Method for distinguishing crops by extracting yellow canopy information of crops

Abstract

The invention discloses a method for distinguishing crops by extracting yellow canopy information of the crops. The method is used for acquiring spatial distribution of at least two different crops growing in the same period in a specified geographic area, and comprises the following steps: step A, in satellite image data covering the specified geographic area, selecting a single-scene high-resolution No.6 WFV image in a time period in which only one crop has a yellow canopy, and preprocessing the single-scene high-resolution No.6 WFV image; step B, obtaining typical ground feature category positions in the single-scene high-resolution No.6 WFV image preprocessed in the step A, and obtaining spectral information of each ground feature; step C, according to the spectral information of each ground feature obtained in the step B, calculating a normalized yellow vegetation index value of each ground feature; and step D, constructing a learning model by using a computer algorithm, and obtaining spatial distribution data of different crops. According to the method provided by the invention, different crops growing in the same period can be accurately distinguished, so that high-precision identification and extraction of spatial distribution of different crops can be realized.

Description

Technical Field

[0001] The invention relates to a satellite remote sensing image processing method, in particular to a method for extracting yellow canopy information of crops by using GF-6 WFV (widefield-of-view) sensor image data to distinguish different crops. Background Art

[0002] By processing high-precision satellite remote sensing images, crops in the remote sensing images can be identified and information such as crop planting area and growth conditions can be obtained, which can provide key information for social and economic activities such as crop yield prediction, growth monitoring, and disaster assessment.

[0003] In the process of processing high-precision satellite remote sensing images, constructing vegetation index based on remote sensing images is an effective means to extract crop distribution information and has been widely used in crop identification research. Among the existing methods for extracting crop distribution information, the green characteristics of vegetation are the main identification basis for the construction of many vegetation indices. Among them, the Normalized Difference Vegetation Index (NDVI) is the most commonly used vegetation index. It uses near-infrared and red band information to capture the reflectivity of green vegetation as the green characteristics of vegetation for crop identification.

[0004] The development of new sensors has explored more discrete spectral channels in the 0.4-2.5μm spectral region, supporting the construction of more efficient vegetation indices. Currently, most studies use green canopy as a typical vegetation feature and focus on building vegetation indices that capture green features under different conditions, while fewer studies use other typical vegetation color features to build indices to achieve high-precision crop identification.

[0005] Many crops show yellowing in the middle and late stages of growth, which is manifested as yellow characteristic information superimposed on the green canopy, and can be used as a significant feature for observation and identification using remote sensing technology. For example, rapeseed presents a canopy morphology characterized by bright yellow flowers during the flowering period; the ears of corn at the top of the stalk are dark yellow compared to the leaves during the heading period; and the ears of rice show yellow characteristics during the maturity period. The Normalized Difference Yellow Index (NDYI) and the Normalized Rapeseed Flowering Index (NRFI) use green, blue, and short-wave infrared bands to construct vegetation indices based on the yellow characteristics of rapeseed flowers, realizing the identification of rapeseed crop types using the yellow characteristics of rapeseed during the flowering period.

[0006] The traditional vegetation index constructed based on the green canopy cannot effectively extract the yellow canopy characteristics. The current vegetation index constructed for the yellow canopy characteristics of crops cannot reflect the green canopy characteristics of vegetation. For example, the application of NDYI can only achieve the effective extraction of rapeseed flowers, but cannot effectively extract the winter wheat information of the same period as rapeseed. In addition, the application of the newly added band information of the Gaofen-6 satellite in crop identification is limited, and lacks effective combination with the key phenological characteristics of crops. Summary of the invention

[0007] The technical problem to be solved by the present invention is to provide a method for extracting yellow canopy information of crops to distinguish crops, so as to reduce or avoid the above-mentioned problems.

[0008] The present invention provides a method for extracting yellow canopy information of crops to distinguish crops, which is used to obtain the spatial distribution of at least two different crops growing in the same period in a specified geographical area, and comprises the following steps:

[0009] Step A, selecting a single-view Gaofen-6 WFV image in a time period in which only one crop has a yellow canopy from satellite image data covering the designated geographic area for preprocessing;

[0010] Step B, for the time period of the single-view GF-6 WFV image selected in step A, based on the existing ground object data, obtain the typical ground object category positions in the single-view GF-6 WFV image preprocessed in step A, and obtain the spectral information of each ground object.

[0011] Step C, calculating the normalized yellow vegetation index value of each object, that is, the NDYVI value, according to the spectral information of each object obtained in step B;

[0012] Step D, after calculating the NDYVI values ​​of various features, a learning model is constructed using a computer algorithm, and sample points for training and verification are provided through existing feature data to obtain spatial distribution data of different crops.

[0013] Preferably, in step C, the calculation process of the normalized yellow vegetation index value is as follows:

[0014] First, the yellow band and red edge I band of the GF-6 WFV image are selected to construct the yellow-red edge I reflectivity. The formula is as follows:

[0015]

[0016] In the above formula, ρ yellow and ρ red-edge1 They represent the reflectance of the yellow band and red edge I band in the GF-6 WFV image, respectively; It is the yellow-red edge I reflectivity, which represents the sum of the reflectivities of the yellow band and the red edge I band.

[0017] Secondly, the normalized standard deviation calculation formula is introduced, the formula is as follows:

[0018]

[0019] In the formula, ρ NIR Represents the reflectivity of the near-infrared band in the GF-6 WFV image.

[0020] Preferably, in step A, a single-view high-resolution WFV-6 image before sowing or emergence of crops, and a single-view high-resolution WFV-6 image after harvesting crops are further selected for preprocessing.

[0021] Preferably, in step A, the preprocessing includes: radiation correction, atmospheric correction, orthorectification and stitching and cropping

[0022] Preferably, in step B, the existing ground object data includes Sentinel / Landsat time-series image data, existing crop spatial distribution data, and field samples collected.

[0023] Preferably, in step D, the computer algorithm is a CART algorithm or a random forest algorithm.

[0024] The method provided by the present invention extracts the yellow canopy information of crops to distinguish them. It innovatively uses the reflection information of the Gaofen-6 satellite in the yellow band. Based on the significant characteristics of yellow canopy crops, it constructs a new vegetation index with high universality and strong effectiveness, which provides an effective means for targeted extraction of crop growth characteristics and high-precision remote sensing identification. It can accurately distinguish different crops growing in the same period, and realize high-precision identification and extraction of the spatial distribution of different crops. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The following drawings are intended only to illustrate and explain the present invention, and do not limit the scope of the present invention.

[0026] Figure 1 A schematic diagram of the technical principle of a method for extracting yellow canopy information of crops to distinguish crops according to a specific embodiment of the present invention;

[0027] Figure 2 for Figure 1 A schematic diagram of spectral information of objects in a specified geographical area corresponding to an embodiment of the invention;

[0028] Figure 3 for Figure 1 Schematic diagram comparing the extraction results of the method with some other methods;

[0029] Figure 4 It is a schematic diagram comparing the extraction results of the method according to another specific embodiment of the present invention and some other methods. DETAILED DESCRIPTION

[0030] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, specific embodiments of the present invention are now described with reference to the accompanying drawings.

[0031] Figure 1 FIG. 1 is a schematic diagram of the technical principle of a method for extracting yellow canopy information of crops to distinguish crops according to a specific embodiment of the present invention; now refer to FIG. Figure 1 The technical principle shown in the figure further illustrates the technical solution of the present invention. Specifically,

[0032] The present invention provides a method for extracting yellow canopy information of crops to distinguish crops, which is used to obtain the spatial distribution of at least two different crops growing in the same period in a specified geographical area, and comprises the following steps:

[0033] Step A: From the satellite image data covering the designated geographical area, a single-view Gaofen-6 WFV image in a time period in which only one crop has a yellow canopy is selected for preprocessing.

[0034] In a specific embodiment of the present invention, Qianjiang City, Hubei Province is used as the experimental area, and rapeseed and winter wheat grown in the same period are used as research objects to extract the distribution information of rapeseed and winter wheat. Qianjiang City, Hubei Province is located in the Jianghan Plain. Agricultural land accounts for a large proportion of the entire area, and there are almost no evergreen forests, which will affect the analysis of crops through satellite images. Since rapeseed is in the flowering period in late March and early April, and winter wheat is still in the filling period of the growth period during this period, only a single-scene high-resolution WFV image is needed to use the yellow canopy feature information of rapeseed during the flowering period to achieve crop differentiation. Specifically, a single-scene high-resolution WFV image of April 7, 2019 was selected for preprocessing. The preprocessing process includes: radiation correction, atmospheric correction, orthorectification, and stitching and cropping.

[0035] The preprocessing process can be implemented by tools such as the US ENVI software. Since it is a relatively mature existing technology, it will not be described in detail here.

[0036] Step B, for the time period of the single-view GF-6 WFV image selected in step A, based on the existing ground object data, obtain the typical ground object category positions in the single-view GF-6 WFV image preprocessed in step A, and obtain the spectral information of each ground object.

[0037] The existing ground object data comes from Sentinel / Landsat time-series image data, existing crop spatial distribution data, and field samples. Because step A selects a single-view GF-6 WFV image, it is necessary to use the existing ground object data to determine the position of each typical ground object in the single-view GF-6 WFV image, that is, the pixel with each typical ground object (the entire pixel contains all the same type of ground objects). In this way, the spectral distribution data of each typical ground object in the single-view GF-6 WFV image can be obtained. Figure 2 for Figure 1 Schematic diagram of ground object spectral information in a specified geographical area corresponding to the embodiment of FIG. Figure 2 As shown, by using the existing ground object data in 2019, the real positions (i.e., specific pixels) of the six ground objects, namely winter wheat, rapeseed, bare land, buildings, water bodies, and vegetation, can be selected from the single-view GF-6 WFV image of April 7, 2019 selected in step A, thereby obtaining the spectral data of the six ground objects, namely winter wheat in the filling period, rapeseed in the flowering period, bare land, buildings, water bodies, and vegetation.

[0038] Step C, calculating the normalized yellow vegetation index value of each object, that is, the NDYVI value, based on the spectral information of each object obtained in step B.

[0039] In the technical solution of the present invention, a Normalized Difference Yellow Vegetation Index (NDYVI) is designed to analyze a single pixel of a single-view GF-6 WFV image. The index is constructed by combining the yellow (0.59μm–0.63μm), red edge I (0.69μm–0.73μm) and near-infrared band (0.77μm–0.89μm) of the GF-6 WFV (wide field-of-view) sensor image data. The specific NDYVI construction process is as follows:

[0040] First, the yellow band and red edge I band of the GF-6 WFV image are selected to construct the yellow-red edge I reflectivity. The formula is as follows:

[0041]

[0042] In the above formula, ρ yellow and ρ red-edge1 They represent the reflectance of the yellow band and red edge I band in the GF-6 WFV image, respectively; It is the yellow-red edge I reflectivity, which represents the sum of the reflectivities of the yellow band and the red edge I band. It can highlight the spectral information displayed by the yellow canopy and further enhance the high reflectivity characteristics of yellow canopy crops in the yellow and red edge I bands.

[0043] Secondly, the normalized standard deviation calculation formula is introduced. Since non-vegetation types such as bare soil and buildings have higher yellow and red edge I reflectance, it is necessary to refer to the near-infrared band that has a higher sensitivity to vegetation to distinguish between vegetation and non-vegetation types. Construct the normalized yellow vegetation index, the formula is as follows:

[0044]

[0045] In the formula, ρ NIR Represents the reflectance of the near-infrared band in the GF-6 WFV image. Since crops with green canopies have higher or similar near-infrared reflectance, and It is much lower than crops with yellow canopy, so it has higher NDYVI. It has been verified that the NDYVI distribution values ​​of different landform types follow the following rule: crops (green canopy) > crops (yellow canopy) > others.

[0046] Specifically, for the pixels of each typical landform determined in step B, according to the spectral data of each pixel, the NDYVI value of each pixel can be calculated according to the above steps, that is, the normalized yellow vegetation index values ​​of the six pixels corresponding to the six landforms of winter wheat in the filling period, rapeseed in the flowering period, bare land, buildings, water bodies, and vegetation can be calculated respectively.

[0047] Step D, after calculating the NDYVI values ​​of various features, a learning model is constructed using a computer algorithm, and sample points for training and verification are provided through existing feature data to obtain spatial distribution data of different crops.

[0048] As mentioned above, after obtaining the benchmark NDYVI values ​​of various features through step C, a learning model can be constructed using a computer algorithm, such as a CART (Classification And Regression Tree) algorithm or a random forest algorithm or other mature algorithms that can construct a learning model, and then the existing feature data is used to provide sample points for training and verification. Usually, 70% of the total sample points are used for computer algorithm training, and the other 30% are used for computer verification. In this way, the computer algorithm can calculate the NDYVI threshold data for judging each pixel. For example, Figure 1 and Figure 2In the example shown, which uses Qianjiang City in Hubei Province as the experimental area and rapeseed and winter wheat grown in the same period as the research objects, the inventor team provided 1,374 sample points through field surveys and existing ground feature data collected by Sentinel / Landsat time series data, which corresponded to 1,374 pixels in a single-view GF-6 WFV image, of which 961 were used for CART training and the remaining 413 were used for CART verification. A threshold decision tree was then constructed based on the CART algorithm to obtain NDYVI threshold data for judging each feature, which was then applied to the entire single-view GF-6 WFV image, so that the spatial distribution of rapeseed and winter wheat can be extracted from the perspective of the pixel using the NDYVI data of each pixel. It has been verified in practice that the overall accuracy of extracting the classification results of rapeseed and winter wheat based on the method of the present invention reached 85.32%, and the Kappa coefficient reached 0.80.

[0049] The inventor team used five existing commonly used vegetation indices: normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), ratio vegetation index (RVI), difference vegetation index (DVI), and normalized yellowness index (NDYI) to Figure 1 and Figure 2 The same image in the embodiment shown uses the same sample data and classification method to extract the spatial distribution of rapeseed and winter wheat, and compares it with the extraction result of NDYVI of the present invention. The accuracy comparison results are shown in the following table:

[0050]

[0051] Figure 3 for Figure 1 Schematic diagram comparing the extraction results of the method with some other methods; Figure 3 The Sentinel-2 satellite image data of the same area shown in the figure is only used as a comparison image to compare the accuracy of each extraction result.

[0052] See above table and Figure 3As shown, the classification result of the method provided by the present invention not only has the highest accuracy, but also has good performance in spatial distribution. The spatial distribution of rapeseed and winter wheat extracted by NDYVI has high discrimination, and the boundary information is relatively complete. In addition, the separability of winter wheat, rapeseed and other vegetation and landform types is significantly improved. Among the contrasting vegetation indices, although NDYI designed for rapeseed has good performance in extracting rapeseed, it is impossible to distinguish winter wheat from other landform types by relying solely on this index, resulting in a classification accuracy of less than 60%. Traditional vegetation indices such as NDVI and EVI can distinguish between crops and other landform types, but the discrimination between winter wheat and rapeseed is relatively low. This shows that the new normalized yellow vegetation index constructed by the present invention can not only efficiently distinguish between crops with yellow canopies and green canopies, but can also effectively identify crops, vegetation and other landform types, which helps to achieve high-precision crop classification.

[0053] In another specific embodiment of the present invention, Kedong County, Heilongjiang Province is used as an experimental area, and corn and soybeans grown in the same period are used as research objects. By using the normalized yellow vegetation index NDYVI designed by the present invention, the distribution information of corn and soybeans is extracted according to the method provided by the present invention. Corn is in the heading stage in late July, showing the yellow-brown ear characteristics at the top of the plant. Soybeans in the same period are in the pod-setting stage, showing a single green canopy state. Kedong County, Heilongjiang Province is located in the northern part of Heilongjiang Province, in the transition zone between the Lesser Khingan Mountains and the Songnen Plain, with undulating terrain, so there are a large number of forests, which are evergreen interference factors in the life cycle of crops. In order to overcome the influence of these interference factors, considering that crops in Northeast China are all one season a year, therefore, the single-view high-resolution No. 6 WFV images of the same area before crop sowing or before emergence and after harvest can be extracted as backup, so as to distinguish interference factors such as forests. According to the method of the present invention, a single-view high-resolution No. 6 WFV image on July 21, 2019 was selected for extracting corn heading features using NDYVI.

[0054] Because there are forests in Kedong County, Heilongjiang Province, which will affect the analysis of crops through satellite images, it is necessary to pre-process the single-view GF-6 WFV images before sowing or emergence of crops (specifically, images before May 1, 2019), the single-view GF-6 WFV images after harvesting crops (specifically, images around September 30, 2019), and the single-view GF-6 WFV images selected on July 21, 2019 mentioned above.

[0055] After that, the feature data of Kedong County, Heilongjiang Province in 2019 were selected from existing feature data such as Sentinel / Landsat time series data, published soybean and corn distribution data sets, and field surveys. After that, the positions of each typical feature, that is, the pixels, were determined in the preprocessed single-view GF-6 WFV image on July 21, 2019, so as to obtain the spectral distribution data of each typical feature in the single-view GF-6 WFV image. Specifically, the real positions of the four types of features, corn, soybeans, forests, and buildings at the heading stage, can be obtained from the single-view GF-6 WFV image on July 21, 2019, so as to obtain the spectral data of these four types of features.

[0056] For the single-view GF-6 WFV image on May 1, 2019 and the single-view GF-6 WFV image on September 30, 2019, since there is no crop information, it is only necessary to determine the pixels of each typical object in the existing object data, and thus determine the spectral distribution data of the pixels of objects such as buildings and forests that will become typical interference factors.

[0057] Next, we can calculate the NDYVI values ​​of various objects in these three phases of single-view GF-6 WFV images.

[0058] Then, according to the existing ground object data, 1580 sample points were provided (1106 of which were used for machine training and the remaining 474 were used for machine verification). Based on the random forest machine learning model, three images of corn heading period (July 21, 2019), before emergence (May 1, 2019) and after harvest (September 30, 2019) were applied. The participation of two images before emergence (May 1, 2019) and after harvest (September 30, 2019) can help the computer to effectively remove the pixels of interference factors such as forests in the image of corn heading period (July 21, 2019) and extract the spatial distribution of corn and soybeans. It has been verified that the overall accuracy of the method based on the present invention reaches 96.21%, and the Kappa coefficient reaches 0.94.

[0059] The inventor team applied five existing commonly used vegetation indices: normalized difference vegetation index (NDVI), enhanced vegetation index (EVI), ratio vegetation index (RVI), difference vegetation index (DVI), and normalized yellow index (NDYI). For the same image in this embodiment, the same sample data and classification method were used to extract the spatial distribution of corn and soybeans, and compared with the extraction results of NDYVI of the present invention. The accuracy comparison results are shown in the following table.

[0060]

[0061] Figure 4FIG. 1 is a schematic diagram comparing the extraction results of a method according to another specific embodiment of the present invention with those of some other methods, see Figure 4 As shown in the above table, the classification results of the method provided by the present invention not only have the highest accuracy, but also have good performance in spatial distribution. Compared with other vegetation indices, the distribution results of corn and soybeans extracted by NDYVI are more complete, with clear boundaries and fewer salt and pepper noise pixels inside the field. In addition, the separability of corn and soybeans from other vegetation and landform types is significantly improved, which is manifested in the extraction of clear field contours with other vegetation as boundaries. Among the contrasting vegetation indices, NDYI designed for rapeseed cannot be effectively applied to the distribution extraction of corn and soybeans, resulting in the lowest extraction accuracy. Traditional vegetation indices such as NDVI and EVI can distinguish between crops and other landform types, but the extraction results have more salt and pepper noise pixels, and the distribution of other vegetation types cannot be efficiently extracted. The results of this case show that the new normalized yellow vegetation index NDYVI constructed by the present invention can not only effectively extract the spatial distribution of rapeseed and winter wheat with simultaneous growth, but can also be applied to high-precision identification of corn and soybeans growing at the same time. Compared with previous vegetation indices, the present invention innovatively applies the reflection information of the Gaofen-6 satellite in the yellow band, and based on the significant characteristics of yellow canopy crops, constructs a new vegetation index with high universality and strong effectiveness, providing an effective means for targeted extraction of crop growth characteristics and realizing high-precision remote sensing identification.

[0062] The method for extracting yellow canopy information of crops to distinguish crops provided by the present invention can accurately distinguish different crops growing in the same period, thereby realizing high-precision identification and extraction of the spatial distribution of different crops.

[0063] Those skilled in the art should understand that although the present invention is described in the form of multiple embodiments, not every embodiment includes only one independent technical solution. Such description in the specification is only for the sake of clarity. Those skilled in the art should understand the specification as a whole and regard the technical solutions involved in each embodiment as being able to be combined into different embodiments to understand the scope of protection of the present invention.

[0064] The above description is only an illustrative embodiment of the present invention and is not intended to limit the scope of the present invention. Any equivalent changes, modifications and combinations made by any person skilled in the art without departing from the concept and principle of the present invention shall fall within the scope of protection of the present invention.

Claims

1. A method for extracting yellow canopy information of crops to distinguish crops, which is used to obtain the spatial distribution of at least two different crops growing in the same period in a specified geographical area, and comprises the following steps: Step A, selecting a single-view Gaofen-6 WFV image in a time period in which only one crop has a yellow canopy from satellite image data covering the designated geographic area for preprocessing; Step B, for the time period of the single-view GF-6 WFV image selected in step A, based on the existing ground object data, obtain the typical ground object category position in the single-view GF-6 WFV image preprocessed in step A, and obtain the spectral information of each ground object; Step C, calculating the normalized yellow vegetation index value of each object, that is, the NDYVI value, according to the spectral information of each object obtained in step B; Step D, after calculating the NDYVI values ​​of various features, a learning model is constructed using a computer algorithm, and sample points for training and verification are provided through existing feature data to obtain spatial distribution data of different crops.

2. The method according to claim 1, characterized in that In step C, the calculation process of the normalized yellow vegetation index value is as follows: First, the yellow band and red edge I band of the GF-6 WFV image are selected to construct the yellow-red edge I reflectivity. The formula is as follows: In the above formula, ρ yellow and ρ red-edge1 They represent the reflectance of the yellow band and red edge I band in the GF-6 WFV image, respectively; is the yellow-red edge I reflectivity, representing the sum of the reflectivity of the yellow band and the red edge I band; Secondly, the normalized standard deviation calculation formula is introduced, the formula is as follows: In the formula, ρ NIR Represents the reflectivity of the near-infrared band in the GF-6 WFV image.

3. The method according to claim 1, characterized in that In step A, a single-view high-resolution WFV-6 image before crop sowing or emergence, and a single-view high-resolution WFV-6 image after crop harvest are further selected for preprocessing.

4. The method according to claim 1, characterized in that In step A, the preprocessing includes: radiation correction, atmospheric correction, orthorectification, and stitching and cropping.

5. The method according to claim 1, characterized in that In step B, the existing ground object data includes Sentinel / Landsat time-series image data, existing crop spatial distribution data, and field samples.

6. The method according to claim 1, characterized in that In step D, the computer algorithm is a CART algorithm or a random forest algorithm.

Images