Multi-temporal remote sensing method for extracting cultivated land planting attributes at plot scale

Abstract

The present invention provides a multi-temporal remote sensing extraction method for cultivated land planting attributes at a plot scale, comprising: a first step: performing data collection and preprocessing; a second step: constructing 10m multi-temporal NDVI time series data; a third step: utilizing the multi-temporal variation characteristics of crop NDVI to perform time series NDVI cluster analysis; a fourth step: based on previous land use survey data and combined with satellite remote sensing data clustering constraints, generating a spatiotemporal standard sample set of surface cover types, and constructing a standard time series curve of ground object NDVI; a fifth step: utilizing the standard time series curve of ground object NDVI to classify the 10m multi-temporal NDVI clustering results, integrating and constraining cultivated land plots on the basis of previous land use maps, and extracting planting attribute information.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing monitoring of cultivated land, and in particular to a multi-temporal remote sensing extraction method for cultivated land planting attributes at a plot scale. Background Art

[0002] Arable land is a fundamental resource for human survival and development. With socioeconomic development and changes in agricultural management practices, crop cultivation activities on arable land have gradually diversified, resulting in a variety of crop attributes within arable land, including grain, forage, forestry, and greenhouse agriculture. These uses not only reduce crop planting area but also impact arable land environmental factors such as soil fertility, hindering the large-scale development of agriculture. The Third National Land Survey explicitly requires a detailed investigation of the current status of arable land, requiring the cleanup and annotation of arable land use within arable land plots. For example, "temporary garden tree planting," "ornamental gardening," "fast-growing trees," and "green grassland" are classified as "other land types." Furthermore, arable land crop attributes are annotated for each plot based on actual cultivation conditions, including "uncultivated," "forest-grain intercropping," "grain crop planting," "non-grain crop planting," "grain and non-grain rotation," and "fallow." Therefore, a method for rapidly and accurately obtaining information on arable land crop attributes is urgently needed in agricultural production management to timely understand various crop use and changes, enabling more scientific and rational formulation of agricultural development and management strategies.

[0003] Satellite remote sensing technology, with its multi-scale, multi-temporal, and multi-spectral Earth observation capabilities, provides accurate and objective information about land features in both space and time, and has been widely used for regional-scale agricultural mapping. Terra / MODIS daily observations capture dynamic changes in crop growth and have become the most commonly used data for large-scale farmland use monitoring. Compared to MODIS data, which has a lower spatial resolution of 500 meters (250 meters in the visible and near-infrared bands), Landsat satellites (15-30 meters) can more accurately monitor crop variations across fields. However, their longer revisit periods, coupled with the influence of meteorological conditions such as clouds, make it difficult to obtain remote sensing data with high spatial and temporal resolution. The European Space Agency's Sentinel-2 satellite, with its five-day revisit period and 10-meter resolution, provides a data foundation for time-series remote sensing monitoring at the field-scale over large areas. Numerous studies have used Sentinel-2 satellite data for large-scale, high-precision crop monitoring in farmland, and their results have demonstrated that this satellite observation data significantly improves the accuracy of regional-scale crop monitoring. Sentinel-2's 10m spatial resolution observation data can, on the one hand, carry out large-scale land use mapping at 10,000 to 50,000 square meters and improve the accuracy of change detection. On the other hand, it also provides an opportunity to expand the application of precise monitoring of cultivated land quality and provides a data foundation for large-area field-scale remote sensing monitoring of cultivated land planting attributes for land management needs.

[0004] In recent years, to address issues such as multi-source satellite observation data collection and data processing efficiency, Google has launched the Google Earth Engine (GEE) remote sensing big data platform. GEE stores nearly all publicly available satellite remote sensing and geospatial datasets from the past 40 years, totaling at the BP level. GEE also provides user-friendly human-computer interaction. Operators can use the platform to preprocess multi-source / multi-temporal satellite observation data without requiring powerful computers or specialized coding skills, downloading preprocessed data without having to download large amounts of raw data. Numerous regional and even global studies based on GEE have demonstrated that GEE has significantly enhanced the capabilities of large-scale, long-term, and high-spatial-resolution remote sensing monitoring, making it possible to monitor cultivated land attributes in a detailed and granular manner over large areas.

[0005] On the other hand, accurately acquiring information on cultivated land crop attributes requires a large number of training and validation samples. Currently, obtaining a large number of ground samples through field surveys or visual interpretation is difficult. Research has shown that the quantity and quality of training samples have the greatest impact on the final classification results, and mapping accuracy improves with increasing sample size. However, regional-scale mapping requires significant manpower and time to produce training sample sets, and it is difficult to update these samples over large areas. To improve sample acquisition efficiency, many countries are selecting samples based on existing land use survey products, such as the United States' Cropland Data Layer (CDL), Canada's National Crop Base Map, and Europe's LPIS data. China's Second Land Use Survey data are the most comprehensive and accurate 1:10,000 land use mapping product. However, there are currently no applications of combining this land use data with high-temporal and spatial resolution remote sensing data to obtain a large number of standard ground feature samples. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to address the above-mentioned shortcomings in the existing technology and provide a multi-temporal remote sensing method for extracting cultivated land planting attributes at the plot scale. This method can utilize existing land use data combined with high-temporal and spatial resolution remote sensing data to obtain a large number of standard samples. Based on the existing land use data, the method extracts cultivated land planting attributes at the plot scale by integrating time series NDVI constraints. According to the present invention, a multi-temporal remote sensing method for extracting cultivated land planting attributes at the plot scale is provided, comprising:

[0007] Step 1: Perform data collection and preprocessing;

[0008] Step 2: Based on the collected data, perform 10m multi-temporal NDVI time series data construction;

[0009] The third step: Based on the constructed 10m multi-temporal NDVI time series data, the multi-temporal variation characteristics of crop NDVI were used to perform time series NDVI cluster analysis;

[0010] Step 4: Based on the previous land use survey data and combined with clustering constraints, a standard sample set of surface cover types is generated, and a standard time series curve of NDVI of land features is constructed;

[0011] Step 5: Use the standard time series curve of ground object NDVI to classify the 10m multi-phase NDVI clustering results. Based on the previous land use map, integrate and constrain the cultivated land plots to extract planting attribute information.

[0012] Preferably, the fifth step utilizes the standard time series curve of the ground object NDVI and applies the dynamic time consolidation algorithm to perform category matching on the cluster analysis results, and extracts the cultivated land planting attribute information for the cultivated land area based on the previous land use data.

[0013] Preferably, in the fifth step, for the object to be classified T, its Wk value and the multiple standard types N are calculated according to the following formula, and the standard type m with the minimum W is the object to be classified:

[0014] γ(i,j)=∥Yi–Sj∥;

[0015] Where Yi is the NDVI of the object to be classified in month i (i = 1, 2, ..., 12); Sj is the standard NDVI of day j (j = 1, 2, ..., 365);

[0016] Di=min[γ(i,j)],j=1,2,…,365;

[0017] Where Di is the regular path, i=1,2,…12

[0018] Wk=∑Di i=1,2,…,12;

[0019] The type of the object to be classified that has the smallest distance from the standard type is the type of the object to be classified:

[0020] T∈m when Wm=min(Wk), k=1,..,m,…,N.

[0021] Preferably, the collected data include multi-temporal satellite remote sensing data with a spatial resolution of 10m or above and existing land use survey data.

[0022] Preferably, the accuracy of the collected preliminary land use data must meet the data accuracy of 150,000 or above, such as the second national land survey data, the third national land survey data, etc.

[0023] Preferably, the construction of 10m multi-temporal NDVI requires the use of, but not limited to, synthesis methods such as monthly maximum, mean, and median values. However, the temporal resolution of the data must be able to cover the growing season of the crops and be able to depict the phenological changes of the crops.

[0024] Preferably, the land use survey data does not include cultivated land planting attribute information.

[0025] Preferably, the surface cover types include: forest land, grassland, bare land, water bodies, construction land, facility agricultural land, garden land, arable land for growing grain crops in one season, arable land for growing grain crops in two seasons, and arable land for other crops.

[0026] Preferably, the collection of multi-temporal satellite remote sensing data, data preprocessing, construction of time series NDVI, cluster analysis of time series NDVI, and generation of standard time series curves of ground objects NDVI can be performed, but not limited to, in a cloud platform, and the remaining processing is performed in a local computer.

[0027] This application proposes a technical method for remote sensing extraction of regional cultivated land planting attribute information based on 10m time series remote sensing data. This method utilizes the multi-temporal variation characteristics of crop NDVI, based on existing land use survey data, combined with satellite remote sensing data clustering constraints, and selects typical category standard samples to quickly obtain a set of ground feature samples of planting attribute types. It also quickly extracts cultivated land planting attribute information through the integrated constraints of cultivated land plot scale (pattern), with an overall accuracy of more than 88%. This method solves the problem of obtaining a large number of samples of cultivated land planting attributes in a large area and the problem of rapid extraction of cultivated land planting attributes, providing a new technical method for monitoring cultivated land planting attributes in my country. BRIEF DESCRIPTION OF THE DRAWINGS

[0028] A more complete understanding of the present invention and its attendant advantages and features will be more readily appreciated by reference to the following detailed description taken in conjunction with the accompanying drawings, in which:

[0029] Figure 1 A flowchart of a method for extracting multi-temporal remote sensing properties of cultivated land at a plot scale according to a preferred embodiment of the present invention is schematically shown.

[0030] Figure 2 The figure schematically shows the location of the study area and the distribution of the field survey sample points according to the multi-temporal remote sensing extraction method of plot-scale cultivated land planting attributes according to the preferred embodiment of the present invention.

[0031] Figure 3 The diagram schematically shows the number of effective observations after cloud removal by Sentinel-2A / B in the Beijing-Tianjin-Hebei region in 2019.

[0032] Figure 4 The spatial distribution of standard samples of land features in Beijing, Tianjin and Hebei is schematically shown.

[0033] Figure 5 The median NDVI value of the standard sample in the northern Beijing-Tianjin-Hebei region in 2019 is shown schematically.

[0034] Figure 6 The median NDVI value of the standard sample in the southern Beijing-Tianjin-Hebei region in 2019 is shown schematically.

[0035] Figure 7 The spatial distribution of the cultivated land planting attribute extraction results of the multi-temporal remote sensing extraction method of cultivated land planting attributes at plot scale according to a preferred embodiment of the present invention is schematically shown; wherein a1 and a2 represent images of area a, b1 and b2 represent images of area b, c1 and c2 represent images of area c, and d1 and d2 represent images of area d.

[0036] Figure 8 The figure schematically shows the precision matrix of the cultivated land planting attribute extraction results of the multi-temporal remote sensing extraction method of cultivated land planting attributes at the plot scale according to the preferred embodiment of the present invention.

[0037] It should be noted that the accompanying drawings are intended to illustrate the present invention, not to limit it. Note that the accompanying drawings showing structures may not be drawn to scale. Furthermore, in the accompanying drawings, identical or similar elements are labeled with identical or similar reference numerals. DETAILED DESCRIPTION

[0038] In order to make the contents of the present invention clearer and easier to understand, the contents of the present invention are described in detail below with reference to specific embodiments and drawings.

[0039] High-precision mapping of cultivated land planting attributes is the basis for monitoring the quantity and quality of cultivated land. In response to the lack of methods for extracting cultivated land planting attributes in large areas, this application proposes a method for quickly extracting cultivated land planting attributes based on clustering and standard sample matching (MSC-Basing on Matching Samples with Clustering Analysis). First, using existing land use survey data and combining it with clustering constraints of satellite remote sensing data, standard samples of typical categories are selected to quickly obtain a set of ground feature samples including those for planting grain crops (1 season), planting grain crops (2 seasons), grassland, water area, bare land, greenhouse, woodland, and garden land, solving the problem of obtaining a large number of samples in a large area. Secondly, using the standard time series curve of the sample NDVI (normalized difference vegetation index), the dynamic time warping algorithm (DTW) is applied to class match the cluster analysis results to extract cultivated land planting attribute information within the cultivated land area. The test results show that based on the previous land use data, a large number of ground feature samples can be quickly obtained using time series remote sensing data, and cultivated land planting attribute information can be quickly extracted. The method of the present invention can realize accurate and rapid extraction of cultivated land planting attribute information in a large area, with an overall accuracy of 0.91, meeting the needs of regional cultivated land planting attribute monitoring.

[0040] Specific preferred embodiments of the present invention will be described below with reference to the accompanying drawings.

[0041] Figure 1 A flowchart of a method for extracting multi-temporal remote sensing properties of cultivated land at a plot scale according to a preferred embodiment of the present invention is schematically shown.

[0042] like Figure 1 As shown, the multi-temporal remote sensing method for extracting plot-scale cultivated land planting attributes according to a preferred embodiment of the present invention includes:

[0043] The first step S1: perform data collection and preprocessing;

[0044] For example, the collected data include multi-temporal satellite remote sensing data and existing land use survey data.

[0045] Here, the data preprocessing method may be performed in any appropriate manner.

[0046] Step 2: Based on the collected data, perform 10m multi-temporal NDVI time series data construction;

[0047] Here, the construction of multi-temporal NDVI requires the use of, but is not limited to, synthesis methods such as monthly maximum, mean, and median values. However, the temporal resolution of the data must be able to cover the growing season of the crops and be able to depict the phenological changes of the crops.

[0048] The third step: based on the constructed 10m multi-temporal NDVI time series data, perform time series NDVI cluster analysis;

[0049] Here, the number of cluster categories is generally set to 2-3 times the number of surface cover types in the study area.

[0050] Step 4: Based on the previous land use survey data and combined with clustering constraints, a standard sample set of surface cover types is generated, and a standard time series curve of NDVI of land features is constructed;

[0051] Step 5: Use the standard time series curve of ground object NDVI to classify the 10m multi-phase NDVI clustering results. Based on the previous land use map, integrate and constrain the cultivated land plots to extract planting attribute information.

[0052] Preferably, the collection of multi-temporal satellite remote sensing data, data preprocessing, construction of time series NDVI, cluster analysis of time series NDVI, and generation of standard time series curves of ground objects NDVI can be performed in the cloud platform, but not limited to, and the rest of the processing is performed in the local computer.

[0053] Below, the implementation methods of the present invention are described in detail using Beijing, Tianjin, and Hebei Province (referred to as Jingjinji) as a research area.

[0054] By leveraging the GEE cloud platform and taking advantage of the Sentinel-2A / B satellite multi-temporal observation data, and combining multi-temporal NDVI data with the constraints of existing land use maps to obtain a large number of training and validation samples, a method for rapid monitoring of cultivated land planting attributes at the regional scale using 10m multi-temporal satellite observations was proposed, providing a decision-making basis for the fine management of regional cultivated land.

[0055] The Beijing-Tianjin-Hebei region (113°E-120°E, 36°N-43°N) is the largest economically developed and most densely populated region in northern China, with a total area of 217.156 km 2 . The terrain is characterized by high altitude in the northwest and low altitude in the southeast. There are great differences in meteorological conditions such as temperature, precipitation, and wind speed between the north and the south of the region. The north is mainly mountainous and has a mid-temperate climate, while the south is mainly plains and has a warm temperate climate. The annual rainfall is between 400 and 800 mm, and rainfall is mainly concentrated in the summer. The region has drastic changes in land use and intensive human-land interaction. There are various forms of arable land use, including: growing food crops, pond aquaculture, greenhouse agriculture, etc. It is a typical area for the study of arable land planting properties.

[0056] A region's thermal resources determine the annual growth period of crops, often expressed as accumulated temperature. Accumulated temperature refers to the sum of daily average temperatures during periods when the average daily temperature in a region reaches or exceeds 10°C and remains above that level. The cropping system in the Beijing-Tianjin-Hebei region varies along the 3400°C annual accumulated temperature line. The northern region has a single cropping system, primarily growing corn, oats, millet, sorghum, and other grain crops, typically sown in April and harvested in October. The southern region has a double cropping system, primarily following a winter wheat-summer corn rotation. Winter wheat is sown around October each year and begins to green up in March or April of the following year, rapidly increasing vegetation cover. After harvesting in June, corn is planted. After harvesting around October, some plots are left fallow while others continue to grow winter wheat. Other crops, such as rice, spring corn, and lotus roots, also have a single cropping system. All crops are sown around May, reach peak growth around September, and are harvested in October.

[0057] Data collection and preprocessing

[0058] (1) Multi-temporal satellite remote sensing data

[0059] Based on the Google Earth Engine (GEE) remote sensing big data platform, we collected 10m spatial resolution Sentinel-2A / B data covering the Beijing-Tianjin-Hebei region in 2019. The data consists entirely of radiometrically corrected top-of-atmosphere reflectance (TOA) data (L1C-level), totaling 5,856 scenes. Sentinel-2A / B data, from the European Space Agency, are high-resolution, multispectral imaging satellites with a five-day return period. They include 12 bands: four 10m visible and NIR bands, six 20m red-edge and SWIR bands, and two 60m atmospheric bands. The QA60 band of the 10m visible and NIR band data from the L1C-level product was used here.

[0060] In order to verify the remote sensing extraction method of cultivated land planting attributes proposed in the present invention, 200 scenes of GF-1 and GF-2 data covering the study area in 2019 were collected. GF-1 data has two widths of 60km and 800km, with a spatial resolution of 2m resolution panchromatic / 8m resolution multispectral and 16m resolution multispectral, and a return period of 41 days (when not tilted). GF-2 data has a width of 45km, a spatial resolution of 4m multispectral and 1m panchromatic bands, and a return period of 69 days (when not tilted). The present invention uses the HSI transformation fusion method to generate 2m resolution GF-1 multispectral data and 1m resolution GF-2 multispectral data. Detailed data is shown in the table below.

[0061]

[0062] (2) Land use and ground survey data

[0063] We collected land use survey data for the Beijing-Tianjin-Hebei region in 2017 from the Second National Land Use Survey. Data includes eight first-level categories, including cultivated land, gardens, woodlands, grasslands, and water bodies, and 38 second-level categories, mapped at a scale of 1:10,000. The country conducts an annual land use change survey, which requires significant human and material resources. This survey updates the previous year's land use data through visual interpretation and ground verification using high-resolution remote sensing imagery (such as High Resolution Image Series (GF), Worldview, Quickbird, and aerial imagery). In this specific example, the data comes from the China National Land Surveying and Planning Institute. The land use data processing and analysis results (shown in this article) are images and text without coordinate information, saved in JPEG and txt formats. All results have been rigorously reviewed.

[0064] The extracted cropland attributes were further verified through ground surveys, which were conducted in the study area in 2018 and 2019. During the surveys, various cropland attributes were recorded and photographed, including 246 typical sample sites for cropland cultivation (single-season), cropland cultivation (two-season), cash crops, grassland, and greenhouse agriculture.

[0065] Specific processing operations

[0066] The growth processes of different vegetation in cultivated land, such as growth, development, harvest or withering, can all be revealed by multi-temporal NDVI data. The present invention proposes a method for directly extracting cultivated land planting attributes using multi-temporal NDVI temporal change information to improve information extraction efficiency. Using the standard time series curve of ground object NDVI, the dynamic time warping algorithm (Dynamic TimeWarping, DTW) is applied to perform category matching on the cluster analysis results, and the cultivated land planting attribute information is extracted for the cultivated land area. The proposed method is called MSC (basing on Matching Samples with Clustering analysis) method. The main process includes time series NDVI construction and cluster analysis, establishment of ground object standard samples, planting attribute extraction, comparison and verification of methods. The method process is as follows: Figure 1 shown.

[0067] Time series NDVI construction and cluster analysis

[0068] After statistical analysis of the coverage frequency of Sentinel-2A / B multi-temporal cloud mask data products in the study area, it was found that although the Sentinel-2A / B observation cycle is 5 days, due to the superposition of observations from nearby orbits, some areas can be observed every 2 days. Overall, effective cloud-free data can be obtained at least every month ( Figure 3 Therefore, using the GEE platform, we collected Sentinel-2A / B data from the study area in 2019, calculated the normalized vegetation index (NDVI = (NIR-Red) / (NIR+Red)) for each scene image, performed monthly NDVI maximum value synthesis processing, and then downloaded the processed multi-temporal NDVI data from the platform.

[0069] In the multi-temporal NDVI changes, different land cover types have different change characteristics, which can be used to identify different land cover types. The K-mean clustering method was used to perform cluster analysis on the NDVI time series data for 12 months in 2019. The number of categories in the K-mean clustering is generally set to 2-3 times the number of land cover types in the study area. The initial clustering is set to 20 categories, the number of iterations is 40, and the transformation threshold is 0.05%. In order to solve the problem of difficulty in clustering operations due to the large regional scope, this study performed clustering according to the administrative divisions of Beijing-Tianjin-Hebei (13 regions).

[0070] Generation of standard samples and NDVI time series data

[0071] Accurately acquiring information on cultivated land planting attributes requires a large number of training and validation samples. This paper uses the 2017 land use map and 2019 time-series NDVI data of the Beijing-Tianjin-Hebei region to quickly extract planting attribute samples to improve the efficiency of extracting cultivated land planting attribute information. Key technical methods in the processing process include time-series NDVI construction and cluster analysis, spatial overlay of patches to extract pure categories, etc. Land use survey data is high-precision data that has been verified through multiple layers of verification, such as visual interpretation of satellite imagery and ground surveys.

[0072] The land use survey data used here does not include information on cultivated land planting attributes such as the number of grain crop planting seasons, greenhouses, and other real-time surface coverage information; and since it is data from the previous year, its land use will also change in that year. To this end, the present invention superimposes the multi-temporal NDVI cluster analysis results of the year and extracts standard samples of surface cover types. The surface cover types include 10 categories: woodland (including forests and shrubs), grassland, bare land, water bodies, construction land, facility agricultural land (vegetable greenhouses and livestock houses, etc.), gardens, cultivated land for growing grain crops (1 season), cultivated land for growing grain crops (2 seasons), and cultivated land for other crops (economic crops, vegetables, etc.). GEE is further used to extract the cloud-free NDVI values obtained by the pixels at different observation times throughout the year for each type of sample (polygon) using Sentinel-2A / B data pixels as units, and perform time series smoothing; for example, the median of the NDVI of all smoothed pixels in the sample is calculated in units of 1 day to obtain various standard NDVI time series values.

[0073] Planting attribute information extraction and verification

[0074] Using the standard NDVI time series curves, the Dynamic Time Warping (DTW) algorithm was applied to classify the cluster analysis results and extract cropping attribute information for cultivated land areas. The DTW algorithm calculates the distance between two time series curves with multiple permutations. The curve with the shortest distance is called the normalization path, and the sum of the normalized paths is called the DTW distance (W). For each object to be classified, T, its Wk value is calculated for each of the multiple standard types (N = 10) according to the following formula. The standard type m with the minimum W is the object to be classified.

[0075] γ(i,j)=∥Yi–Sj∥…(1)

[0076] Where Yi is the NDVI of the object to be classified in month i (i = 1, 2, ..., 12); Sj is the standard NDVI of day j (j = 1, 2, ..., 365)

[0077] Di=min[γ(i,j)],j=1,2,…,365…(2)

[0078] Where Di is the regular path, i=1,2,…12

[0079] Wk=∑Di i=1,2,…,12…(3)

[0080] Where Di is the sum of regular paths, k = 1, .., N

[0081] The type of the object to be classified with the minimum distance from the standard type (Formula (4)) is the type of the object to be classified.

[0082] T∈m when Wm=min(Wk),k=1,..,m,…,N…(4)

[0083] The accuracy confusion matrix was used to calculate various indicators to verify the accuracy of the two methods, including overall accuracy, mapping accuracy, user accuracy and Kappa coefficient.

[0084] Spatiotemporal distribution of ground feature standard sample set

[0085] Using the combined method of land use survey data and NDVI cluster analysis, a total of 47,120 standard samples were screened and extracted (see Figure 4 ), including 10,217 forest lands, 4,434 grasslands, 394 bare lands, 4,098 construction lands, 1,230 water bodies, 2,381 gardens, 4,617 facility agricultural lands, 8,616 cultivated lands for growing grain crops (one season), 7,756 cultivated lands for growing grain crops (two seasons), and 3,377 cultivated lands for other crops.

[0086] According to the spatial location of the samples, the GEE platform was used to calculate the median NDVI of each type of sample in 2019 to obtain the surface type NDVI time series sample data ( Figure 5 The Beijing-Tianjin-Hebei study area is large, and the temperature difference between the north and the south leads to different vegetation phenology such as crop planting and harvesting time, which shows different NDVI time series variation characteristics. For this reason, when extracting NDVI time series samples, we refer to the meteorological data of active accumulated temperature above 10°C, and use the active accumulated temperature of 3400°C, the temperature requirement node for crop planting, as the boundary to divide the Beijing-Tianjin-Hebei region into two districts, north and south. For 10 categories, the 2019 NDVI time series data standard samples were established in the two districts ( Figure 5 ). Figure 5 The NDVI temporal variation characteristics of each land cover type are shown. The NDVI temporal variation among different types shows obvious differences. Among them, the NDVI temporal variation of crops in cultivated land is most different from that of other land features. Therefore, the NDVI temporal variation can be used as a reference standard for land cover classification.

[0087] Cultivated land planting attribute extraction results

[0088] The total cultivated land area in the Beijing-Tianjin-Hebei region is 75,407 km 2 In 2019, the actual area of cultivated land planted with grain crops in the Beijing-Tianjin-Hebei region was 49,840 km2. 2 Of this total, the planting area for grain crops (single season) and grain crops (two seasons) accounted for 36% and 29% respectively, while other crop planting and greenhouse and other facility agricultural land accounted for 13% and 4% respectively. Non-crop land accounted for 18%, with gardens, fallow land, construction and development land, and ponds accounting for 7%, 7%, 2.7%, and 1.3% respectively. Among non-crop land uses, Zhangjiakou had the largest area of fallow land, accounting for 47% of the total fallow land in the Beijing-Tianjin-Hebei region. This is due to the region's fragile, semi-arid environment. To maintain soil fertility, local farmers often use fallow land in rotation.

[0089] Spatially, grain crops (two seasons) are primarily distributed in southern regions, including Baoding, Hengshui, Handan, Xingtai, and Cangzhou. The remaining regions are dominated by grain crops (one season), consistent with the geographical pattern of the 3400°C accumulated temperature line. Gardens occupying arable land primarily occur in the southeastern regions of Beijing and Shijiazhuang, leading to the problem of farmland being converted to non-grain crops. Furthermore, to protect the tillage layer and maintain soil fertility, the proportion of fallow and crop rotation on cultivated land has gradually increased in recent years, particularly in Zhangjiakou, a northern region of the study area where fallow land is predominant. These areas should be the focus of monitoring the conversion of cultivated land to non-agricultural and non-grain crops.

[0090] Figure 7The area of each type of cropping attribute in different municipalities is shown. Except for Beijing, the cultivated land in each municipality is mainly planted in one or two seasons, with the largest area, while Beijing has the largest garden area. Among the municipalities, Xingtai has the largest area of two-season crop planting, exceeding 3000km 2 Zhangjiakou has the largest crop planting area, exceeding 5,000 km2. 2 Baoding has the largest planting area of other crops, about 1,600 km 2 .

[0091] Accuracy verification

[0092] In order to better verify the proposed method for extracting cultivated land planting attributes, the present invention uses 246 field survey sample points and 1200 field sample points combined with remote sensing visual interpretation sample points to verify the cultivated land planting attribute results. The accuracy matrix shows that the field survey sample point set has higher verification accuracy, with an overall accuracy of 91% and a kappa coefficient of 0.87. The verification accuracy of the field sample point combined with remote sensing visual interpretation sample point set is slightly lower ( Figure 7 Validation results show that the proposed MSC method can effectively classify food crops (1 season), food crops (2 seasons), water areas, gardens, and construction land, with user accuracies exceeding 90%. However, the classification accuracy for fallow land, facility farmland, and other crops is slightly lower. In terms of user accuracy, gardens and food crops (2 seasons) have the highest accuracy. In terms of producer accuracy, water areas and construction land have the highest accuracy.

[0093] High-precision mapping of cultivated land planting attributes is the basis for monitoring the quantity and quality of cultivated land. Satellite remote sensing technology has the ability to observe the earth at multiple scales, multiple time phases, and multiple spectral bands, and can provide accurate and objective ground object information in space and time, providing an effective way to extract high-precision cultivated land planting attribute information at the regional scale. The present invention uses time-series remote sensing data and integrates unsupervised clustering, spatial overlay analysis, and time dynamic regularization to extract cultivated land planting attribute information. On the basis of existing land use data, the time-series NDVI change characteristics of the ground objects are used to quickly extract cultivated land planting attribute information, providing a new extraction strategy for actual cultivated land planting attribute monitoring. Through two verification methods of ground sampling and ground combined with remote sensing judgment, the overall accuracy is greater than 88%, and the extraction accuracy is relatively high.

[0094] Sentinel-2A / B satellite imagery boasts high temporal and spatial resolution. Compared to the more commonly used LandsatTM / OLI and MODIS time series data, Sentinel-2A / B data offer advantages such as shorter regression periods and higher temporal and spatial resolution. These data hold great potential for precise crop identification and crop attribute extraction at the regional scale. Compared to single-phase land cover classification, monthly fused NDVI can better characterize crop growth characteristics, significantly improving crop classification accuracy. By integrating this with existing land use data, it can address the challenges of class division and labeling in previous unsupervised classification, resulting in more accurate classification results.

[0095] The proposed method still exhibits minor misclassifications and omissions, with some instances of misclassification and misclassification of bare land, grassland, and buildings. Manual review and modification of the classification results can address the misclassification of roads, grassland, and forest belts attached to fields as crops. Further research can further explore the spatial information contained in Sentinel-2A / B imagery, such as texture and shape features, to improve crop classification accuracy.

[0096] In summary, this application proposes a technical method for remote sensing extraction of regional cultivated land planting attribute information based on 10m time series remote sensing data. This method utilizes the multi-temporal change characteristics of crop NDVI, based on existing land use survey data, combined with satellite remote sensing data clustering constraints, and selects typical category standard samples to quickly obtain a set of ground feature samples of planting attribute types, and through the integrated constraints of cultivated land plot scale (pattern), quickly extract cultivated land planting attribute information. The overall accuracy can reach more than 88%, which solves the problem of obtaining a large number of samples of cultivated land planting attributes in a large area and the rapid extraction of cultivated land planting attributes, and provides a new technical method for monitoring cultivated land planting attributes in my country.

[0097] It should be noted that, unless otherwise specified, the terms "first", "second", "third", etc. in the specification are only used to distinguish the various components, elements, steps, etc. in the specification, and are not used to indicate the logical relationship or sequential relationship between the various components, elements, steps, etc.

[0098] It is understood that although the present invention has been disclosed above with reference to preferred embodiments, the above embodiments are not intended to limit the present invention. For any person skilled in the art, without departing from the scope of the technical solution of the present invention, the technical content disclosed above can be used to make many possible changes and modifications to the technical solution of the present invention, or to modify it into an equivalent embodiment with equivalent changes. Therefore, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A multi-temporal remote sensing method for extracting cultivated land planting attributes at a plot scale, characterized in that: The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes includes: Step 1: Perform data collection and preprocessing; Step 2: Based on the collected data, perform 10m multi-temporal NDVI time series data construction; The third step: Based on the constructed 10m multi-temporal NDVI time series data, the 10m multi-temporal NDVI cluster analysis was performed using the multi-temporal variation characteristics of crop NDVI; Step 4: Based on the previous land use survey data and combined with the clustering constraints of satellite remote sensing data, a spatiotemporal standard sample set of surface cover types is generated, and a standard time series curve of NDVI of land features is constructed; Step 5: Using the standard time series curve of ground object NDVI, a dynamic time normalization algorithm is applied to classify the 10m multi-temporal NDVI clustering results. Based on the previous land use patches, the cultivated land plots are integrated and constrained to extract the cultivated land planting attribute information. In the fifth step, for the object to be classified T, the Wk values of the object and multiple standard types N are calculated according to the following formula. The standard type m with the minimum W is the object to be classified: γ(i,j) = ∥Yi – Sj∥; Where Yi is the NDVI of the object to be classified in month i (i = 1, 2, …, 12); Sj is the standard NDVI of day j (j = 1, 2, …, 365); Di = min[γ(i,j)],j=1,2,…,365; Where Di is the regular path, i=1,2,…12 Wk= ∑Di, i=1,2,…,12; The type of the object to be classified that has the smallest distance from the standard type is the type of the object to be classified: T∈m when Wm = min (Wk), k = 1,..,m,…,N.

2. The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes according to claim 1 is characterized in that: The data collected in the first step include multi-temporal satellite remote sensing data with a spatial resolution of 10m or above and existing land use survey data.

3. The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes according to claim 1 is characterized in that: The accuracy of the preliminary land use survey data described in the fourth step meets the data accuracy of 150,000 or above.

4. The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes according to claim 1, characterized in that: In the second step, the 10m multi-phase NDVI time series data are constructed using the monthly maximum, mean and / or median synthesis method. The temporal resolution of the data can cover the growing season of crops and depict the phenological changes of crop growth.

5. The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes according to claim 1, characterized in that: The preliminary land use survey data in the fourth step includes cultivated land planting attribute information.

6. The multi-temporal remote sensing extraction method for plot-scale cultivated land planting attributes according to claim 1, characterized in that: The surface cover types described in the fourth step include: forest land, grassland, bare land, water bodies, construction land, facility agricultural land, garden land, arable land for growing grain crops in one season, arable land for growing grain crops in two seasons, and arable land for other crops.

Images