A remote sensing automatic extraction method for winter wheat planting structure

Through the vegetation index calculation and S-G filtering technology of remote sensing images, combined with dynamic threshold method, high-precision automatic extraction of winter wheat planting structure is achieved, solving the problems of low accuracy and sample database dependence in the existing technology, and improving the universality of the method and credibility.

CN117830860BActive Publication Date: 2024-12-06JIANGSU BASIC GEOGRAPHIC INFORMATION CENT
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410252941.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-03-06
Publication Date
2024-12-06
Estimated Expiration
2044-03-06

AI Technical Summary

Technical Problem

The existing crop classification methods have low accuracy, low efficiency, poor universality, and are difficult to meet the needs of precise management of seasonal crop production. Especially in the automatic extraction of winter wheat planting structures, there is a lack of high quality that does not rely on sample databases and manual experience. Accuracy, migratory method.

Method used

By obtaining remote sensing images, calculating the vegetation index of ten-scale, and reconstructing the data set using S-G filtering, determining the filter parameters with the fitting effect index, and using dynamic threshold method to extract the phenological vegetation index data of winter wheat to realize automatic extraction of winter wheat planting structure , and does not require sample database support, reducing labor costs.

Benefits of technology

It significantly improves the accuracy of winter wheat planting structure extraction, achieves high reliability and migration of the model, reduces labor costs, and is suitable for crop monitoring in different regions and times.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117830860B_ABST
    Figure CN117830860B_ABST
Patent Text Reader

Abstract

The present invention provides a remote sensing automatic extraction method for winter wheat planting structure, which obtains remote sensing images, and calculates the vegetation index at the ten-day scale based on the remote sensing images to synthesize the initial ten-day scale long-time series vegetation index data set, and uses the fitting effect index as an evaluation index to reconstruct the data set using S-G filtering; winter wheat is extracted using the reconstructed ten-day scale long-time series vegetation index data set, and when winter wheat is extracted, vegetation index data of each phenological period is extracted according to the winter wheat phenological period, and a dynamic threshold is set based on the vegetation index data of different phenological periods to achieve winter wheat extraction. The method of the present invention fully considers the winter wheat phenological period, and can significantly improve the accuracy of winter wheat planting structure extraction. In addition, the present invention does not require the support of a sample database, does not rely on manual experience, greatly reduces labor costs, and has a high model credibility, transferability, and strong universality.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention belongs to the technical field of remote sensing monitoring of natural resources, and in particular relates to a remote sensing automatic extraction method for winter wheat planting structure. Background Art

[0002] The ever-changing global situation has led to an increasing demand for accurate and timely spatial distribution of food crop types. Accurate crop spatial distribution and planting area information is the basic base map for regional-scale crop growth monitoring and productivity forecasting, and is crucial for food security risk assessment. At present, due to factors such as fragmented fields and complex planting methods, existing crop classification and mapping methods generally have problems such as low accuracy, low efficiency, and poor universality. Large-scale, high-precision crop classification products are still scarce. In addition, the public release time of most existing crop classification products is seriously delayed, making it difficult to meet the needs of precise management of seasonal crop production.

[0003] As a cutting-edge technology of modern information technology, remote sensing has the advantages of large coverage, short detection cycle, strong timeliness and low cost, making it possible to quickly and accurately extract crop planting information over a large area, and has become an important tool for producing land cover products from local to global scales. With the development of remote sensing technology, high spatial resolution remote sensing images have gradually become the main data source for extracting spatial distribution information of crops. The use of remote sensing technology to extract crop planting area is mainly based on the spectral characteristics, phenological characteristics, texture characteristics, etc. of crops, relying on single-phase remote sensing image recognition or multi-phase remote sensing change monitoring. There are mainly three methods:

[0004] 1) Visual interpretation method based on single-phase high-resolution images. This method mainly uses single-phase remote sensing images, based on single-period remote sensing images of the key growth period of crops, and adopts different classification methods, such as manual visual interpretation, unsupervised classification, decision tree classification support vector machine, etc., or extracts vegetation index, leaf area index, etc. through single-period remote sensing images, and sets thresholds to extract crop information. Due to the influence of the phenomenon of "same object, different spectrum, different objects, same spectrum" of ground objects, this method will cause a lot of mixed classification in the identification of crops in the same period, and the recognition accuracy is difficult to guarantee.

[0005] 2) Crop mapping technology that relies on prior knowledge. This method uses existing crop type map products of various crop planting structures or similar crop planting structures published worldwide as training data to generate prior knowledge and then complete the mapping of crop planting structures in the study area. This method is affected by the spatial distribution range of existing crop planting structures. In addition, since most of the existing results are based on sample generation, and since the samples have regional characteristics, migration to different places or migration between different times in the same region will affect the accuracy of the results.

[0006] 3) Crop mapping techniques through training data or classification model transfer. This is achieved by directly migrating historical training data or transferring classification models trained on existing training data using historical field survey data. Since mapping accuracy is easily affected by intra- and inter-class variations in reflectance and phenology, model transfer methods still require the original ground truth to train supervised models and determine robust classification features.

[0007] In summary, it is urgent to develop a model for automatic extraction of winter wheat planting structure that does not require the support of a sample database, does not rely on manual experience, greatly reduces labor costs, has high model credibility, is transferable, and has strong universality. Single-phase remote sensing data can only obtain the spectral characteristics of surface crops in a certain period, and crops have different spectral characteristics at different growth stages. Since time series satellite observations can capture the detailed phenological and spectral characteristics of crop types, they reflect the growth of crops at different growth stages and avoid the phenomenon of different objects with the same spectrum. Phenological-based methods are generally considered to be an alternative to making crop type maps without using training data. Summary of the invention

[0008] The purpose of the present invention is to provide a remote sensing automatic extraction method for winter wheat planting structure. When used for winter wheat planting structure extraction, the winter wheat phenological period is fully considered, and the vegetation index data of the key phenological period is screened for extraction model construction, which significantly improves the accuracy of winter wheat planting structure extraction. At the same time, the present invention does not require the support of a sample database, does not rely on manual experience, and can greatly reduce labor costs. In addition, the model has high credibility, is transferable, and has strong universality.

[0009] In order to achieve the above technical objectives and the above technical effects, the present invention adopts the following technical solutions:

[0010] A remote sensing automatic extraction method for winter wheat planting structure, comprising:

[0011] Obtain remote sensing images and calculate the vegetation index at the ten-day scale based on the remote sensing images;

[0012] The constructed ten-day-scale vegetation index is used to synthesize the initial ten-day-scale long-time series vegetation index data set, and the data set is reconstructed using SG filtering. When reconstructing the SG filtering, the fitting effect index is used as an evaluation index to fit different SG filtering parameter combinations, and the filtering parameter combination corresponding to the minimum value of the fitting effect index is taken as the SG filtering parameter; the fitting effect index is constructed based on the deviation between the reconstructed vegetation index value and the initial vegetation index value;

[0013] Winter wheat is extracted using the reconstructed ten-day long-term series vegetation index dataset. When extracting winter wheat, the vegetation index data of each phenological period is extracted according to the winter wheat phenological period. Dynamic thresholds are set based on the vegetation index data of different phenological periods to achieve winter wheat extraction.

[0014] As a preferred implementation, the vegetation index is the Normalized Difference Vegetation Index NDVI.

[0015] As a preferred implementation, the calculated vegetation index is used to calculate the daily vegetation index using the median method, and then the decadal vegetation index is calculated using the maximum value synthesis method.

[0016] As a preferred implementation, the fitting effect index is calculated as follows:

[0017]

[0018]

[0019] In the formula, F k is the fitting effect index; n is the total number of data in the data set; is the vegetation index value corresponding to the i-th data of the k-th fitting time series; is the initial vegetation index value corresponding to the i-th data; W i is the weight coefficient of the i-th data; for The absolute value of the difference between max is d i The maximum value of .

[0020] As a preferred implementation, the least square method is used to fit different filter parameter combinations; the filter parameters include filter window size and fitting polynomial coefficients.

[0021] As a preferred embodiment, the winter wheat extraction model is as follows:

[0022]

[0023] In the formula, NDVI B NDVI C NDVI S They are the vegetation index values ​​corresponding to the sowing period, maturity period and harvesting period of winter wheat respectively; threshold_1, threshold_2 and threshold_3 are the corresponding extraction thresholds, which are determined by using the single scene data images of the sowing period, maturity period and harvesting period respectively, and are obtained by the dynamic threshold method.

[0024] As a preferred implementation, the vegetation index values ​​corresponding to the sowing period, maturity period, and harvest period of winter wheat are determined as follows:

[0025] The time point when NDVI reaches the maximum value is taken as the beginning time of winter wheat heading period; in the rising stage of NDVI, the time point when NDVI reaches NDVIlim The time point is the beginning of the greening period of winter wheat; when the NDVI reaches The time point is the beginning of the maturity period of winter wheat, NDVI lim Based on the following calculation:

[0026]

[0027] In the formula, NDVI lim is a dynamic threshold; NDVI max NDVI min are the maximum and minimum values ​​of NDVI in the growing season, respectively; C is the threshold coefficient, which is a self-set percentage;

[0028] The ten-day data of the beginning of heading period is selected as the NDVI value of maturity period; the minimum value of all ten-day data before the beginning of green period is selected as the NDVI value of sowing period; the minimum value of all ten-day data after the beginning of maturity period is selected as the NDVI value of harvesting period.

[0029] As a preferred implementation, the remote sensing image uses Sentinel-L1C product.

[0030] As a preferred implementation, the method further includes obtaining cultivated land map patches of the study area, and using the cultivated land map patches of the study area to clip the extracted winter wheat planting structure distribution to obtain the winter wheat planting structure distribution within the cultivated land range.

[0031] As a preferred implementation, the spatial resolution of the cultivated land patches is better than the spatial resolution of remote sensing images.

[0032] The present invention has the following beneficial effects:

[0033] (1) The present invention introduces the "fitting effect index" to automatically obtain the optimal SG filter fitting parameters and realize the fully automatic reconstruction of the long-term vegetation index data set without the need for human intervention.

[0034] (2) The present invention comprehensively considers the changing characteristics of the long-term vegetation index of winter wheat, and uses the dynamic threshold method to automatically extract the key phenological periods in the growth process of winter wheat, thereby assisting in the selection of time-series images of key phenological periods in the later stage.

[0035] (3) The present invention fully considers the phenological period of winter wheat and can significantly improve the accuracy of winter wheat planting structure extraction. Specifically, the three key phenological points of the winter wheat growth phenological period, namely "sowing period", "maturity period" and "harvest period", are used to introduce the idea of ​​hierarchical classification. The NDVI value of the maturity period is used to classify the study area into winter crop planting areas and non-winter crop planting areas by using the threshold method. For the winter crop planting areas, the early and late growth stages of winter wheat are modeled separately, and the winter wheat planting structure is extracted by using the dynamic threshold method.

[0036] (4) The present invention does not require the support of a sample database and does not rely on manual experience, which greatly reduces labor costs. The model is highly credible, transferable, and has strong universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 This is a schematic diagram of the winter wheat planting structure extraction results in Example 1. DETAILED DESCRIPTION

[0038] The technical solution of the present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0039] Example 1

[0040] The remote sensing automatic extraction method of winter wheat planting structure of the present invention specifically comprises the following steps:

[0041] S1: Acquire Sentinel-L1C data of the study area, obtain complete remote sensing images of the study area and process the remote sensing images, calculate the Normalized Difference Vegetation Index (NDVI) data on a daily and ten-day scale; obtain the cultivated land patches of the study area; the spatial resolution of the cultivated land patches of the study area obtained is better than the optical remote sensing images of Sentinel-L1C data; the specific steps include the following:

[0042] S11: According to the location and scope of the study area, collect and obtain high-resolution Sentinel-L1C data for subsequent NDVI calculation; obtain the cultivated land patches of the study area for the final spatial distribution mask of the winter wheat planting structure; the spatial resolution of the cultivated land patches is better than the optical remote sensing image of the Sentinel-L1C data, which is used for the final manual visual interpretation; the cultivated land patch data source used in the present invention comes from the annual routine land change survey tasks issued nationwide by the natural resources system, or the user's own cultivated land patch data can also be used.

[0043] S12: Perform atmospheric correction on the acquired Sentinel original top atmosphere apparent reflectance data (L1C) to obtain the surface reflectance L2A level data;

[0044] S13: Calculate NDVI data using the Sentinel-L2A data generated in step S12;

[0045] The NDVI calculation formula is as follows:

[0046] (1)

[0047] In the formula, They represent the surface reflectivity of the near-infrared band (NIR) and the red band (RED) respectively.

[0048] S13: Based on the NDVI data calculated in S12, the daily NDVI data is calculated using the median method, and the ten-day NDVI data is calculated using the maximum value synthesis method;

[0049] The median method NDVI calculation formula is as follows:

[0050] (2)

[0051] In the formula, NDVI i is the NDVI value of the i-th day, NDVI ij is the NDVI value of the j-th image on the i-th day.

[0052] The calculation formula of NDVI by the maximum value synthesis method is as follows:

[0053] (3)

[0054] In the formula, NDVI i is the NDVI value of the ith decade, NDVI ij is the daily NDVI value in the i-th decade.

[0055] S14: Use smoothing filtering algorithm to fill in the pixels without values ​​in the ten-day-scale NDVI data.

[0056] S2: Use the constructed ten-day-scale NDVI data to construct a long-term vegetation index dataset, and use the SG (Savitzky-Golay) filtering algorithm to reconstruct the long-term NDVI vegetation index dataset; specifically, the following steps are included:

[0057] S21: Use the calculated ten-day scale NDVI data to synthesize the initial ten-day scale long-term series vegetation index dataset;

[0058] S22: Using the fitting effect index F k As an evaluation index, the least squares fitting method is used to best fit the combination of the filter window size (m) and the fitting polynomial coefficient (d) in the SG filter, and the optimal fitting parameters are automatically selected to take the fitting effect index Fk The combination of the filter window size (m) and the fitting polynomial coefficient (d) corresponding to the minimum value is the final SG filter parameter;

[0059] Fitting effect index F k The calculation formula is as follows:

[0060] (4)

[0061] In the formula, n is the total number of data in the data set; is the NDVI value corresponding to the i-th data of the k-th fitting time series; is the initial NDVI value corresponding to the i-th data; W i is the weight coefficient of the i-th data.

[0062] Weight coefficient The calculation formula is as follows:

[0063] (5)

[0064] In the formula, is the NDVI value corresponding to the i-th data of the k-th fitting time series; is the initial NDVI value corresponding to the i-th data; ,for The absolute value of the difference between max is d i The maximum value of .

[0065] S23: Combine the interpolation coefficients (m, d) and weight coefficients ( ), use the SG filtering algorithm to interpolate the long-term vegetation index dataset at the ten-day scale;

[0066] (6)

[0067] In the formula, is the NDVI value corresponding to the i-th data of the k-th fitting time series; is the initial NDVI value corresponding to the i-th data; m is the window size.

[0068] S24: Reconstruct the long-term vegetation index dataset using the following formula;

[0069] (6)

[0070] In the formula, is the NDVI value of the i-th data in the final dataset, is the NDVI value corresponding to the i-th data of the k-th fitting time series; is the initial NDVI value corresponding to the i-th data.

[0071] S3: Extract the phenological period of winter wheat crops using the reconstructed long-term NDVI vegetation index dataset; specifically, the following steps are included:

[0072] S31: Determine the threshold for extracting winter wheat phenological period;

[0073] (7)

[0074] In the formula, NDVI lim is a dynamic threshold; NDVI max NDVI min are the maximum and minimum values ​​of NDVI in the growing season respectively; C is the threshold coefficient, which is generally 20%. The threshold is calculated pixel by pixel, that is, one threshold for each pixel.

[0075] S32: Determine the key growth period in the winter wheat phenological period: take the time point when NDVI reaches the maximum value as the start time of winter wheat heading period; in the NDVI rising stage, take the time point when NDVI reaches NDVI lim The time point is the beginning of the greening period of winter wheat; when the NDVI reaches NDVI lim The time point is the beginning of the maturity period of winter wheat;

[0076] S4: Screen the NDVI data of key points in the phenological period to construct a winter wheat planting structure extraction model and complete the extraction of winter wheat planting structure;

[0077] S41: Select the ten-day data of the beginning time of the heading period as the NDVI value of the maturity period (NDVI C , which is equivalent to the aforementioned NDVI max ); select the minimum value of all the ten-day data before the start of the green period as the NDVI value of the sowing period (NDVI B ); select the minimum value of all the ten-day data after the beginning of the maturity period as the NDVI value at the harvest period (NDVI S );

[0078] S42: Construct a winter wheat planting structure extraction model using NDVI at the sowing, maturity and harvesting stages of winter wheat;

[0079] (8)

[0080] In the formula, NDVI B NDVI C NDVI Sare the NDVI values ​​corresponding to the sowing period, maturity period, and harvest period of winter wheat; threshold_1, threshold_2, and threshold_3 are the corresponding thresholds, respectively using the sowing period (NDVI B )、Maturity(NDVI C ), harvest period (NDVI S ) is determined by a single scene data image, and the dynamic threshold method is used when it is used. The threshold is calculated pixel by pixel, that is, one threshold for each pixel.

[0081] In this formula, the NDVI value at maturity is first used (NDVI C ) to classify the study area, NDVI C >threshold_1 is for winter crop planting areas; for winter crop planting areas, the early growth period of winter wheat (NDVI C -NDVI B ) and late growth period (NDVI C -NDVI S ) were modeled separately to extract the winter wheat planting structure.

[0082] S43: Using the constructed long-term NDVI dataset and winter wheat planting structure extraction model, complete the winter wheat planting structure extraction in the study area;

[0083] S5: Use the cultivated land data of the study area to clip the extracted winter wheat planting structure distribution to obtain the winter wheat planting structure distribution within the cultivated land range;

[0084] Specifically, the extracted winter wheat planting structure data were masked using the cultivated land patches of the study area to obtain the spatial distribution of winter wheat planting structure within the cultivated land area;

[0085] S6: Use manual visual inspection methods to check the results and form the final winter wheat planting structure distribution results.

[0086] Specifically, high-resolution optical remote sensing images were used to eliminate the erroneously extracted water bodies and artificial structures in the spatial distribution of winter wheat planting structure within the cultivated land through manual visual interpretation, so as to form the final results of the spatial distribution of winter wheat planting structure in the study area, such as Figure 1 .

Claims

1. A remote sensing automatic extraction method for winter wheat planting structure, characterized in that: include: Acquire remote sensing images, and calculate the vegetation index at a ten-day scale based on the remote sensing images; the remote sensing images are Sentinel-L1C products; The constructed ten-day-scale vegetation index is used to synthesize the initial ten-day-scale long-time series vegetation index data set, and the data set is reconstructed using SG filtering. When reconstructing the SG filtering, the fitting effect index is used as an evaluation index, and the least square method is used to fit different SG filtering parameter combinations, and the filtering parameter combination corresponding to the minimum value of the fitting effect index is taken as the SG filtering parameter; the filtering parameters include the filter window size and the fitting polynomial coefficient; the fitting effect index is constructed based on the deviation between the reconstructed vegetation index value and the initial vegetation index value; Winter wheat was extracted using the reconstructed ten-day long-term series vegetation index dataset. When extracting winter wheat, the vegetation index data of each phenological period was extracted according to the winter wheat phenological period. A dynamic threshold was set based on the vegetation index data of different phenological periods to achieve winter wheat extraction. The winter wheat extraction model is as follows: ; In the formula, NDVI B , NDVI C , NDVI S They are the vegetation index values ​​corresponding to the sowing period, maturity period, and harvest period of winter wheat; threshold _1, threshold _2. threshold _3 is the corresponding extraction threshold, which is determined by using the single scene data image of the sowing period, maturity period and harvest period respectively, and is obtained by using the dynamic threshold method; the threshold is calculated pixel by pixel; The vegetation index values ​​corresponding to the sowing period, maturity period and harvest period of winter wheat are determined as follows: The time point when NDVI reaches the maximum value is taken as the beginning time of winter wheat heading period; in the rising stage of NDVI, the time point when NDVI reaches NDVI lim The time point is the beginning of the greening period of winter wheat; when the NDVI reaches NDVI lim The time point is the beginning of the maturity period of winter wheat. NDVI lim Based on the following calculation: ; In the formula, NDVI lim is the dynamic threshold; NDVI max , NDVI min are the maximum and minimum values ​​of NDVI in the growing season respectively; C is the threshold coefficient, which is a self-set percentage; the threshold is calculated pixel by pixel; The ten-day data of the beginning of heading period is selected as the NDVI value of maturity period; the minimum value of all ten-day data before the beginning of green period is selected as the NDVI value of sowing period; the minimum value of all ten-day data after the beginning of maturity period is selected as the NDVI value of harvesting period.

2. The method according to claim 1, characterized in that: The vegetation index is the Normalized Difference Vegetation Index (NDVI).

3. The method according to claim 1, characterized in that The calculated vegetation index is used to calculate the daily vegetation index using the median method, and then the ten-day scale vegetation index is calculated using the maximum synthesis method.

4. The method according to claim 1, characterized in that: The calculation method of the fitting effect index is as follows: ; ; In the formula, F k is the fitting effect index; n is the total number of data in the dataset; For the k The first fitting time series i The vegetation index value corresponding to the data; is the initial vegetation index value corresponding to the i-th data; W i For the i The weight coefficient of each data; for The absolute value of the difference between d max for d i The maximum value of .

5. The method according to claim 1, characterized in that It also includes obtaining the cultivated land map patches of the study area, using the cultivated land map patches of the study area to clip the extracted winter wheat planting structure distribution, and obtaining the winter wheat planting structure distribution within the cultivated land area.

6. The method according to claim 5, characterized in that The spatial resolution of the cultivated land patches is better than that of remote sensing images.