Urban green land vegetation extraction method based on unmanned aerial vehicle multispectral remote sensing image

Abstract

The invention discloses an urban green land vegetation extraction method based on an unmanned aerial vehicle multispectral remote sensing image. Comprising the following steps: 1) determining an image acquisition area, and performing unmanned aerial vehicle path planning; 2) performing data acquisition through an unmanned aerial vehicle to obtain a high-resolution multispectral image; (3) splicing the acquired images through measurement software Pix4D mapper 4.4 to generate a digital orthoimage; and 4) combining the orthoimages of the spectral bands by using ENVI 6.0 software to obtain images of five bands including blue, green, red, red edge and near-infrared. According to the method, a background area which does not meet the vegetation spectral reflectivity characteristic is effectively inhibited, and the abnormal phenomenon that the reflectivity of a red-edge wave band is lower than that of a red wave band or higher than that of a near-infrared wave band is also solved, so that noise interference is effectively reduced; meanwhile, no obvious intersection phenomenon exists between the gray statistical characteristic values of the vegetation and the non-vegetation characteristics, and the accuracy of green land vegetation collection is improved.

Description

Technical Field

[0001] The invention relates to a method for extracting urban green space vegetation based on multi-spectral remote sensing images of unmanned aerial vehicles. Background Art

[0002] With the rapid advancement of global urbanization, urban green space is becoming increasingly important to people. Therefore, understanding the coverage, distribution and changing trends of urban green space vegetation is crucial for urban planning, environmental protection and urban development. Urban green space is an important part of the urban ecosystem, and has many ecological benefits such as improving air quality, alleviating urban heat island effect, and protecting biodiversity. Urban green space also provides many social benefits, such as providing entertainment and social space, reducing mental stress, and ensuring the health of urban residents. However, the process of urbanization is accompanied by many negative effects, such as crowded habitats and water pollution, among which the problem of urban green space coverage is particularly prominent.

[0003] Remote sensing technology provides strong support for the efficient and rapid determination of the spatial distribution and structural characteristics of vegetation. Remote sensing satellites have collected a large amount of surface data with different spatial, spectral and temporal resolutions, becoming the main source for the extraction of surface vegetation and its change analysis in recent decades. Many studies on urban green space vegetation extraction using satellite remote sensing use images with spatial resolutions ranging from 1 meter to 30 meters. This is prone to mixed pixel problems (the phenomenon of containing two or more different types of land objects in one pixel) in the face of the complex interweaving and fragmentation of urban ecosystems, making it impossible to accurately distinguish between different vegetation types and other land object coverage areas. In addition, satellite images are easily affected by the atmosphere, clouds and rainfall, which reduces the accuracy of remote sensing images.

[0004] In recent years, UAV remote sensing technology has been widely used in urban mapping, traffic management, disaster management and precision agriculture. Compared with traditional remote sensing technology, UAV remote sensing has unique advantages such as high resolution (e.g., reaching centimeter level in space), strong timeliness (e.g., UAV can transmit the collected data back to the ground station in real time during flight), and flexible and convenient data acquisition (e.g., preparing and taking off quickly in a short time, and providing fast data acquisition capabilities for emergency or temporary needs). UAV can operate at a low flight altitude (e.g., 100 meters) close to the target, obtain centimeter-level extremely high-resolution images, effectively reduce the problem of mixed pixels of complex urban coverage types, and gradually become an important data source for urban green space monitoring and vegetation information extraction. UAV high-resolution images provide rich ground features, which are conducive to more refined green space vegetation extraction. However, due to the complexity of the urban environment, there are often more interferences, that is, hidden noises, in high-resolution images, which brings challenges to the accurate extraction of urban green space.

[0005] At the same time, a large number of methods for extracting and classifying urban green space vegetation based on vegetation index have been studied and developed. The main advantages of these methods are high efficiency and easy implementation. The vegetation index method is based on the difference in spectral reflectance of different objects. It enhances vegetation information through band combination calculation, while suppressing other object information, thereby realizing the extraction of target vegetation information. At present, the vegetation index that uses visible light and near-infrared bands to evaluate the spectral characteristics of plant chlorophyll has been used in the field of remote sensing to evaluate vegetation distribution and growth characteristics. Most studies based on UAV remote sensing images only use images of the three RGB channels. Many scholars have proposed a series of vegetation indices based on the visible light band using the reflectance characteristics of green vegetation in the visible light band. However, since the RGB vegetation index can only determine the type of ground cover based on color, it is difficult to distinguish the characteristics of similar vegetation colors, resulting in less than ideal recognition accuracy.

[0006] Therefore, a method for extracting urban green space vegetation based on UAV multispectral remote sensing images was proposed. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method in which the background area that does not meet the spectral reflectance characteristics of vegetation is effectively suppressed, and the abnormal phenomenon that the reflectance of the red edge band is lower than that of the red band or higher than that of the near-infrared band is also solved, which effectively reduces noise interference; at the same time, there is no obvious crossover phenomenon between the grayscale statistical characteristic values ​​of vegetation and non-vegetation features, which provides favorable assistance for determining a suitable threshold to complete the extraction of urban green space vegetation. The method is practical and widely used for extracting urban green space vegetation based on unmanned aerial vehicle multispectral remote sensing images.

[0008] To solve the above problems, the present invention adopts the following technical solutions:

[0009] A method for extracting urban green space vegetation based on multispectral remote sensing images from unmanned aerial vehicles includes the following steps:

[0010] 1) Determine the image collection area and plan the drone path;

[0011] 2) Collect data through drones to obtain high-resolution multispectral images;

[0012] 3) The collected images were stitched using the measurement software Pix4D mapper 4.4 to generate digital orthophotos;

[0013] 4) ENVI 6.0 software was used to combine the orthophotos of each spectral band to obtain images of five bands: blue, green, red, red edge, and near infrared.

[0014] Preferably, the drone operation process in step 2) is automatically completed through a route flight mode, and an equidistant interval photo taking mode is adopted.

[0015] Preferably, before stitching the images in step 3), calibration is performed using a radiation calibration panel taken before the flight to eliminate or correct systematic and random radiation distortion or aberration problems caused by external factors such as the nonlinear response of the sensor itself, atmospheric refraction, and cloud cover.

[0016] Preferably, in order to effectively distinguish the characteristics of green vegetation areas and non-vegetation areas in high-resolution multispectral images of drones, the spectral reflectance curves of typical landforms are analyzed by manually selecting the area of ​​interest on the drone multispectral orthophotos, and the vegetation types of trees, shrubs and grasslands in the area, as well as other types of landforms such as non-vegetation, are selected to calculate the average value of the relative reflectance of the spectral bands of various landforms.

[0017] Preferably, in the process of combining image bands, based on the influence of shadow coverage and special areas on reflectivity when extracting vegetation, the reflectivity of green vegetation type in the red band is low, and the reflectivity in the red edge band is significantly higher; there is a large difference between the average reflectivity of the red edge and red bands of vegetation, while the reflectivity difference of non-vegetation types between these bands is small; the basis for constructing the red edge vegetation index is calculated as follows:

[0018]

[0019] where ρ RE represents the red edge band reflectivity, ρ Red Represents the red band reflectance.

[0020] Preferably, based on the fact that vegetation has a lower reflectivity in the red band, a higher reflectivity in the near-infrared band, and a medium level between the two in the red edge band, an experimental attempt is made on the spectral reflectivity curve characteristics of green vegetation to obtain a combination of ideal vegetation indices, which are mathematically expressed as follows:

[0021]

[0022] Preferably, multiple vegetation indices are constructed for different scenes using the reflectance characteristics of visible light and multispectral bands; 5 existing main vegetation indices are selected to extract and compare the accuracy of urban green spaces; they are the Normalized Difference Vegetation Index NDVI, the Red Edge / Red Light Vegetation Index NDVIrededge, the Near Infrared / Red Edge Vegetation Index NDRE, the Shadow Vegetation Index SVI, and the Improved Green / Red Vegetation Index MGRVI. In order to better compare the reliability of these vegetation index maps in extracting vegetation information, the statistics of typical objects in each vegetation index are calculated by using the same ROI area in the calculated vegetation index map, including the mean and standard deviation.

[0023] Preferably, with the goal of optimizing the extraction of vegetation information and suppressing the background, after multiple experimental analyses and comparisons, the optimal segmentation threshold for each vegetation index is obtained, and then the image creation is completed.

[0024] The beneficial effects of the present invention are:

[0025] 1. The expression of vegetation index method NDVI / NDVIrededge not only effectively suppresses the background area that does not meet the vegetation spectral reflectance characteristics, but also solves the abnormal phenomenon that the red edge band reflectance is lower than the red band or higher than the near-infrared band reflectance, which effectively reduces noise interference. At the same time, there is no obvious crossover phenomenon between the grayscale statistical characteristic values ​​of vegetation and non-vegetation features, which provides favorable help for determining the appropriate threshold to complete the extraction of urban green space vegetation.

[0026] 2. The vegetation index method NDVI / NDVIrededge performs well in urban areas with complex land features, effectively improving the difficulty of extracting vegetation information in complex urban scenes. The over-extraction and under-extraction situations are significantly reduced. The accuracy rate and Kappa coefficient of vegetation information extraction both reach the highest 98.6% and 0.97, respectively, which are 1.6% and 4% higher than NDVIrededge, which ranks second in accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below, but this does not limit the protection scope of the present invention.

[0028] Figure 1 A schematic diagram of a UAV digital orthophoto image of the research area of ​​the present invention;

[0029] Figure 2 It is a schematic diagram of the relative reflectance of typical ground objects in different bands of the present invention;

[0030] Figure 3 It is a schematic diagram of the relative amounts of the reflection coefficients of typical ground objects in five wavebands of the present invention;

[0031] Figure 4 It is a schematic diagram of the red edge vegetation index and the adjusted vegetation index of the present invention;

[0032] Figure 5 This is a schematic diagram of vegetation extracted by the vegetation index of the present invention. DETAILED DESCRIPTION

[0033] Take a certain area as the target area of ​​this study, such as Figure 1As shown in the figure, the area is rich in landform types. Vegetation areas include trees, shrubs and lawns, while non-vegetation areas include buildings, roads, water bodies and various sports venues. The buildings on campus are mainly concentrated in the central and western areas, including teaching buildings, libraries, dormitories and administrative office buildings. The vegetation coverage rate on campus is high. Trees are mainly distributed on both sides of the road and around buildings, forming a good shading effect. Shrubs and lawns are widely distributed in open areas and around sports venues, forming a multi-level green landscape. The diverse landform types and good vegetation coverage make the ecological environment on campus an ideal research area for this study.

[0034] The "DJI M300 RTK" drone was used for data collection. The aircraft has a built-in RTK module (Real-Time Kinematic), which is a high-precision positioning technology based on carrier phase observations, ensuring high positioning accuracy. The MicaSense RedEdge MX camera is mounted on the M300 RTK drone to collect high-resolution multispectral images. Images of five bands, including blue, green, red, red edge, and near-infrared, are obtained. The resolution at an altitude of 120 meters is about 8 cm. The central wavelength and main parameters are shown in Tables 1 and 2, where the half-peak full bandwidth is the wavelength range corresponding to the peak height in the spectral curve dropping to half.

[0035]

[0036] Table 1. Spectral bands of MicaSense RedEdge MX;

[0037]

[0038] Table 2Micasense RedEdge MX specifications;

[0039] The entire flight operation process of the UAV is automatically completed through the route flight mode, and the equidistant interval photo mode is adopted to ensure the efficiency of image acquisition. The specific flight parameters are shown in Table 3.

[0040]

[0041] Table 3: Basic parameters of the flight mission;

[0042] The collected images were stitched using professional aerial photogrammetry software Pix4D mapper 4.4 to generate digital orthophotos, namely DOM (Digital Orthophoto Map). At the same time, the radiation calibration panel taken before the flight was used for calibration. Radiation calibration is to convert the original digital value (DN value) recorded by the sensor into a quantity with physical meaning, such as radiation brightness or ground reflectivity; then radiation correction is performed, which is to eliminate or correct the systematic and random radiation distortion or aberration caused by external factors such as the nonlinear response of the sensor itself, atmospheric refraction, and cloud cover.

[0043] ENVI 6.0 software was used to combine the orthophotos of each spectral band to obtain various vegetation index images, and the image GSD (Ground Sample Distance) resolution was set to 8 cm / pixel.

[0044] In the implementation process, in order to effectively distinguish the characteristics of green vegetation areas and non-vegetation areas in the high-resolution multispectral images of drones, it is necessary to analyze the reflectance of different bands; by manually selecting the region of interest (ROI) on the multispectral orthophoto of drones, the spectral reflectance curves of typical objects were analyzed. These include 11 types of vegetation types such as trees, shrubs and grasslands, as well as other types such as non-vegetation. The average value of the relative amount of reflectance of the spectral bands of various objects was calculated to obtain the reflectance curves of 5 bands.

[0045] from Figure 2 From the reflectivity curve, it can be seen that green vegetation (color marked as: red, red dots, and red virtual) has the highest reflectivity in the near-infrared band. The reflectivity in the near-infrared band and the red edge band is much greater than that in the visible light band (blue, green, and red bands), while the reflectivity in the red and blue bands is absorbed, presenting a lower level.

[0046] During the image acquisition process, some problems will be encountered: Usually, the reflectivity of typical vegetation areas in the near-infrared and red bands is quite different, so it is easy to distinguish. However, for high-resolution drone images taken near the ground, there is an obvious shadow interference problem, which is caused by dense objects in urban areas and tall buildings or trees blocking the sunlight. From the spectral reflectivity curve, the reflectivity of the shadow-covered area is lower than that of other object areas, but it has similar spectral characteristics with vegetation, so it is difficult to distinguish vegetation. Due to the influence of shadow occlusion, the reflectivity of green vegetation under the shadow is reduced, and the decrease in reflectivity in the near-infrared band and the red edge band is particularly significant.

[0047] On the other hand, it was also found that the blue badminton court and the roof had a relatively high reflectivity in the near-infrared band (ground feature number 9) in the study area, so vegetation could not be extracted in the near-infrared band. Therefore, for the extraction of green vegetation from high-resolution drone images in complex urban scenes, a complete solution is needed to address the above issues in order to facilitate image reconstruction.

[0048] To solve the above problems, an effective UAV multispectral image vegetation index is established to conduct a detailed analysis of the reflectance differences of typical objects in five bands, including the mean value, standard deviation, etc. The statistical results are shown in Table 4.

[0049]

[0050] Table 4 Relative reflectance of typical objects in 5 bands;

[0051] Studies have shown that the red edge band has a high sensitivity to vegetation, which reflects the vegetation growth status. The red edge area is defined as between 707-737nm. From Table 4 (and Figure 4 ) It can be seen that the green vegetation type has a lower reflectivity in the red band and a significantly higher reflectivity in the red edge band. There is a large difference between the average reflectivity of the red edge and red bands of vegetation, while the reflectivity difference between these bands for non-vegetation types is small. Therefore, the normalized difference between the red edge band and the red band has been evaluated and applied to vegetation analysis in some studies. This feature is used as the basis for constructing a new vegetation index, which is calculated as follows:

[0052]

[0053] where ρ RE represents the red edge band reflectivity, ρ Red Represents the red band reflectance.

[0054] At the same time, due to the three-dimensional structure of urban areas, the interference of shadows in urban environments is greater than that in natural environments. It is important to consider the impact of shadows on the reflectivity of vegetation objects. Compared with other bands, the decrease in reflectivity of shadows in the near-infrared band is extremely obvious.

[0055] Therefore, from the statistical results in Table 4, it can be seen that vegetation has a lower reflectivity in the red band, a higher reflectivity in the near-infrared band, and a medium level between the two in the red edge band. Referring to this rule, an experimental attempt was made on the spectral reflectivity curve characteristics of green vegetation to obtain a combination of ideal vegetation indices, which is expressed as follows (2):

[0056]

[0057] During the implementation process, in order to reflect the actual effect on the multi-spectral RGB band, the above vegetation index is used and verified in the following way.

[0058] Using the reflectance characteristics of visible light and multispectral bands, a variety of vegetation indices were constructed for different scenes. As shown in Table 5, five existing main vegetation indices were selected to extract urban green space and compare their accuracy. They are the Normalized Difference Vegetation Index NDVI, the Red Edge / Red Light Vegetation Index NDVIrededge, the Near Infrared / Red Edge Vegetation Index NDRE, the Shadow Vegetation Index SVI, and the Improved Green / Red Vegetation Index MGRVI. These vegetation indices have been widely used to extract vegetation in previous studies.

[0059]

[0060] Table 5 has a comparison of the main vegetation indices;

[0061] In order to illustrate the effectiveness and universal applicability of the vegetation index of this algorithm, three highly representative areas with different surface characteristics in the study area (badminton court area, shadow area, and water area) were selected for comparison and explanation. Figure 5 It contains a variety of typical landform types, for example, large areas of tree cover such as Figure 5 (a) Column, scattered bushes, withered grass, blue badminton court and red basketball court, buildings of various colors and styles, shadow interference caused by building occlusion Figure 5 Column (b) shows the dark green pond water surface. Figure 5 (c) Column.

[0062] According to the vegetation index calculation formula listed in Table 5, the inter-band calculation is performed on the high-resolution multispectral remote sensing images of the UAV in the area to obtain the vegetation index extraction formula. Figure 5 shown.

[0063] from Figure 5It can be seen that the vegetation area is relatively bright, and the non-vegetation area is darker. For the vegetation area, the tone of the tree area is the brightest, and the brightness of the shrub and grass area is relatively reduced. From the calculation result diagram, it can be seen that each vegetation index can well display the vegetation coverage area. However, under complex scene conditions, the results of the existing vegetation index have produced varying degrees of confusion. In the calculation of NDVI, NDVIrededge, NDRE, SVI, and MGRVI, the outer circle or the entire range of the badminton court on the blue ground presents a higher brightness and cannot be distinguished from the vegetation area; NDVI, NDRE, NDVIrededge, and MGRVI mistakenly overestimate the shadow area of ​​the building. Because the reflectivity of the vegetation area blocked by the shadow is significantly reduced, the corresponding area in the SVI vegetation index map shows a lower grayscale, which is difficult to effectively distinguish from non-vegetation. Except for MGRVI, other vegetation indices can distinguish water areas.

[0064] In order to better compare the reliability of these vegetation index maps in extracting vegetation information, the statistics of typical objects in each vegetation index, including the mean and standard deviation, were calculated by using the same ROI (Region of Interest) area in the calculated vegetation index map. The statistical characteristics of each vegetation index are shown in Table 6.

[0065]

[0066] Table 6 Statistical characteristics of vegetation index

[0067] As can be seen from Table 6, SVI effectively suppressed shadows, but at the same time caused the vegetation under the shadows to have a lower index value than some background types. MGRVI had a large standard deviation for shrubs and grassland types, which would confuse them with the background, and had a poor suppression effect on various non-vegetation types in the experimental area, and could not effectively extract vegetation information. From the statistical results of this combined method, no cross-overlap phenomenon was found between the grayscale statistical feature values ​​of vegetation types and non-vegetation types.

[0068] In order to further evaluate each vegetation index qualitatively and quantitatively, and extract the performance of urban green space vegetation from the high-resolution multispectral image of the UAV, a threshold is needed for the vegetation index to generate the urban green space vegetation image as the result of vegetation information. With the goal of optimizing the extraction of vegetation information and suppressing the background, after multiple experimental analysis and comparison, the optimal segmentation threshold was selected for each vegetation index, as shown in Table 7.

[0069]

[0070] Table 7 Vegetation index thresholds.

[0071] The proposed combined vegetation index has taken advantage of the fact that the actual spatial distribution of vegetation pixels is closest to the UAV remote sensing image. The background area except the vegetation is excellently suppressed, which not only makes the boundaries of the vegetation area more complete, but also effectively reduces the void and noise phenomena.

[0072] In order to verify the accuracy of each vegetation index, 500 verification points were randomly selected in the experimental area. After manual visual interpretation, it was determined that there were 204 verification points of green vegetation types including trees, shrubs, grass and shaded vegetation, and the remaining 296 verification points were all background types. Through the error matrix, the classification accuracy of vegetation and non-vegetation types is easy to obtain, and the results are shown in Table 8.

[0073]

[0074]

[0075] Table 8 Vegetation extraction accuracy estimation

[0076] After determining the optimal thresholds of each vegetation index, an urban green space vegetation image can be created. Figure 5 (a) In the vegetation information extraction map of NDVI, EVI, NDRE and SVI indexes, the badminton court on the blue ground is misclassified as a vegetation type, resulting in a large error. Except for MGRVI, which fails to exclude the characteristics of water bodies from vegetation types, other vegetation indices effectively suppress water body information, see Figure 5 (c); NDVI, NDRE, NDVIrededge, and MGRVI failed to successfully avoid the interference of shadows, such as Figure 5 (b) Figure 5 (c).

[0077] At the same time, it can be seen from Table 8 that NDVI incorrectly classifies background pixels such as shadows and blue objects as vegetation, which increases the extraction error of vegetation information, but it can better give vegetation a higher index value, and the overall accuracy still reaches 90.2%. From the perspective of vegetation extraction accuracy, SVI can extract vegetation information more accurately, but it cannot effectively extract vegetation obscured by shadows, which reduces the overall accuracy. NDRE failed to effectively suppress background information and mistakenly confused many background types with vegetation information, resulting in a relatively low extraction accuracy. NDVIrededge performed well overall, and the vegetation area extraction was relatively comprehensive, but it was still confused by shadow interference, which increased the extraction error.

[0078] The advantages of the present invention are as follows: the proposed NDVI / NDVIrededge extracts vegetation information with higher completeness, and has better recognition effect on lawns with wilting than NDVIrededge, stronger suppression ability for shadow interference, fewer holes and noise phenomena, and higher extraction accuracy in both vegetation and non-vegetation areas, reaching a maximum overall accuracy of 98.6%. The kappa coefficient can be used to evaluate the consistency and reliability of the vegetation information extraction accuracy of each vegetation index with the actual situation. For NDVI, NDRE, SVI, and MGRVI, the kappa coefficient is between 0.7 and 0.8, and the accuracy reaches a high consistency level. NDVIrededge reaches a kappa coefficient of 0.93, indicating that the accuracy of vegetation information extraction reaches a high level. NDVI / NDVIrededge achieves the highest kappa coefficient of 0.97, proving the superiority of the vegetation index proposed in this paper in terms of vegetation information extraction accuracy. Overall, the accuracy of the method in this paper for extracting urban green space from high-resolution images is better than that of other mainstream vegetation indices compared, providing a new and reliable way for extracting vegetation information in complex urban areas.

[0079] At the same time, the spectral characteristics of typical objects in the study area were analyzed in detail, and it was found that vegetation has similar characteristics in the red to near-infrared bands, with low reflectivity in the red band, high reflectivity in the near-infrared band, and medium reflectivity in the red edge band. Therefore, a new joint vegetation index method was constructed. Since the blue coated roof and the blue badminton court have similar reflectivity differences with vegetation in the near-infrared band and the red band, NDVI cannot avoid such confusion, but the red edge band can be effectively distinguished because the reflectivity of such areas in the red edge band is extremely low. Shadows have a spectral reflectivity curve similar to that of vegetation, which is why most indices are difficult to eliminate interference. However, the reflectivity of the shadow area is low, and the shadow area is effectively suppressed by multiplying it with the near-infrared band. Therefore, SVI effectively suppresses the shadow, but at the same time makes the vegetation under the shadow unrecognizable. The vegetation under the shadow has a higher reflectivity in the near-infrared band than the non-vegetation type. The index value of vegetation under shadow occlusion calculated by MREVI is still significantly higher than that of the non-vegetation area. Therefore, the vegetation area under the shadow can still be effectively distinguished from the background.

[0080] The expression of the combined vegetation index method NDVI / NDVIrededge not only effectively suppresses the background area that does not meet the vegetation spectral reflectance characteristics, but also solves the abnormal phenomenon that the red edge band reflectance is lower than the red band or higher than the near-infrared band reflectance, which effectively reduces noise interference. At the same time, there is no obvious crossover between the grayscale statistical characteristic values ​​of vegetation and non-vegetation features, which provides favorable help for determining a suitable threshold to complete the extraction of urban green space vegetation.

[0081] NDVI / NDVIrededge performed well in urban areas with complex terrain features, effectively improving the difficulty of extracting vegetation information in complex urban scenes. The over-extraction and under-extraction situations were significantly reduced. The accuracy rate and Kappa coefficient of vegetation information extraction both reached the highest 98.6% and 0.97, respectively, which were 1.6% and 4% higher than NDVIrededge, which ranked second in accuracy.

[0082] The calculation results of the vegetation index largely determine the ability to extract vegetation information, but the choice of segmentation threshold is also crucial; in addition, the image data analyzed in this study is limited in scope, and the accuracy of vegetation information extraction in other regions based on the newly constructed NDVI / NDVIrededge needs further research and demonstration.

[0083] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that are not conceived through creative work should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope defined in the claims.

Claims

1. A method for extracting urban green space vegetation based on multispectral remote sensing images from unmanned aerial vehicles; it is characterized by: The following steps are involved: 1) Determine the image collection area and plan the drone path; 2) Collect data through drones to obtain high-resolution multispectral images; 3) The collected images were stitched using the measurement software Pix4D mapper 4.4 to generate digital orthophotos; 4) ENVI 6.0 software was used to combine the orthophotos of each spectral band to obtain images of five bands: blue, green, red, red edge, and near infrared.

2. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1 is characterized in that: In step 2), the drone operation process is automatically completed through the route flight mode and the equidistant interval photo taking mode is adopted.

3. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1 is characterized in that: In step 3), before stitching the images, the radiation calibration panel taken before the flight is used for calibration to eliminate or correct the systematic and random radiation distortion or distortion caused by external factors such as the nonlinear response of the sensor itself, atmospheric refraction, and cloud cover.

4. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: When calibrating the radiation calibration panel in step 3), in order to effectively distinguish the characteristics of green vegetation areas and non-vegetation areas in the high-resolution multispectral image of the drone, the spectral reflectance curves of typical objects are analyzed by manually selecting the area of ​​interest on the multispectral orthophoto of the drone. At the same time, the vegetation types of trees, shrubs and grasslands in the area, as well as other types of objects such as non-vegetation, are selected, and the average value of the relative reflectance of the spectral bands of various objects is calculated to obtain the reflectance of the five bands.

5. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: Step 4) In the process of image combination, based on the influence of shadow coverage and special areas on reflectivity when extracting vegetation, the reflectivity of green vegetation type in the red band is low, and the reflectivity in the red edge band is significantly higher; there is a large difference between the average reflectivity of the red edge and red bands of vegetation, while the reflectivity difference between non-vegetation types in these bands is small; a new vegetation index combination is constructed, which is calculated as follows: where ρ RE represents the red edge band reflectivity, ρ Red Represents the red band reflectance.

6. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: Step 4) Based on the fact that vegetation has a lower reflectivity in the red band, a higher reflectivity in the near-infrared band, and a medium level between the two in the red edge band, an experimental attempt was made on the spectral reflectivity curve characteristics of green vegetation to obtain an ideal vegetation index combination, which is expressed as the following formula (2):

7. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: Using the reflectance characteristics of visible light and multispectral bands, a variety of vegetation indices are constructed for different scenes. Five existing main vegetation indices are selected to extract urban green spaces and compare their accuracy. The five vegetation indices are the Normalized Difference Vegetation Index (NDVI), the Red Edge / Red Light Vegetation Index (NDVIrededge), the Near Infrared / Red Edge Vegetation Index (NDRE), the Shadow Vegetation Index (SVI) and the Improved Green / Red Vegetation Index (MGRVI). At the same time, in order to verify the reliability of these vegetation index maps in extracting vegetation information, the same ROI area method is used in the vegetation index maps to calculate and count the information of typical objects in each vegetation index, including the mean and standard deviation.

8. The method for extracting urban green space vegetation based on multispectral remote sensing images of unmanned aerial vehicles according to claim 1, characterized in that: With the goal of optimizing the extraction of vegetation information and suppressing the background, after multiple experimental analyses and comparisons, the optimal segmentation threshold for each vegetation index was obtained, and then the image creation was completed.

Images