Anthocyanidin reflection vegetation index based on enhanced near-infrared and blue wave bands and method for extracting anthocyanin reflection vegetation index in farmland protection forest

Abstract

The invention relates to the field of remote sensing mapping, and discloses a farmland protection forest extraction method based on an anthocyanin reflection vegetation index of an enhanced near-infrared and blue wave band, and the method comprises the steps: 1, obtaining images, and splicing the obtained images through employing a tool of embedding to a new grid in ArcGIS software; step 2, calculating a vegetation index, and providing RMARI-NIR and RMARI-RGB indexes suitable for different scenes; 3, making a farmland protection forest label data set; and step 4, uploading a label data set and a research area vector range on a Google Earth Engine platform, dividing a training set and test set proportion, calculating a Kappa coefficient and total precision as evaluation indexes, using the evaluation indexes as a verification method of innovation feasibility, carrying out farmland protection forest identification training classification through a random forest (RF) model and additionally fusing a characteristic wave band and an index wave band, and carrying out farmland protection forest identification training classification. The classification result and the field boundary range are overlapped, mask operation is applied to extract and obtain the distribution position information of the farmland protection forests, the spectral characteristics of the farmland protection forests are highlighted according to the extraction requirements, the distinguishing degree between the farmland protection forests and other ground objects is enhanced, and the method is suitable for generalization in various environments.

Description

Technical Field

[0001] The invention relates to the field of remote sensing mapping, in particular to an anthocyanin reflectance vegetation index based on enhanced near infrared and blue bands and a method for extracting the index in farmland shelterbelts. Background Art

[0002] In the 1980s and 1990s, researchers established classification standards through visual interpretation methods based on the coupling of remote sensing images and non-remote sensing information, and realized the extraction research of farmland shelterbelts. However, this is an interpretation method that relies on expert experience, with high accuracy but relatively low efficiency. The traditional survey method has a long cycle, high cost, time-consuming and labor-intensive, and is not suitable for large-scale research and the rapid development of informatization in today's world.

[0003] With the development of remote sensing technology and its combined application with computer technology, researchers have combined the spectral information of remote sensing images with machine learning classification methods such as maximum likelihood (ML), support vector machine (SVM), random forest (RF), etc. to extract farmland shelterbelt distribution information. However, remote sensing data itself has limitations and cannot fully reflect all the characteristics of the ground objects. The lack of additional contextual information leads to poor performance of the model, especially in some complex scenes. It is impossible to effectively distinguish similar categories using only remote sensing data. There are situations of "same object, different spectrum" and "same spectrum, different objects", that is, different objects may have similar spectral characteristics, while the spectral characteristics of the same object may be different under different conditions, resulting in decreased classification accuracy.

[0004] In the prior art, remote sensing features such as spectral features, vegetation index, and texture features are combined, the wavelength range selected for the vegetation index band is changed, and the RF algorithm is used for feature selection and classification to extract the distribution of farmland shelterbelts. However, the spectral features unique to shelterbelts are not effectively captured, and the correlation between farmland shelterbelts and bands is not clearly reflected, resulting in poor extraction results. The lack of targeted indexes makes the extraction process more susceptible to the influence of other environmental factors (such as soil type, climate change, etc.), and there are problems affecting the accuracy of extraction. Texture features usually involve calculations at multiple scales and directions. Different algorithms may lead to inconsistent results, increasing the complexity of interpretation and application. The texture features between different landforms and farmland shelterbelts may be similar in some cases, leading to confusion and erroneous extraction. For this reason, the present application proposes an anthocyanin reflectance vegetation index based on enhanced near-infrared and blue bands and a method for extracting it in farmland shelterbelts. Summary of the invention

[0005] The object of the present invention is to provide an anthocyanin reflectance vegetation index based on enhanced near-infrared and blue bands and a method for extracting the same in farmland shelterbelts, so as to solve the problems raised in the above-mentioned background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: an anthocyanin reflectance vegetation index based on enhanced near infrared and blue bands, including an index RMARI-NIR and an index RMARI-RGB, wherein the index RMARI-NIR and the index RMARI-RGB are based on the index ARI, which is an indicator of plant health monitoring, and the index RMARI-RGB and the index RMARI-NIR are respectively expressed as:

[0007]

[0008] Among them, Red is the red band, Green is the green band, NIR is the near infrared band, Blue is the blue band, and L refers to the correction value.

[0009] According to the above method for extracting farmland shelterbelt based on enhancing anthocyanin reflectance vegetation index in near infrared and blue bands, the method comprises the following steps:

[0010] Step 1: Obtain images, select image months, obtain Sentinel-2 images of farmland shelterbelts during their growth period, perform atmospheric correction and geometric correction on remote sensing images, and use the “Mosaic to New Raster” tool in ArcGIS software to stitch the images together;

[0011] Step 2: Calculate the vegetation index. In the UAV remote sensing scenario, that is, the sensor only includes the visible light RGB wavelength range, the RMARI-RGB formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts; in the satellite remote sensing scenario, that is, the satellite-level sensor includes a multi-band wavelength range, the RMARI-NIR formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts. The proposed new index is also used as an additional input band to participate in the RF classification task;

[0012] Step 3: Create label data, define the labeling method according to the input of machine learning, use ArcGIS software to create a classified vector label file, perform image annotation and interpretation through human-computer interaction, and use the calculated vegetation index as an auxiliary reference to create a farmland shelterbelt dataset;

[0013] Step 4: Upload the dataset labels and the vector range of the study area on the Google Earth Engine platform, divide the training set and test set ratio, calculate the Kappa coefficient and overall accuracy as evaluation indicators, and as a verification of the feasibility of the innovation, use the RF model and additionally integrate the characteristic bands and index bands to perform farmland shelterbelt identification training and classification, superimpose the classification results with the field boundary range, and apply the mask operation to extract the farmland shelterbelt distribution location information.

[0014] Preferably, in step 1, remote sensing images with a cloud content of less than 10% are screened, atmospheric correction, geometric correction and the like are performed, and then the images are exported to Google Drive and saved locally. The images are spliced ​​using the "Mosaic to New Raster" tool in ArcGIS software according to the processed images to synthesize a complete image of the study area.

[0015] Preferably, the new index calculated in step 2 is used as an additional input band to participate in the RF classification task.

[0016] Preferably, the step 3 prepares a training and verification data set for applying farmland shelterbelt labels to the RF model.

[0017] Preferably, the step 4 adopts the RF model to divide the ratio of the training set and the validation set, and calculates the evaluation index of the Kappa coefficient and the overall accuracy (OA) as the evaluation criteria for the feasibility of the innovation.

[0018] Compared with the prior art, the present invention has the following beneficial effects:

[0019] The RMARI-NIR and RMARI-RGB vegetation indices proposed in this invention can be used to improve accuracy and stability by introducing enhanced near-infrared bands and blue bands, respectively, for extracting farmland shelterbelts. According to the extraction requirements, the spectral characteristics of farmland shelterbelts are highlighted, the degree of distinction between them and other landforms is enhanced, and they are suitable for generalization in various environments. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flow chart of a method in an embodiment of the present invention;

[0021] Figure 2 NDVI time series curve diagram of different ground objects in the embodiment of the present invention;

[0022] Figure 3 4 is a comparison diagram of the original image and the original index extraction effect in an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] The embodiment of the present invention will enhance the near-infrared and blue band anthocyanin reflectance vegetation index for farmland shelterbelt identification and extraction, and on the basis of the existing vegetation index, improve the formula structure, add additional bands, add correction value adjustment, and then optimize and improve it and participate in the RF model classification task, thereby achieving a more accurate and faster farmland shelterbelt extraction and identification solution.

[0025] Example 1

[0026] Anthocyanin reflectance vegetation index RMARI-NIR and RMARI-RGB based on enhanced near infrared and blue bands are indices based on the index ARI of plant health monitoring. The ARI vegetation index was originally used to monitor the health of vegetation leaves. The function formula of the ARI vegetation index is:

[0027]

[0028] Among them, Red is the red band, Green is the green band, and combined with the spectral characteristics of farmland shelterbelts, two improvement schemes are proposed in the present invention based on spectral mechanism analysis to extract index RMARI-NIR and index RMARI-RGB suitable for farmland shelterbelts in different scenarios.

[0029] Furthermore, the index RMARI-RGB can effectively capture the nonlinear relationship between bands by replacing the existing formula with a logarithmic structure. The logarithmic transformation can reduce the impact of high values, make the data distribution closer to the normal distribution, reduce the skewness of the data, and help improve the effectiveness of statistical analysis. Avoid the impact of extreme values ​​on vegetation indexes, and introduce the blue band to improve the stability of the formula. The index RMARI-RGB is specifically:

[0030]

[0031] Among them, Blue is the blue band, and L refers to the correction value, which is taken as 1 in the present invention.

[0032] At the same time, the index RMARI-NIR is based on the index RMARI-RGB. By introducing an additional near-infrared band on the basis of the index RMARI-RGB, more spectral information can be provided and the ability to identify vegetation features can be enhanced. The index RMARI-NIR is specifically:

[0033]

[0034] Among them, NIR is the near infrared band.

[0035] Example 2

[0036] Reference Figure 1As shown, this embodiment provides a method for extracting farmland shelterbelt based on anthocyanin reflectance vegetation index enhanced in near infrared and blue bands, comprising the following steps:

[0037] Step 1, combining the phenological analysis of the farmland shelterbelt with its significant characteristic moments, selecting the identified time window, the present invention obtains the remote sensing image of the study area corresponding to Sentinel-2 in May.

[0038] Step 2, the index RMARI-NIR and the index RMARI-RGB are proposed, including the improvement of the original ARI vegetation index formula structure, the addition of additional bands, and the addition of correction value adjustment. In the UAV remote sensing scene, the index RMARI-RGB formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts; in the satellite remote sensing scene, the index RMARI-NIR formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts, and the new index is also used as an additional input band to participate in the RF classification task. The present invention mainly uses the RMARI-NIR improved index as a feature band to participate in subsequent classification tasks and analysis.

[0039] Step 3: Define the annotation method based on the input of machine learning, and create a farmland shelterbelt dataset by annotating and interpreting images through human-computer interaction.

[0040] Step 4: The characteristic bands and indexes are integrated into the RF model for identification and classification, and the trained results are superimposed on the field boundary to obtain the distribution location information of the farmland shelterbelt.

[0041] In the above step 1, the Sentinel-2 farmland shelterbelt image in May is obtained, and the vector range (ROI) of the relevant study area is uploaded through the Google Earth Engine (GEE) platform. In order to prevent large-area cloud interference, the present invention screens the images with a cloud content of less than 10%, selects the required band, performs atmospheric correction, geometric correction and other processing on the remote sensing image, exports the image to Google Drive and saves it locally, and uses the "Mosaic to New Raster" tool in ArcGIS software to splice the image according to the processed image to synthesize a complete image of the study area.

[0042] In step 2 above, a good vegetation index can maximize the sensitivity to vegetation characteristics and model accuracy. The most commonly used spectral bands for developing vegetation indices are located in the red region of chlorophyll absorption, about 670nm, and the near-infrared region, 750-900nm. Vegetation has strong reflection in this region. The specific wavelengths of visible light and near-infrared have been determined, which can be used to monitor biochemical substances in leaves at the canopy scale. It is further shown that the spectral absorption characteristics of the leaves of farmland shelterbelts are non-specific, have a wide coverage range, and have strong spectral signals, which are easy to distinguish from other landforms. The spectral characteristics of farmland shelterbelts in May mainly depend on its leaves, which also belong to green vegetation and are suitable for vegetation indices. There are always certain differences in spectral curves for different landform categories, which is the basis for identifying landform classification information, such as Figure 2 The figure shows the difference in the NDVI average values ​​of the band spectra between different land objects under the NDVI time series. By introducing additional spectral bands into the anthocyanin reflectance vegetation index and optimizing it through logarithmic enhancement transformation, its discrimination ability and applicability for farmland shelterbelt extraction are increased. Band Math function calculates index image, generates .tif format file, imports into ArcGIS software to display detail effect, selects appropriate index with greater distinction between farmland shelterbelt and different ground objects pixel value as additional input band of RF, combines spectral characteristics of farmland shelterbelt and vegetation index analysis, for complex vegetation remote sensing, multispectral remote sensing data is often selected for mathematical analysis and calculation, in this invention, two improved schemes are proposed based on spectral mechanism analysis to be applicable to the extraction index of farmland shelterbelt, namely formula 2 and 3, index RMARI-NIR and index RMARI-RGB, combined with the difference of spectral reflectance of farmland shelterbelt and other ground objects, because the internal structure of vegetation leaves has strong scattering effect on near infrared, wavelength of about 700-1300nm, and shows very high reflectance, and this part of light is There is no direct effect on the photosynthesis of plants, showing the high reflectivity characteristics of vegetation in the near-infrared band. Therefore, in order to make it easier to distinguish farmland shelterbelts from other landforms and improve the classification effect and accuracy, near-infrared and blue bands are introduced to enhance the strong correlation between vegetation; the original improved index ARI only uses red and green bands. In order to reduce the existence of unexplained accidental factors, limitations and avoid linear relationships between bands, additional near-infrared and blue bands are added to the original formula, the formula is changed into a logarithmic square structure and a correction value is added. The best distinction scheme is selected, and the new index is also used as an additional input band to participate in the RF classification task. The present invention mainly uses the RMARI-NIR improved index as a characteristic band to participate in subsequent classification tasks and analysis. The effect comparison with the original index formula is shown in the attached figure. Figure 3 , where a1-a10 are the original images, b1-b10 are the original indices, and c1-c10 are the improved indices.

[0043] In the above step 3, after calculating the vegetation index, there is a significant difference between the farmland shelterbelt and other landforms. The farmland shelterbelt labels are made and applied to the training and verification data set of the RF model. Vectors are created, and the projected geographic coordinate system is selected to label the farmland shelterbelt. About 300-500 data labels are made, and typical and representative samples are selected. The farmland shelterbelt must be completely covered inside the label vector. Suitable lengths are selected for breakpoints. Even if there are adjacent areas, they need to be labeled separately to avoid interference from the spectral information of other landforms. At the same time, in order to ensure the accuracy of classification and avoid contingency, labels for other landforms are made, such as buildings, farmland, water bodies, etc. A total of about 100 data labels are made, and the attribute table is opened to create a type field for assigning values ​​to the category. To match the typical representation range of image pixel values, the category threshold is set between 0 and 255, and the label vector .shp and other files are uploaded to Google Earth Engine (GEE) platform, call the Sentinel-2 satellite, select the time, area of ​​interest, select the red, green, blue bands and the bands needed for vegetation index, read the type field category of the uploaded label vector, add vegetation index as a feature band to the classified image, enter the selected band, enter the calculation formula, and finally integrate the added bands.

[0044] In step 4 above, the RF model was used to divide the proportions of the training set and the validation set, and the Kappa coefficient and the overall accuracy (OA) evaluation index were calculated as the evaluation criteria for the feasibility of the innovation. After the above steps were completed, the model training was started, and the results were saved to Google Drive. After downloading to the local computer, it was opened through ArcGIS software, and different unique value categories were selected, that is, the farmland shelterbelt labeling categories after the model classification results were selected. Then, according to the field boundary data in the land use, the non-farmland shelterbelt part was removed by masking operation to extract the distribution of the farmland shelterbelt.

[0045] In the above steps 1 to 4, it can be seen that the classification accuracy is slightly improved after adding the index as a characteristic band, but the evaluation index of the improved RMARI-NIR index extraction model is slightly lower than the ARI index extraction accuracy, which is reduced by 0.0013 and 0.0012 respectively. The difference in change is not obvious. The Kappa coefficient and the overall accuracy (OA) are similar. The reason is that a small number of other land object category labels are added to participate in the training classification during the classification process to improve the robustness and stability of the model in the classification process, improve the obvious difference between different land objects, and make the contours of different categories of extraction results more complete and clear. Under the same image and label, the improvement scheme focuses on improving the extraction effect and accuracy of farmland shelterbelts, and ignores the band feature sensitivity extraction of soil, buildings, and water bodies in non-vegetated areas to a certain extent. The overall evaluation index will change according to the results of other land object classification, so the Kappa coefficient and the overall accuracy OA will be affected, and there is a slightly decreasing trend.

[0046] Table 1 Accuracy of ARI, RMARI-NIR and RMARI-RGB indices extracted in the RF model

[0047]

[0048] For the extraction of farmland shelterbelt, the proposed RMARI-NIR and RMARI-RGB vegetation indices can be used to improve accuracy and stability. According to the extraction requirements, the spectral characteristics of farmland shelterbelt are highlighted, the degree of distinction between farmland shelterbelt and other land objects is enhanced, the generalization under various environments is suitable, and the stability and universality are verified. The feasibility of this method is verified, and the RMARI-NIR and RMARI-RGB indices and the ARI index are respectively used as characteristic bands to participate in RF classification. The accuracy comparison is shown in Table 1, which verifies the feasibility of the innovative improvement.

[0049] The index RMARI-NIR and index RMARI-RGB respectively combine the existing red and green band features, introduce additional near-infrared and blue bands, highlight the vegetation characteristics, enhance the sensitivity of the spectral bands, and suppress other background information such as soil, buildings, bare land, etc.; introduce correction values ​​to adjust the formula using logarithms and squares as a whole, so that the index can maintain stable extraction results under different images and environmental conditions. The original ARI index is used for vegetation health monitoring. The improved vegetation index is now used for farmland shelterbelt extraction to achieve high-precision extraction. It provides a method for establishing a farmland shelterbelt extraction model that can achieve high-precision distinction between farmland shelterbelt and other ground information over a large range while reducing the amount of engineering work, and is semi-automatic and transferable. It realizes a method model for large-scale monitoring at the remote sensing technology level, and proposes a more effective, identification and extraction method. The new vegetation index improvement technical scheme with higher extraction accuracy, higher degree of distinction from other landforms, more complete edge contours of farmland shelterbelt extraction, and reduced noise from other landforms improves the inapplicability of existing commonly used indices for farmland shelterbelt extraction. The index RMARI-NIR involves four commonly used basic bands of red, green, blue, and near-infrared. While it is suitable for different series of remote sensing satellite images, it is also suitable for equipment such as drones that only have RGB and near-infrared sensors; the index RMARI-RGB only involves red, green, and blue visible light bands, and can also be applied to digital cameras and other equipment that do not have near-infrared band sensors. The improvements of the two schemes and the RF model classification and extraction method are to provide more accurate agricultural support for relevant industry workers, ensure the long-term sustainable development of farmland, and enhance the overall economic benefits of agriculture.

[0050] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0051] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An anthocyanin reflectance vegetation index based on enhanced near infrared and blue bands, including the index RMARI-NIR and the index RMARI-RGB, characterized in that: The index RMARI-NIR and the index RMARI-RGB are improved indices proposed based on the index ARI of plant health monitoring indicators in combination with the vegetation spectrum mechanism, and the index RMARI-RGB and the index RMARI-NIR are respectively expressed as: Among them, Red is the red band, Green is the green band, NIR is the near infrared band, Blue is the blue band, and L refers to the correction value.

2. The method for extracting farmland shelterbelt based on enhancing anthocyanin reflectance vegetation index in near infrared and blue bands according to claim 1, characterized in that: The steps include: Step 1: Obtain images, select image months, obtain and download Sentinel-2 farmland shelterbelt growing period images, perform atmospheric correction and geometric correction on remote sensing images, and use the "Mosaic to New Raster" tool in ArcGIS software to stitch the acquired images; Step 2: Calculate the vegetation index. In the UAV remote sensing scenario, that is, the sensor only includes the visible light RGB wavelength range, the RMARI-RGB formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts; in the satellite remote sensing scenario, that is, the satellite-level sensor includes multi-band wavelength ranges, the RMARI-NIR formula is used to calculate the vegetation index to extract the distribution of farmland shelterbelts. The proposed new index is also used as an additional input band to participate in the random forest (RF) classification task; Step 3: Create label data, define the labeling method according to the input of machine learning, use ArcGIS software to create a classified vector label file, perform image annotation and interpretation through human-computer interaction, and use the calculated vegetation index as an auxiliary reference to create a farmland shelterbelt label dataset; Step 4: Upload the label dataset and the vector range of the study area on the Google Earth Engine platform, divide the ratio of the training set to the test set, calculate the Kappa coefficient and overall accuracy as evaluation indicators, and as a verification method for the feasibility of the innovation, use the RF model and additionally incorporate characteristic bands and index bands to perform farmland shelterbelt identification training and classification, superimpose the classification results with the field boundary range, and apply mask operations to extract the distribution location information of the farmland shelterbelt.

3. The method for extracting farmland shelterbelt based on enhanced near infrared and blue band anthocyanin reflectance vegetation index according to claim 2, characterized in that: The step 1 screens remote sensing images with cloud content less than 10%, that is, reduces the interference of clouds on image analysis, performs atmospheric correction, geometric correction and other processing, and then exports the images to Google Drive and saves them locally. According to the processed images, the "Mosaic to New Raster" tool in ArcGIS software is used to stitch the images to synthesize a complete image of the study area.

4. The method for extracting farmland shelterbelt based on enhancing anthocyanin reflectance vegetation index in near infrared and blue bands according to claim 3, characterized in that: The new index calculated in step 2 is used as an additional input band to participate in the RF classification task.

5. The method for extracting farmland shelterbelt based on enhancing anthocyanin reflectance vegetation index in near infrared and blue bands according to claim 4, characterized in that: The step 3 prepares a training and verification data set for the farmland shelterbelt label to be applied to the RF model.

6. The method for extracting farmland shelterbelt based on enhanced near-infrared and blue band anthocyanin reflectance vegetation index according to claim 5, characterized in that: The step 4 adopts the RF model to divide the ratio of the training set and the validation set, and calculates the Kappa coefficient and the overall accuracy evaluation index as the evaluation criteria for the feasibility of the innovation.

Images