A lawn identification and extraction method based on remote sensing big data

Abstract

The present invention provides a lawn identification and extraction method based on remote sensing big data, comprising the following steps: collecting Planet multispectral satellite images; directly obtaining all image data of Planet since its launch in September 2015; obtaining all surface reflectance data; fitting a smooth time series change trend line; obtaining all trend turning points using a linear segmentation method, which include local extreme points and inflection points, and calculating and counting the differences between the maximum and minimum values of lawn types and annual average NDVI in different seasons; removing local noise extreme points; calculating the NDVI difference between two adjacent points based on all trend turning points, and removing non-inflection point differences with smaller differences; after removing the noise extreme points and non-inflection points, obtaining the true inflection points of the long-term trend change of the lawn, and finally obtaining the parameters required for lawn identification, removing non-lawn interference, and calculating the lawn-related parameters extracted based on the time series inflection point technology to generate a mapping map of the study area.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing big data, and in particular to a lawn identification and extraction method based on remote sensing big data. Background Art

[0002] With the continuous development and application of remote sensing technology in the field of urban ecological environments, the identification, mapping, and dynamic monitoring of urban lawns have achieved unprecedented precision and spatial and temporal scales. However, current research, both domestically and internationally, has focused less on distinguishing between lawns and non-lawns within urban areas. Spatial data on the distribution of urban lawns is scarce, to the point where they are not reflected in existing large-scale land use and land cover maps. Indicators such as the distribution, frequency, and duration of urban lawn use are crucial indicators of urban land degradation, biodiversity conservation, and supply chain changes. The lack of this information will severely impact national and regional urban land resource planning and the development of policies to improve ecological and biodiversity conservation.

[0003] In recent years, the rapid development of remote sensing big data and related technologies has provided unprecedented opportunities for large-scale, high-precision, and real-time satellite monitoring of land use and land cover changes. Therefore, building on past related research, this paper proposes to utilize remote sensing big data and super cloud computing technology to perform high-spatial resolution, real-time identification, and range extraction of urban lawns. This approach aims to fill the gaps in existing urban lawn mapping methods and technologies, and further quantify the effectiveness of urban lawn policy implementation. This research will help promote research and technological innovation in the application of remote sensing big data technology in land use and land cover changes. It can also provide an important scientific basis for comprehensive regional urban resource planning, and provide valuable scientific support for the formulation of important national policies related to sustainable urban development and ecological security.

[0004] Research has been conducted both domestically and internationally on real-time monitoring of neighborhood-scale lawns and the conservation of biodiversity in designated areas, focusing on the extraction of lawns from space and ground-based remote sensing. Significant progress and achievements have been made. However, the resulting atlases, due to low spatiotemporal resolution, time lags, and a lack of urban-scale studies, struggle to provide accurate references for lawn biodiversity conservation for government decision-makers and non-governmental environmental protection organizations. However, significant deficiencies remain in the understanding of the impact mechanisms and large-scale research based on high-resolution aerospace remote sensing data. Scientific questions urgently need to be answered, such as the temporal and spatial pattern evolution of urban-scale lawns and the extent to which their temporal and spatial variations influence the assessment of their biodiversity conservation value. These findings are primarily manifested in the following aspects: First, remote sensing extraction methods for urban lawns need improvement, hindering large-scale, high-resolution monitoring of lawns. Second, due to technical bottlenecks in efficient real-time monitoring of urban lawns and the construction of high-resolution mapping models, the mechanisms for evaluating the biodiversity conservation value of urban-scale lawns remain unclear, necessitating further theoretical exploration and practical modeling.

[0005] Therefore, a new method is needed to solve the above technical problems. Summary of the Invention

[0006] The purpose of this invention is to address the deficiencies of the existing technology and provide a lawn identification and extraction method based on remote sensing big data. The specific scheme is as follows:

[0007] A lawn identification and extraction method based on remote sensing big data comprises the following steps:

[0008] Step 1: Collect Planet multispectral satellite images;

[0009] Step 2: Use all Planet image data from its launch in September 2015 to date directly from the cloud computing platform;

[0010] Step 3: Comprehensively remove clouds, shadows, water bodies, and ice and snow;

[0011] Step 4: Obtain all surface reflection data based on the SIAC method of the atmospheric transport model 6s;

[0012] Step 5: Based on data reconstruction technology, remove noise and fit the smoothed Planet 3m and 24-hour NDVI time series change trend lines. The change trend between two adjacent time series points can be divided into three forms: "increasing", "maintaining" and "decreasing".

[0013] Step 6: Use the linear segmentation method to obtain all trend turning points, including local extreme points and inflection points, and calculate and count the differences between the maximum and minimum annual average NDVI values of lawn types and different seasons;

[0014] Step 7: Take 50% of the minimum difference as the median greenness value of all lawns in the study area, and then use this phenological index to remove local noise extreme points;

[0015] Step 8: Calculate the NDVI difference between two adjacent points based on all trend turning points after denoising, and then use the median green value again to remove the non-inflection point differences with smaller differences;

[0016] Step 9: After removing the noise extreme points and non-inflection points, the true inflection points of the long-term trend change of the lawn are obtained, and finally the parameters required to identify the lawn are obtained. The latest land use and cover data of the study area are used to mask out non-lawn interference. The lawn-related parameters extracted based on the time series inflection point technology are calculated at the pixel scale to produce a drawing of the study area.

[0017] Based on the above, step 4 requires supplementary surface reflectance data from Landsat and Sentinel-2. For example, in years with heavy cloud cover, Landsat, Sentinel-2, and Planet fusion technology can be used to compensate for missing data and to roughly estimate historical urban lawn changes before the launch of the Planet satellite using Landsat and Sentinel-2.

[0018] Based on the above, there are six situations in which the trend turning points between the three adjacent time series points in step 5 are "rise-maintain", "rise-fall", "hold-rise", "hold-fall", "fall-rise", and "fall-maintain".

[0019] The present invention has outstanding substantial features and significant progress compared to the prior art. Specifically, the present invention has the following advantages:

[0020] This invention is based on Planet satellite images with high spatial resolution (3 meters) and temporal resolution (24 hours), utilizes remote sensing fine identification technology of artificial lawns and natural lawns, and with the help of a geographic cloud computing platform, constructs an innovative, fully automatic, real-time lawn identification and extraction method framework, providing favorable scientific support for the formulation of important policies related to national urban sustainable development and ecological security. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Flowchart for high temporal and spatial resolution urban lawn remote sensing fine identification and automatic mapping.

[0022] Figure 2 a is the type of trend change of three adjacent points in the time series; Figure 2 b is the time series formula; Figure 2 c is the turning point that satisfies the formula.

[0023] Figure 3 This is a schematic diagram of urban lawn types and corresponding remote sensing interpretation characteristics.

[0024] Figure 4 It is a real-time diagram of urban lawns. DETAILED DESCRIPTION

[0025] The technical solution of the present invention is further described in detail below through specific implementation methods.

[0026] Example

[0027] like Figure 1 As shown, an embodiment of the present invention provides a lawn identification and extraction method based on remote sensing big data, which is characterized by comprising the following steps:

[0028] Step 1: Collect Planet multispectral satellite imagery (MSI), which has a spatial resolution of up to 3 meters in some bands (red, green, blue, and near-infrared bands). Because the entire satellite constellation has 32 small satellites in orbit around the International Space Station (ISS) and 100 small satellites in sun-synchronous orbit (SSO), a one-day repetition period can be achieved;

[0029] Step 2: Use all Planet image data from its launch in September 2015 to date directly from the cloud computing platform;

[0030] Step 3: Comprehensively remove clouds, shadows, water bodies, and ice and snow;

[0031] Step 4: Obtain all surface reflectance data using the SIAC method using the 6-second atmospheric transport model. This step requires supplementary surface reflectance data from Landsat (30 m, 16 days) and Sentinel-2 (10 m, 5 days). For example, in years with high cloud cover, Landsat, Sentinel-2, and Planet fusion techniques can be used to compensate for missing data. Landsat and Sentinel-2 can also be used to roughly estimate historical urban lawn changes before the launch of the Planet satellite.

[0032] Step 5: Based on data reconstruction technology, remove noise and fit the smooth Planet 3m and 24-hour NDVI time series change trend lines. The change trend between two adjacent time series points can be divided into three forms: "increase", "maintain" and "decrease". Among them, the trend turning situations include: "increase-maintain", "increase-decline", "maintain-increase", "maintain-decline", "decline-increase" and "decline-maintain" ( Figure 2 a), so in the NDVI time series X ( Figure 2 b) If xi is a trend turning point, then xi satisfies Figure 2 c formula;

[0033] The present invention refers to some key points in the trend turning points that can reflect key information such as the reversal of the long-term trend of the time series as "inflection points". This "inflection point" refers to the time point when the shape of the lawn vegetation index time series curve changes significantly. It is different from the "inflection point" defined in mathematics - the point where the sign of the curve derivative changes. The lawn vegetation index time series curve has obvious shape differences in different seasons. The key inflection point information of the time series can potentially identify urban lawns.

[0034] Step 6: Use the linear segmentation method (PLR) to obtain all trend turning points, including local extreme points (peaks or concave points) and inflection points, calculate and count the differences between the maximum and minimum values of the annual average NDVI of lawn types and different seasons, and divide the lawn types and corresponding remote sensing interpretation features into two categories, namely artificial lawn and natural lawn (such as Figure 3 ).

[0035] Step 7: Although the algorithm in step 6 is computationally efficient and suitable for big data calculations, and can better preserve the change pattern of the original time series, the results are often trivial and retain a large number of unfiltered detail changes. Therefore, these trend turning points must be further processed to extract the inflection points required by the present invention; based on the spatial range data of the lawns in the study area, statistical yearbooks, and data on lawn types collected in the field, the difference between the maximum and minimum annual average NDVI of vegetation of different types and seasons is calculated and counted, and 50% of the minimum difference amplitude is used as the median greenness value of all lawns in the study area. This phenological indicator is then used to remove local noise extreme points (such as local low peaks and high concave values of the curve);

[0036] Step 8: Calculate the NDVI difference between two adjacent points based on all trend turning points after denoising, and then use the median green value again to remove the non-inflection point differences with smaller differences;

[0037] Step 9: Since the NDVI curve of lawn vegetation index is more regular than that of natural vegetation, the NDVI of lawn has significant differences during the artificial lawn renovation period and in different seasons, so the inflection points with high NDVI differences can be accurately retained; lawns that cannot be harvested due to drought, pests and diseases during vegetation growth will also be automatically removed because they have very small NDVI differences or a very short artificial lawn renovation period (less than 1 month); in short, after removing noise extreme points and non-inflection points, the true inflection points of long-term trend changes in lawns are obtained, and finally the parameters required for lawn identification are obtained. Using the latest land use and land cover data of the study area, the mask is removed to remove non-lawn interference (the remaining non-lawn areas can also be removed by inflection point technology in the future). The lawn-related parameters extracted based on the time series inflection point technology are calculated at the pixel scale to produce a drawing of the study area: a real-time drawing (one scene every day) of a high spatial resolution (3m) schematic diagram of the urban lawn (such as Figure 4 ), and then draw a map of the city lawn for each day.

[0038] This method can verify the accuracy of the framework model and mapping results of the method in multiple aspects through different channels: ① field data collection; ② consulting and collecting urban lawn information from local agricultural departments; ③ comparing with other relevant spatial auxiliary data.

[0039] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to preferred embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or some technical features can be replaced by equivalents without departing from the spirit of the technical solutions of the present invention. They should all be included in the scope of the technical solutions claimed for protection by the present invention.

Claims

1. A lawn identification and extraction method based on remote sensing big data, characterized in that The following steps are involved: Step 1: Collect Planet multispectral satellite images; Step 2: Use all Planet image data from its launch in September 2015 to date directly from the cloud computing platform; Step 3: Comprehensively remove clouds, shadows, water bodies, and ice and snow; Step 4: Obtain all surface reflection data based on the SIAC method of the atmospheric transport model 6s; Step 5: Using data reconstruction technology, remove noise and fit smoothed Planet 3m and 24-hour NDVI time series trend lines. The trend between two adjacent time series points can be divided into three forms: "increasing", "maintaining", and "decreasing". Step 6: Use the linear segmentation method to obtain all trend turning points, including local extreme points and inflection points, and calculate and count the differences between the maximum and minimum annual average NDVI values of lawn types and different seasons; Step 7: Take 50% of the minimum difference as the median greenness value of all lawns in the study area, and then use this phenological index to remove local noise extreme points; Step 8: Calculate the NDVI difference between two adjacent points based on all trend turning points after denoising, and then use the median green value again to remove the non-inflection point differences with smaller differences; Step 9: After removing the noise extreme points and non-inflection points, the true inflection points of the long-term trend change of the lawn are obtained, and finally the parameters required to identify the lawn are obtained. The latest land use and cover data of the study area are used to mask out non-lawn interference. The lawn-related parameters extracted based on the time series inflection point technology are calculated at the pixel scale to produce a drawing of the study area.

2. The lawn identification and extraction method based on remote sensing big data according to claim 1, characterized in that: Step 4 requires the addition of Landsat and Sentinel-2 surface reflectance data.

3. The lawn identification and extraction method based on remote sensing big data according to claim 1, characterized in that: There are six situations in which the trend turning points between the three adjacent time series points in step 5 are "rise-maintain", "rise-fall", "maintain-rise", "maintain-fall", "fall-rise", and "fall-maintain".

Images