A decision tree extraction method for bare sand land based on high-resolution satellite remote sensing
Through the high-segment satellite remote sensing method of exposed sand decision tree, the DBSI and NDVI indexes are used to solve the problems of low accuracy and high cost of bare sand recognition in the existing technology, and the accurate identification of exposed sand and large-scale areas are achieved. Monitoring, suitable for high-score series satellite data.
Patent Information
- Application Number
- CN202310896583.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-07-21
- Publication Date
- 2024-11-12
- Estimated Expiration
- 2043-07-21
AI Technical Summary
The existing remote sensing recognition method for exposed sand is based on MODIS and Landsat satellite data, with low accuracy and difficulty in distinguishing between buildings, exposed land and sand barriers. Machine learning methods such as random forests rely on subjective experience, and the training area has a large workload and large results are different. , cannot meet the needs of high-score series satellite data.
Using the method of extracting the bare sand decision tree based on high-segment satellite remote sensing, the decision tree algorithm is established to determine the NDVI by calculating the bare sand index DBSI and the normalized vegetation index NDVI using the reflectivity data of the 8-band of high-segment satellite. and DBSI thresholds to identify exposed sand, suitable for different underfloor cover types.
Accurate identification and area estimation of exposed sand is achieved, monitoring accuracy is improved, suitable for remote sensing monitoring of large-scale and long-term changes, reducing costs, and suitable for areas with high spatial heterogeneity.
Smart Images

Figure CN116977873B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to a decision tree extraction method, in particular to a bare sand land decision tree extraction method based on high-resolution satellite remote sensing. Background Art
[0002] Due to factors such as climate change and human activities, land desertification has become more serious. Over the years, the ecological environment monitoring and management departments have mainly relied on ground manual measurements, low-resolution remote sensing inversion, or statistical methods based on land type, ecological vegetation, climate precipitation, etc. to monitor the long-term changes in the coverage area of desertification and bare sand. The calculation results obtained by different methods vary greatly, and compared with the field measurement results, the accuracy cannot meet the requirements of ecological sustainable development in the new era.
[0003] Traditionally, polar-orbiting meteorological satellites such as MODIS are used to monitor desertification. NDVI and EVI indexes are the most commonly used monitoring methods and are widely used in large-scale vegetation and desert monitoring. Kearney et al. used the characteristics of soil, such as the highest reflectivity in the mid-infrared spectrum and the easiest distinction from vegetation and water bodies in this spectrum, to construct the Normalized Difference Soil Index (NDSI) in the mid-infrared and near-infrared spectrum, which can also be used to extract information about bare sand. Zhao et al. used Landsat TM / ETM+ images to extract bare soil in the Yellow River Delta, and created the Normalized Difference Bareness Index (NDBaI) using thermal infrared and mid-infrared 5 bands. In addition, machine learning classification techniques represented by the random forest classification method are also widely used in the extraction of sand information.
[0004] At present, there is no research on the algorithm for extracting desert sand index from high-resolution satellite data in China. Most of the existing remote sensing identification indices for bare sand are based on MODIS and Landsat satellite data. The key band is the mid-infrared band, which is not suitable for domestic high-resolution satellites. Traditional identification methods based on NDVI or EVI have low accuracy and are difficult to distinguish between low vegetation areas such as buildings, bare land, and sand barriers. Although machine classification learning technologies such as random forests and support vector machines have good extraction effects, the workload of demarcating training areas is large and relies on subjective experience and judgment. The measurement results obtained by different training areas and different methods vary greatly. Summary of the invention
[0005] In order to solve the shortcomings of the above-mentioned technology, the present invention provides a bare sand decision tree extraction method based on high-resolution satellite remote sensing.
[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: a decision tree extraction method for exposed sand land based on high-resolution satellite remote sensing, including the following extraction process:
[0007] 1) Prepare high-resolution satellite remote sensing image data;
[0008] 2) Preprocess the acquired remote sensing image data;
[0009] 3) Calculate the bare sand index DBSI;
[0010] 4) Introduce NDVI to construct a decision tree algorithm and calculate the normalized difference vegetation index NDVI;
[0011] 5) At each remote sensing pixel, divide it into low vegetation area, vegetation or water body according to the NDVI calculated value;
[0012] 6) For each remote sensing pixel that meets the low vegetation zone, bare sand is extracted based on the DBSI calculation value.
[0013] Furthermore, in 1), the prepared high-resolution satellite remote sensing data covers the target monitoring area, including reflectance data of 8 bands, namely: B1-blue spectrum, B2-green spectrum, B3-red spectrum, B4-near infrared spectrum, B5-red edge spectrum 1, B6-red edge spectrum 2, B7-purple spectrum, and B8-yellow spectrum.
[0014] Furthermore, the wavelength of B1-blue spectrum is between 0.45-0.52um, the wavelength of B2-green spectrum is between 0.52-0.59um), the wavelength of B3-red spectrum is between 0.63-0.69um, the wavelength of B4-near infrared spectrum is between 0.77-0.89um, the wavelength of B5-red edge spectrum 1 is between 0.69-0.73um, the wavelength of B6-red edge spectrum 2 is between 0.77-0.89um, the wavelength of B7-purple spectrum is between 0.40-0.45um, and the wavelength of B8-yellow spectrum is between 0.59-0.63um.
[0015] Furthermore, in 2), the data preprocessing methods include but are not limited to orthorectification, radiation calibration, and atmospheric correction.
[0016] Further, in 3), the reflectivity Blue of the B1-blue spectrum segment B1 , B2-reflectivity of green spectrum GReen B2 、B7-Reflectivity of purple spectrum B7 , calculate the bare sand index DBSI, see formula 1:
[0017] DBSI=Blue B1 +Green B2 -Purple B7(1).
[0018] Further, in 4), from the reflectance Red of the B3 - red spectral band B3 and the reflectance NIR of the B4 - near - infrared spectral band B4 , the normalized difference vegetation index NDVI is calculated, as shown in Equation 2:
[0019]
[0020] Further, in 5), it is determined whether the value of NDVI satisfies the condition: 0 < NDVI < 0.2. If it is satisfied, it is a low - vegetation area; if not, it is vegetation or water body.
[0021] Further, in 6), on each remote - sensing pixel that satisfies the low - vegetation area, it is determined whether the value of DBSI satisfies the condition: DBSI > C = 0.15, where C = 0.15 is the empirical threshold of DBSI; if it is satisfied, it is bare sand; if not, it is building, sand barrier or bare land.
[0022] In view of the disadvantages of the existing method for identifying bare sand, such as low accuracy, inability to monitor large areas, high cost, and inability to be applied to areas with high spatial heterogeneity, the present invention proposes a method that takes the high - resolution satellite remote - sensing data (reflectance data of 8 bands of the WFV sensor) of remote - sensing target pixels as input, calculates the normalized difference vegetation index NDVI and the new bare - land sand - land index DBSI in the target pixels, and identifies the target pixels as bare - land sand - land through the decision - tree method of judging DBSI > 0.15 and 0 < NDVI < 2. The present invention innovatively proposes a new calculation method for the bare - land sand - land index DBSI, that is: bare - land sand - land index = blue spectral band + green spectral band - purple spectral band; for the set of decision - tree recognition technologies for bare - land sand - land based on high - resolution satellite remote - sensing images formed by the present invention, it can be applied to the extraction and area estimation of bare - land sand - land within pixels of different underlying - surface coverage types (including forests, grasslands, farmlands, wetlands, shrubs, and deserts, etc.), and realize the large - area and long - term change remote - sensing monitoring of severely desertified areas in China. BRIEF DESCRIPTION OF THE DRAWINGS
[0023] Figure 1 It is a schematic diagram of the overall technical process of the present invention.
[0024] Figure 2 It is a broken - line graph of the typical spectral characteristics of bare sand and related land types in the study area of the present invention.
[0025] Figure 3 It is the high - resolution No. 6 optical image on September 6, 2022 in the embodiment of the present invention.
[0026] Figure 4 It is the result map of the extracted bare - land sand - land coverage range in the embodiment of the present invention. Detailed implementation manners
[0027] The present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation manners.
[0028] The present invention provides a decision tree recognition method for bare sandy land based on high-resolution satellite remote sensing images. This method first proposes a new bare sandy land index DBSI (Difference Bare Sand Index), and when the bare land sandy land index DBSI of a remote sensing pixel in a low vegetation area (0 < NDVI < 0.2) is greater than 0.15, it can be determined that the pixel is bare sandy land.
[0029] As Figure 1 shown, it is the flowchart of the technical solution of the present invention, and the specific recognition method process is as follows:
[0030] 1) Prepare high-resolution satellite remote sensing image data;
[0031] The prepared data is the reflectance data of 8 bands of the high-resolution satellite WFV sensor covering the target monitoring area, which are: B1 - blue spectral band (0.45 - 0.52um), B2 - green spectral band (0.52 - 0.59um), B3 - red spectral band (0.63 - 0.69um), B4 - near-infrared spectral band (0.77 - 0.89um), B5 - red edge spectral band 1 (0.69 - 0.73um), B6 - red edge spectral band 2 (0.77 - 0.89um), B7 - purple spectral band (0.40 - 0.45um), B8 - yellow spectral band (0.59 - 0.63um);
[0032] 2) Preprocess the remote sensing image data;
[0033] Perform preprocessing such as orthorectification, radiometric calibration, and atmospheric correction on the obtained remote sensing image data; the preprocessing work is completed on the ENVI5.6.2 software platform.
[0034] Among them, orthorectification is a processing process for correcting the spatial and geometric distortion of an image to generate a multi-center projection plane orthoimage. In addition to correcting the distortion caused by general system factors, it can also correct the geometric distortion caused by terrain.
[0035] Radiometric calibration is the process of converting the voltage or digital quantization value recorded by the sensor into radiance.
[0036] The purpose of atmospheric correction is to eliminate the influence of factors such as the atmosphere and light on the reflection of ground objects, and obtain real physical model parameters such as ground object reflectance, radiation, and surface temperature, to eliminate the influence of water vapor, oxygen, carbon dioxide, methane, and ozone in the atmosphere on the reflection of ground objects, and eliminate the influence of atmospheric molecular and aerosol scattering. Atmospheric correction is the process of retrieving the true reflectance of ground objects.
[0037] 3) Calculate the bare land and sandy land index DBSI;
[0038] From the reflectance of the B1 - blue spectral band Blue B1 , the reflectance of the B2 - green spectral band Green B2 , and the reflectance of the B7 - purple spectral band Purple B7 , calculate the bare land and sandy land index DBSI, as shown in Equation 1:
[0039] DBSI = Blue B1 + Green B2 - Purple B7 (1)
[0040] The calculation formula of DBSI is obtained through a large amount of data analysis. Sample points are taken for various typical ground objects within the study area for analysis. The typical ground object types include bare sandy land, artificial buildings, bare land, saline - alkali beaches, vegetation, etc. Analysis shows that within the B1 to B8 band range, the reflectance of bare sandy land is between that of saline - alkali beaches and artificial buildings and bare land. The trend of reflectance height is different in different bands, generally higher, and increases with the increase of wavelength. Through further analysis, it is found that the reflectance of bare sandy land in the B1 blue light band is lower than that of artificial buildings, exceeds it in the B2 green light band, and on the B7 purple light band, the reflectance of bare sandy land is significantly lower than that of artificial buildings; compared with bare land, the reflectance of bare sandy land is higher in the B1 and B2 bands, but the reflectance of bare sandy land drops faster in the B7 band and is close to that of bare land. Based on this characteristic, it provides an important basis for better extracting the information of bare sandy land and realizes the acquisition of the calculation formula (1) of DBSI.
[0041] 4) Calculate the normalized difference vegetation index NDVI;
[0042] To further improve the accuracy of the monitoring results and reduce the misjudgment of DBSI in the desert transition zone, the NDVI is introduced to construct a decision tree algorithm. From the reflectance of the B3 - red spectral band Red B3 , and the reflectance of the B4 - near - infrared spectral band NIR B4 , calculate the normalized difference vegetation index NDVI, as shown in Equation 2:
[0043]
[0044] 5) Judgment of low - vegetation areas;
[0045] According to multiple experiments and field investigations, in the Kubuqi Desert, Mu Us Sandy Land, etc., the value of NDVI = 0.2 is the transition threshold between bare sandy land and desertified grassland. Therefore, for each remote - sensing pixel, judge whether the value of NDVI satisfies the condition: 0 < NDVI < 0.2. If it is satisfied, it is a low - vegetation area; if not, it is vegetation or water body.
[0046] 6) Determination of bare sand;
[0047] At each remote sensing pixel that meets the low vegetation zone, it is determined whether the DBSI value meets the condition: DBSI>C=0.15, C=0.15 is the empirical threshold of DBSI in Kubuqi Desert based on experiments and field investigations; if it meets the condition, it is bare sand, if not, it is a building, sand barrier or bare land.
[0048] The following is a further explanation of the decision tree identification method for exposed sand land based on high-resolution satellite remote sensing images in conjunction with a specific application example.
[0049] In this embodiment, the Kubuqi Desert in Ordos, Inner Mongolia is taken as the research object, and the 16-meter resolution data of the WFV sensor of the Gaofen-6 satellite is obtained. The shooting data is at 12:10:05 on September 6, 2022, 17 / 66 (scene path / Row), the solar altitude angle is 34.956 degrees, and the solar azimuth angle is 161.902 degrees, including blue spectrum (0.45-0.52um), green spectrum (0.52-0.59um), red spectrum (0.63-0.69um), near-infrared spectrum (0.77-0.89um), red edge spectrum 1 (0.69-0.73um), red edge spectrum 2 (0.77-0.89um), purple spectrum (0.40-0.45um), yellow spectrum (0.59-0.63um) and other 8 bands, with a width of ≥800km. The scene data completely covers the target monitoring area.
[0050] In the study area, 10 sample points were selected for evaluation of various typical landforms, and the typical values of various typical landforms in the 8 bands of GF-6 satellite WFV were obtained (see Table 1). Figure 2 As shown in the figure, a line graph is produced by averaging 10 sample points of various typical landforms within the study area.
[0051] Table 1 Mean reflectivity of bare sand and related landform types in the study area
[0052]
[0053] According to Figure 1 The technical process of the present invention shown in the figure calculates the NDVI and DBSI indexes; uses the calculated NDVI and DBSI indexes to judge the remote sensing pixels in the study area, and finally obtains a complete bare sand distribution map of the Kubuqi Desert, as shown in FIG. Figure 4 The original optical image of the study area is shown in Figure 3 shown.
[0054] Under visual interpretation, this embodiment has a good effect on filtering out artificial buildings, sand barriers and other objects. The desert boundary in the remote sensing image extracted by the algorithm is highly fitted with the true color synthesis. Figure 3 and Figure 4 As shown in Figure 2, the 2-meter high-resolution remote sensing image of the same year was used to evaluate the accuracy of the confusion matrix, and the overall accuracy reached 96.8%.
[0055] It can be seen that the decision tree identification method for exposed sand based on high-resolution satellite remote sensing images disclosed in the present invention has the following technical advantages compared with the prior art:
[0056] 1) A new calculation algorithm for the bare sand index (DBSI) is proposed, which has a clear physical meaning;
[0057] 2) The recognition method of the present invention is mainly based on the reflectivity information of satellite remote sensing images, and the amount of calculation is low;
[0058] 3) The advantages of clear physical meaning and low computational complexity make the method of the present invention widely applicable, and can be applied to the monitoring of bare sand changes in large areas or the entire land;
[0059] 4) The judgment threshold C can be optimized for different soil types to improve the estimation accuracy;
[0060] 5) It has spatial independence, only the remote sensing data corresponding to the target pixel is used as input, and the data of the neighboring pixels of the pixel are not required to participate in the calculation, and the error information of other pixels is not introduced, making the recognition result more accurate;
[0061] 6) No need for ground monitoring, it can cover sparsely populated areas, saving a lot of costs.
[0062] The above implementation modes are not limitations of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions or substitutions made by technicians in this technical field within the scope of the technical solution of the present invention also belong to the protection scope of the present invention.
Claims
1. A decision tree extraction method for bare sandy land based on high-resolution satellite remote sensing, characterized in that: It includes the following extraction processes: 1) Prepare high-resolution satellite remote sensing image data; 2) Preprocess the obtained remote sensing image data; 3) Calculate the bare land sandy land index DBSI; Among them, the reflectivity of the blue spectrum segment of B1 is Blue B1 , B2-reflectivity of green spectrum GReen B2 、B7-Reflectivity of purple spectrum B7 , calculate the bare sand index DBSI, see formula 1: DBSI=Blue B1 +Green B2 -Purple B7 (1); 4) Introduce NDVI to construct a decision tree algorithm and calculate the normalized difference vegetation index NDVI; 5) On each remote sensing pixel, divide the low vegetation area, vegetation or water body according to the calculated value of NDVI; that is, determine whether the value of NDVI satisfies the condition: 0 < NDVI < 0.
2. If it is satisfied, it is the low vegetation area. If it is not satisfied, it is vegetation or water body; 6) On each remote sensing pixel that satisfies the low vegetation area, extract the bare sandy land according to the calculated value of DBSI; on each remote sensing pixel that satisfies the low vegetation area, determine whether the value of DBSI satisfies the condition: DBSI > C = 0.15, where C = 0.15 is the empirical threshold of DBSI; if it is satisfied, it is the bare sandy land. If it is not satisfied, it is buildings, sand barriers or bare land.
2. The decision tree extraction method for bare sandy land based on high-resolution satellite remote sensing according to claim 1, characterized in that: In 1), the prepared high-resolution satellite remote sensing data covers the target monitoring area and includes the reflectance data of 8 bands of the high-resolution satellite WFV sensor, which are: B1 - blue spectral band, B2 - green spectral band, B3 - red spectral band, B4 - near-infrared spectral band, B5 - red edge spectral band 1, B6 - red edge spectral band 2, B7 - purple spectral band, B8 - yellow spectral band.
3. The decision tree extraction method for bare sandy land based on high-resolution satellite remote sensing according to claim 2, characterized in that: The wavelength of the B1 - blue spectral band is between 0.45 - 0.52um, the wavelength of the B2 - green spectral band is between 0.52 - 0.59um, the wavelength of the B3 - red spectral band is between 0.63 - 0.69um, the wavelength of the B4 - near-infrared spectral band is between 0.77 - 0.89um, the wavelength of the B5 - red edge spectral band 1 is between 0.69 - 0.73um, the wavelength of the B6 - red edge spectral band 2 is between 0.77 - 0.89um, the wavelength of the B7 - purple spectral band is between 0.40 - 0.45um, and the wavelength of the B8 - yellow spectral band is between 0.59 - 0.63um.
4. The decision tree extraction method for bare sandy land based on high-resolution satellite remote sensing according to claim 1, characterized in that: In 2), the data preprocessing methods include orthorectification, radiometric calibration, and atmospheric correction.
5. The decision tree extraction method for bare sandy land based on high-resolution satellite remote sensing according to claim 1, characterized in that: 4), the reflectance of the red band of B3 B3 、B4-NIR reflectivity B4 , calculate the normalized difference vegetation index NDVI, see formula 2:
