pixels and soil pixels.

11. The UAV remote sensing monitoring device for measuring vegetation cover in desert steppe according to claim 1, characterized in that: In step S5, the process for calculating the multispectral reflectance data of each sub-block is as follows: the preprocessed multispectral image is subjected to the same block processing, and the average value of effective pixels with pixel values ​​greater than zero is statistically calculated for each band in each sub-block, which is used as the average reflectance of each band in each sub-block.

Images