Farmland shelterbelt image extraction method and device and electronic equipment

Abstract

The application provides a farmland shelterbelt image extraction method and device and electronic equipment, and belongs to the technical field of computers.The method comprises the following steps: performing feature extraction on a time sequence remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, and acquiring a time sequence feature set corresponding to each target image feature category; inputting the time sequence feature set corresponding to each target image feature category into a target random forest classification model, and acquiring a classification result output by the target random forest classification model; and based on the classification result, extracting a first farmland shelterbelt image from a remote sensing image of the target region. Through feature extraction on multi-temporal data, the extracted features can fully represent the seasonal changes and growth dynamics of the farmland shelterbelt, thereby helping the classification model to identify the ground object classification to which a pixel belongs, and the accuracy of extracting farmland shelterbelt information from the remote sensing image can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and in particular to a farmland shelterbelt image extraction method and device and electronic equipment. BACKGROUND

[0002] In related art, in terms of methods for extracting farmland shelterbelt information from remote sensing images, common methods tend to use single-phase high-resolution images for analysis, but due to seasonal changes and growth dynamics of farmland shelterbelts, single-phase images cannot accurately capture these changes, resulting in low accuracy. SUMMARY

[0003] To solve the problems in the prior art, the embodiments of the present application provide a farmland shelterbelt image extraction method, device and electronic equipment.

[0004] In a first aspect, the present application provides a farmland shelterbelt image extraction method, comprising:

[0005] Based on a plurality of target image feature categories and a plurality of target months, performing feature extraction on a time series remote sensing image set of a target area to obtain a time series feature set corresponding to each target image feature category, the image feature category being a spectral feature category, a terrain feature category or a texture feature category;

[0006] Inputting the time series feature set corresponding to each target image feature category into a target random forest classification model to obtain a classification result output by the target random forest classification model, the classification result being used to represent a ground object classification corresponding to each pixel in the remote sensing image of the target area;

[0007] Based on the classification result, extracting a first farmland shelterbelt image from the remote sensing image of the target area.

[0008] Optionally, according to the farmland shelterbelt image extraction method provided by the present application, before the feature extraction on the time series remote sensing image set of the target area based on the plurality of target image feature categories and the plurality of target months to obtain the time series feature set corresponding to each target image feature category, the method further comprises:

[0009] Based on a preset time step, performing multiple image fusion on the remote sensing images of the target area in time sequence to obtain a plurality of target time periods respectively corresponding to the fused remote sensing images, one image fusion being used to process a plurality of remote sensing images within one preset time step, and the length of the target time period being equal to the preset time step;

[0010] Based on the fused remote sensing images respectively corresponding to the plurality of target time periods, determining the time series remote sensing image set.

[0011] Optionally, the application provides a farmland shelter forest image extraction method, and the image fusion comprises the following steps:

[0012] Based on the preset time step, a plurality of to-be-fused remote sensing images in a target period are acquired;

[0013] Based on the plurality of to-be-fused remote sensing images, a plurality of pixel remote sensing data corresponding to each pixel are acquired;

[0014] For each pixel, target pixel remote sensing data with the highest normalized difference vegetation index (NDVI) are selected from the plurality of pixel remote sensing data corresponding to the pixel;

[0015] Based on the target pixel remote sensing data corresponding to each pixel, a fused remote sensing image corresponding to the target period is generated.

[0016] Optionally, in the farmland shelter forest image extraction method provided by the application, before the feature extraction of the time sequence remote sensing image set of the target region based on the plurality of target image feature categories and the plurality of target months, and the acquisition of the time sequence feature set corresponding to each target image feature category, the method further comprises the following steps:

[0017] Based on the plurality of preset image feature categories and the time sequence remote sensing image set, the feature importance of each preset image feature category in each target period is acquired by random forest evaluation;

[0018] Based on the feature importance of each preset image feature category in each target period, the plurality of preset image feature categories are screened to determine the plurality of target image feature categories;

[0019] Based on the feature importance of each preset image feature category in each target period, the months in a year are screened to determine the plurality of target months.

[0020] Optionally, in the farmland shelter forest image extraction method provided by the application, the screening of the plurality of preset image feature categories based on the feature importance of each preset image feature category in each target period and the determination of the plurality of target image feature categories comprise the following steps:

[0021] For each preset image feature category, the feature importance average value corresponding to the preset image feature category is determined by calculating the average value based on the feature importance of the preset image feature category in each target period;

[0022] Based on the first feature importance threshold and the feature importance average value corresponding to each preset image feature category, the plurality of preset image feature categories are screened to determine the plurality of target image feature categories.

[0023] Optionally, the application provides a farmland shelterbelt image extraction method, which comprises the following steps:

[0024] For each month, based on the second feature importance threshold and the feature importance value of each preset image feature category corresponding to each target period, the target category number of the preset image feature category with the feature importance value greater than the second feature importance threshold is counted to determine the target category number corresponding to the month.

[0025] Based on the category number threshold and the target category number corresponding to each month, the months in a year are screened to determine the target months.

[0026] Optionally, according to the farmland shelterbelt image extraction method provided by the application, after the first farmland shelterbelt image is extracted from the remote sensing image of the target area based on the classification result, the method further comprises the following steps:

[0027] Based on the known ground feature type data, the farmland shelterbelt image is processed by a mask removal method to obtain a second farmland shelterbelt image.

[0028] The connected pixels in the second farmland shelterbelt image with the pixel number less than the pixel number threshold are filtered out, and the holes in the second farmland shelterbelt image are eliminated to obtain a third farmland shelterbelt image.

[0029] Optionally, according to the farmland shelterbelt image extraction method provided by the application, after the connected pixels in the second farmland shelterbelt image with the pixel number less than the pixel number threshold are filtered out and the holes in the second farmland shelterbelt image are eliminated to obtain a third farmland shelterbelt image, the method further comprises the following steps:

[0030] Based on the remote sensing image of the target area, a land parcel segmentation processing is performed to obtain a land parcel segmentation result image.

[0031] Based on the land parcel segmentation result image and the third farmland shelterbelt image, image fusion is performed to obtain a land parcel surrounding image, which is used to represent the state of the land parcel surrounded by the shelterbelt.

[0032] Based on the land parcel surrounding image, the farmland shelterbelt of the target area is evaluated to obtain at least one farmland shelterbelt parameter.

[0033] The at least one farmland shelterbelt parameter comprises one or more of the following parameters: land parcel surrounding degree, shelterbelt angle and effective rate of main-harm wind prevention.

[0034] In a second aspect, the present application further provides a farmland shelterbelt image extraction device, comprising:

[0035] a feature extraction module configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, wherein the image feature category is a spectral feature category, a terrain feature category or a texture feature category;

[0036] a feature ground object classification module configured to input the time series feature set corresponding to each target image feature category into a target random forest classification model, to obtain a classification result output by the target random forest classification model, wherein the classification result is used to represent a ground object classification corresponding to each pixel in the remote sensing image of the target region;

[0037] an image extraction module configured to extract a first farmland shelterbelt image from the remote sensing image of the target region based on the classification result.

[0038] In a third aspect, the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the farmland shelterbelt image extraction method according to any one of the above aspects when executing the program.

[0039] The farmland shelterbelt image extraction method, device and electronic device provided by the present application can obtain a time series feature set corresponding to each target image feature category from multi-temporal data by performing feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, and then input the time series feature set corresponding to each target image feature category into a target random forest classification model, classify each pixel by the target random forest classification model, obtain a classification result, and extract a farmland shelterbelt image from the remote sensing image of the target region. The extracted features can fully represent the seasonal changes and growth dynamics of the farmland shelterbelt, which helps the classification model to identify the ground object classification to which the pixel belongs, and improves the accuracy of extracting farmland shelterbelt information from the remote sensing image. BRIEF DESCRIPTION OF DRAWINGS

[0040] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0041] Figure 1 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0042] Figure 2 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0043] Figure 3 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0044] Figure 4 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0045] Figure 5 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0046] Figure 6 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0047] Figure 7 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0048] Figure 8 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0049] Figure 9 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0050] Figure 10 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application;

[0051] Figure 11 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application; DETAILED DESCRIPTION

[0052] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions in the present application will be described below in connection with the drawings in the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the scope of protection of the present application.

[0053] Figure 1 is one of flowcharts of the method for extracting the image of the shelter forest in the farmland provided by the application, as shown in Figure 1 the execution subject of the extraction method can be an electronic device, such as a server, etc. The method comprises:

[0054] In step 101, feature extraction is performed on the time-series remote sensing image set of the target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time-series feature set corresponding to each target image feature category. The image feature category is a spectral feature category, a terrain feature category, or a texture feature category.

[0055] Specifically, the plurality of target image feature categories can include one or more of a spectral feature category, a terrain feature category, and a texture feature category. The spectral feature category relates to spectral information of different wavebands in the image, the terrain feature category relates to terrain elevation, slope, aspect, etc., and the texture feature category describes texture changes in the image.

[0056] In order to obtain the time-series remote sensing image set of the target region, a period of time (e.g., one year or more) can be selected, and a plurality of remote sensing images of the target region can be obtained. Then, the plurality of remote sensing images can be sorted in time sequence, and the time-series remote sensing image set of the target region can be obtained.

[0057] Since the time-series remote sensing image set of the target region covers a period of time, remote sensing images corresponding to each target month can be filtered from the time-series remote sensing image set of the target region based on a plurality of target months. Then, feature extraction can be performed on the remote sensing images corresponding to each target month based on a plurality of target image feature categories, and a time-series feature set corresponding to each target image feature category can be obtained.

[0058] For example, the plurality of target months includes September and October, remote sensing images corresponding to September and October are filtered from the time-series remote sensing image set of the target region, and feature extraction is performed on the remote sensing images corresponding to September and October based on a plurality of target image feature categories, to obtain a time-series feature set corresponding to each target image feature category. The time-series feature set corresponding to the target image feature category includes features extracted from the September remote sensing images and features extracted from the October remote sensing images. The features extracted are sorted in time sequence to obtain the time-series feature set.

[0059] In step 102, the time-series feature set corresponding to each target image feature category is input into a target random forest classification model, and a classification result output by the target random forest classification model is obtained. The classification result is used to represent a ground object classification corresponding to each pixel in the remote sensing image of the target region.

[0060] Specifically, the time-series feature set corresponding to each target image feature category is input into the target random forest classification model to obtain the classification result. The random forest is composed of a plurality of decision trees, and classification is performed through collective decision.

[0061] The time series feature set input to the target random forest classification model contains feature information from different feature categories and different time points. These features can include spectral features, topographic features, or texture features, etc., which are used to describe the properties of each pixel in the remote sensing image.

[0062] The random forest model can classify each pixel based on these features and output the corresponding classification results. The classification results can correspond to different land cover categories, such as farmland shelterbelt, farmland, or water body, etc. These classification results reflect the land cover categories represented by each pixel in the remote sensing image.

[0063] In step 103, based on the classification results, a first farmland shelterbelt image is extracted from the remote sensing image of the target area.

[0064] Specifically, in step 103, according to the classification results obtained in step 102, the operation of extracting the first farmland shelterbelt image from the remote sensing image of the target area can be performed.

[0065] The classification results are output by the target random forest classification model, which represent the land cover classification corresponding to each pixel in the remote sensing image. In step 103, the focus is on the specific land cover category of farmland shelterbelt.

[0066] According to the classification results, pixels belonging to the farmland shelterbelt category can be screened out. These pixels have characteristics related to farmland shelterbelt in the remote sensing image, such as specific spectral reflectance, shape, or texture, etc. By extracting these pixels, they can be combined into a separate image, i.e., the first farmland shelterbelt image.

[0067] Such operations help to more intuitively observe and analyze the distribution and characteristics of farmland shelterbelt in the target area. This image can be used for further research and application, such as evaluating the coverage and changes of farmland shelterbelt, understanding the impact of farmland shelterbelt on the environment and ecological system, etc.

[0068] The application provides an agricultural shelterbelt image extraction method, which can obtain a time sequence feature set corresponding to each target image feature category from multi-temporal data by performing feature extraction on a time sequence remote sensing image set of a target region based on multiple target image feature categories and multiple target months, and then inputting the time sequence feature set corresponding to each target image feature category into a target random forest classification model, classifying each pixel by the target random forest classification model, and obtaining a classification result, which can represent a ground object classification corresponding to each pixel in the remote sensing image of the target region, so that the agricultural shelterbelt image can be extracted from the remote sensing image of the target region. The extracted features can fully represent the seasonal changes and growth dynamics of the agricultural shelterbelt, and then help the classification model to identify the ground object classification to which the pixel belongs, thereby improving the accuracy of extracting agricultural shelterbelt information from the remote sensing image.

[0069] Optionally, according to the agricultural shelterbelt image extraction method provided by the application, before the feature extraction on the time sequence remote sensing image set of the target region based on multiple target image feature categories and multiple target months, and obtaining the time sequence feature set corresponding to each target image feature category, the method further comprises:

[0070] Based on a preset time step, the remote sensing images of the target region are sequentially fused for multiple times to obtain fused remote sensing images corresponding to multiple target time periods, one image fusion is used to process multiple remote sensing images within one preset time step, and the length of the target time period is equal to the preset time step.

[0071] Based on the fused remote sensing images corresponding to the multiple target time periods, the time sequence remote sensing image set is determined.

[0072] Specifically, Figure 2 is a flowchart of the agricultural shelterbelt image extraction method provided by the application, as shown in Figure 2 The method comprises steps 201 to 205.

[0073] Step 201: Based on a preset time step, the remote sensing images of the target region are sequentially fused for multiple times to obtain fused remote sensing images corresponding to multiple target time periods.

[0074] Specifically, image fusion is a process of combining multiple remote sensing images into one image. Here, a specific time span, i.e. a preset time step, such as 7 days, 15 days or one month, is selected in chronological order. Within this time span, multiple remote sensing images are collected and then fused.

[0075] By fusing multiple remote sensing images in this period of time, a fused remote sensing image can be obtained, which contains multiple features and information in this period of time. Such fusion can provide more comprehensive and accurate ground features, and help better analyze the spatio-temporal changes of the target region.

[0076] According to the selection of the preset time step, the image fusion process can be repeated, and multiple remote sensing images in a preset time step are processed each time. In this way, multiple fused remote sensing images corresponding to target time periods can be obtained, and each fused image represents a specific time period with a time length equal to the preset time step.

[0077] In step 202, based on the multiple fused remote sensing images corresponding to the target time periods respectively, the time sequence remote sensing image set is determined.

[0078] Specifically, in step 202, the fused remote sensing images can be summarized and formed into a time sequence remote sensing image set. The set contains multiple fused images corresponding to the target time periods respectively, and each fused image represents a specific time period with a time length equal to the preset time step.

[0079] In step 203, based on the multiple target image feature categories and the multiple target months, feature extraction is performed on the time sequence remote sensing image set of the target region to obtain a time sequence feature set corresponding to each target image feature category.

[0080] In step 204, the time sequence feature set corresponding to each target image feature category is input into the target random forest classification model to obtain a classification result output by the target random forest classification model.

[0081] In step 205, based on the classification result, a first farmland shelterbelt image is extracted from the remote sensing image of the target region.

[0082] In an optional example A, the remote sensing image of the target region can be obtained through the following data sources.

[0083] Data sources: (1) Harmonized Sentinel-2MSI: MultiSpectral Instrument, Level-2A; (2) SRTM DEM; (3) ALOS DSM; (4) Google Maps.

[0084] Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A refers to the harmonized processing of Sentinel-2 satellite's MultiSpectral Instrument (MSI) data. Sentinel-2 is a series of Earth observation satellites by the European Space Agency (ESA), which carry the MSI instrument to collect multispectral data of the Earth's surface. Level-2A refers to a series of preprocessing and correction steps applied to the Sentinel-2 MSI data, so that users can more easily use these data for analysis and application. Level-2A data includes atmospheric correction, radiometric correction, geometric correction, etc. steps to provide high-quality images and data products. Therefore, Harmonized Sentinel-2 MSI: MultiSpectral Instrument, Level-2A can be understood as the unified processing of the MSI data of the Sentinel-2 satellite, and the generation of preprocessed and corrected Level-2A data, so that users can better use these data for remote sensing analysis and application.

[0085] SRTM DEM refers to the Digital Elevation Model (DEM) obtained by the Shuttle Radar Topography Mission (SRTM). DEM represents a digital elevation model, which is a digitalized data used to represent the elevation of the Earth's surface. DEM can provide topographic information such as ground height, mountains or valleys, etc.

[0086] ALOS DSM refers to the Digital Surface Model (DSM) obtained by the ALOS (Advanced Land Observing Satellite), which is an Earth observation satellite used to obtain remote sensing data of the Earth's surface. DSM represents a digital surface model, which is a model used to represent the Earth's surface, including the elevation information of objects such as ground, vegetation, buildings, etc. DSM is similar to DEM, but in DEM, objects below the ground surface such as buildings, trees, etc. are removed, only the elevation information of the ground is retained, while DSM retains the elevation information of objects above the ground surface.

[0087] Optionally, in the above example A, for the remote sensing image of the target area, cloud removal can be achieved using the QA60 band marker.

[0088] Optionally, in the above example A, a remote sensing image of more than one year (e.g., 390 days) of the target region can be acquired by the data source, and the preset time step can be 15 days. Remote sensing image fusion can be performed every 15 days to generate a fused remote sensing image. A plurality of fused remote sensing images can be acquired, i.e., a plurality of fused remote sensing images corresponding to a plurality of target time periods, and the time length of a target time period is equal to 15 days. For example, for the remote sensing image of 390 days of the target region, remote sensing image fusion can be performed multiple times with 15 days as the preset time step, and 26 fused remote sensing images can be acquired, i.e., 26 fused remote sensing images corresponding to 26 target time periods.

[0089] Optionally, in the above example A, the target image feature category can be any one of the image feature categories in Table 1.

[0090] Table 1: Preset image feature category table

[0091]

[0092] The normalized difference vegetation index (NDVI) is a commonly used vegetation index for evaluating the degree of vegetation coverage.

[0093] The enhanced vegetation index (EVI) is an improved vegetation index that corrects for atmospheric effects and soil background reflection in pixels.

[0094] The green chlorophyll vegetation index (GCVI) is an index for estimating the chlorophyll content of vegetation.

[0095] The land surface water index (LSWI) is an index for detecting surface water bodies.

[0096] The bare soil index (BSI) is an index for detecting bare soil.

[0097] The red edge position (REP) is an index for estimating the chlorophyll content and physiological state of vegetation leaves.

[0098] The modified simple ratio (MSR705) is an index for detecting the chlorophyll content and growth state of vegetation.

[0099] Blue light is a part of the visible spectrum, with a wavelength range typically between 450 and 495 nanometers.

[0100] Green light: Green light is a part of the visible spectrum, with a wavelength range typically between 495 and 570 nanometers.

[0101] Red light: Red light is a part of the visible spectrum, with a wavelength range typically between 620 and 750 nanometers.

[0102] Edge1, Edge2, Edge3 (edge ​​bands): Edge bands usually refer to the bands located between the visible light and near-infrared spectra.

[0103] Slope: Slope refers to the degree of inclination of the ground surface or terrain.

[0104] SOS (Slope of slope, second-order slope): SOS refers to the rate of change of slope or the slope of slope.

[0105] DSM_slope (Digital Surface Model Slope): DSM_slope refers to the slope calculated based on the Digital Surface Model (DSM).

[0106] NDVI_ASM (Normalized Difference Vegetation Index Angular Second Moment): NDVI_ASM is a statistic used to describe the texture features of images based on the Normalized Difference Vegetation Index (NDVI).

[0107] NDVI_CORR (Normalized Difference Vegetation Index Correlation): NDVI_CORR is used to measure the correlation between pixel values ​​in an NDVI image.

[0108] NDVI_DENT (Normalized Difference Vegetation Index Difference Entropy): NDVI_DENT is used to describe the differences in pixel values ​​in an NDVI image.

[0109] NDVI_ENT (Normalized Difference Vegetation Index Entropy): NDVI_ENT is used to evaluate the uncertainty and randomness of pixel values in the NDVI image.

[0110] NDVI_IDM (Normalized Difference Vegetation Index Inverse Difference Moment): NDVI_IDM is used to describe the local uniformity and smoothness of pixel values in the NDVI image.

[0111] NDVI_IMCORR (Normalized Difference Vegetation Index Image Correlation): NDVI_IMCORR1 / NDVI_IMCORR2 is used to measure the correlation between pixels at different locations in the NDVI image.

[0112] NDVI_SENT (Normalized Difference Vegetation Index Sentropy): NDVI_SENT is used to describe the complexity and randomness of vegetation distribution in the NDVI image.

[0113] wherein the Gray-Level Co-occurrence Matrix (GLCM) is a statistical tool used to describe the texture features of an image, commonly used in satellite remote sensing image analysis. GLCM is based on the gray value relationship between pixels in the image, providing information about the texture of the image.

[0114] Optionally, according to the image extraction method of the farmland shelter forest provided by the present application, the image fusion comprises:

[0115] Based on the preset time step, a plurality of to-be-fused remote sensing images in a target period are obtained;

[0116] Based on the plurality of to-be-fused remote sensing images, a plurality of pixel remote sensing data corresponding to each pixel are obtained;

[0117] For each pixel, the target pixel remote sensing data with the highest normalized vegetation index NDVI is selected from the plurality of pixel remote sensing data corresponding to the pixel;

[0118] Based on the target pixel remote sensing data corresponding to each pixel, a fused remote sensing image corresponding to the target period is generated.

[0119] Specifically, Figure 3 is a flowchart of image fusion provided by the present application, as Figure 3 shown, the image fusion process includes steps 301 to 304.

[0120] Step 301, based on the preset time step, obtaining a plurality of to-be-fused remote sensing images within a target period.

[0121] Specifically, in step 301, a plurality of remote sensing images are selected according to the preset time step, and the time span (i.e. the target period) of these images is consistent with the preset time step. These to-be-fused remote sensing images are used as input for subsequent step 302.

[0122] Step 302, based on the plurality of to-be-fused remote sensing images, obtaining a plurality of pixel remote sensing data corresponding to each pixel.

[0123] Specifically, in step 302, a plurality of pixel remote sensing data corresponding to each pixel are obtained by sampling each pixel using a plurality of to-be-fused remote sensing images. This means that for the position of each pixel, corresponding data is collected in different remote sensing images.

[0124] Step 303, for each pixel, filtering out the target pixel remote sensing data with the highest normalized difference vegetation index (NDVI) from the plurality of pixel remote sensing data corresponding to the pixel.

[0125] Specifically, the normalized difference vegetation index (NDVI) is a commonly used remote sensing index for evaluating and quantifying the condition of ground vegetation. It is calculated based on remote sensing data of different bands and can reflect the growth condition and coverage of vegetation.

[0126] In step 302, a plurality of pixel remote sensing data corresponding to each pixel are obtained, which may include reflectance or radiation values of different bands, etc.

[0127] In step 303, the normalized difference vegetation index (NDVI) of each pixel can be calculated using these pixel remote sensing data, and the highest value can be filtered out. This means that the data with the highest vegetation coverage among the plurality of pixel remote sensing data corresponding to each pixel can be found.

[0128] By selecting the target pixel remote sensing data with the highest normalized difference vegetation index (NDVI), the ground feature information with the highest vegetation coverage can be obtained, and the data other than the target pixel remote sensing data among the plurality of pixel remote sensing data can be filtered out. This can reduce the data input to the classification model while preserving the ground feature of the pixel, thereby improving the operation efficiency.

[0129] At step 304, a fused remote sensing image corresponding to the target period is generated based on the target pixel remote sensing data corresponding to each pixel.

[0130] Optionally, according to the image extraction method of the farmland shelter forest provided by the present application, before the feature extraction of the time series remote sensing image set of the target region based on the plurality of target image feature categories and the plurality of target months, and the acquisition of the time series feature set corresponding to each target image feature category, the method further comprises:

[0131] Based on the plurality of preset image feature categories and the time series remote sensing image set, the feature importance is evaluated by random forest, and the feature importance value corresponding to each preset image feature category in each target period is acquired.

[0132] Based on the feature importance value corresponding to each preset image feature category in each target period, the plurality of preset image feature categories are screened, and the plurality of target image feature categories are determined.

[0133] Based on the feature importance value corresponding to each preset image feature category in each target period, the months in a year are screened, and the plurality of target months are determined.

[0134] Specifically, Figure 4 is a flowchart of the image extraction method of the farmland shelter forest provided by the present application, as shown in Figure 4 The extraction method comprises steps 401 to 406.

[0135] At step 401, based on the plurality of preset image feature categories and the time series remote sensing image set, the feature importance is evaluated by random forest, and the feature importance value corresponding to each preset image feature category in each target period is acquired.

[0136] Specifically, random forest is a machine learning algorithm that can be used for feature selection and predictive modeling. In this step 401, the random forest algorithm is used to evaluate the feature importance value corresponding to each target period of different preset image feature categories. By analyzing the remote sensing image data and the preset image feature categories, the importance value of each feature category in each target period can be obtained.

[0137] At step 402, based on the feature importance value corresponding to each preset image feature category in each target period, the plurality of preset image feature categories are screened, and the plurality of target image feature categories are determined.

[0138] Specifically, in step 401, the feature importance is evaluated by random forest, and the feature importance value corresponding to each preset image feature category in each target period is obtained. These feature importance values represent the contribution degree of each preset image feature category in each target period.

[0139] In step 402, the plurality of preset image feature categories can be screened according to the feature importance values. Selecting those preset image feature categories that perform most important in each target period as the target image feature categories of interest can reduce the amount of data input to the classification model and improve the operation efficiency.

[0140] In step 403, the months in a year are screened based on the feature importance values corresponding to each preset image feature category in each target period, and the plurality of target months are determined.

[0141] Specifically, in step 401, the feature importance values corresponding to each preset image feature category in each target period are obtained by evaluating the feature importance through random forest. The feature importance values represent the importance of each preset image feature category in each target period.

[0142] In step 403, the months in a year are screened using the feature importance values. Those months that perform most important in each target period are selected as the target months of interest.

[0143] By screening the important target months, more attention can be paid to the remote sensing image data in these months, which can reduce the amount of data input to the classification model and improve the operation efficiency.

[0144] In step 404, the time series feature set corresponding to each target image feature category is obtained by performing feature extraction on the time series remote sensing image set of the target region based on the plurality of target image feature categories and the plurality of target months.

[0145] In step 405, the time series feature set corresponding to each target image feature category is input to the target random forest classification model, and the classification result output by the target random forest classification model is obtained.

[0146] In step 406, the first farmland shelterbelt image is extracted from the remote sensing image of the target region based on the classification result.

[0147] Optionally, in the above example A, the 26 target periods can be numbered in time sequence, the number of the target period ranked first in time sequence is 0, and the number of the target period ranked last in time sequence is 25. The feature importance values corresponding to each preset image feature category in each target period can be obtained by evaluating the feature importance through random forest based on the plurality of preset image feature categories and the time series remote sensing image set. Figure 5 is a schematic diagram of the random forest feature importance provided by the present application, as shown in Figure 5 the horizontal axis represents the image feature category, the vertical axis represents the number of the target period (from 0 to 25), and the number in the box represents the feature importance.

[0148] Optionally, the feature importance can be assessed by a random forest, which can be achieved by the following steps A1 to A5.

[0149] A1. Data Preparation: First, divide the dataset into training and testing sets, and preprocess the data, such as data cleaning, normalization or standardization, etc.

[0150] A2. Build Random Forest Model: Using the training set data, build a random forest classification or regression model. Random forest is composed of multiple decision trees, each of which is trained using a different subset of features.

[0151] A3. Feature Importance Assessment: Assess the importance of each feature through the feature importance indicators in the random forest model. Commonly used feature importance indicators include Gini Importance and Mean Decrease Impurity, etc.

[0152] Gini Importance: Gini importance is to calculate the number of times each feature is used to divide nodes in the random forest, and normalize it to evaluate the importance of the feature. The higher the importance value, the more important the feature.

[0153] Mean Decrease Impurity: Mean Decrease Impurity is to calculate the average value of the impurity reduction when each feature is used to divide nodes in the random forest to evaluate the importance of the feature. The more impurity reduction, the more important the feature.

[0154] A4. Feature Importance Ranking: According to the value of the feature importance indicator, rank the features from high to low to determine the importance order of each feature.

[0155] A5. Result Analysis: According to the results of feature importance ranking, it can be judged which features play a key role in the prediction ability of the model, and then feature selection, feature engineering or model optimization, etc. are carried out.

[0156] It should be noted that feature importance assessment is based on the internal mechanism of the random forest model, which can help analyze the contribution of features to the model.

[0157] Optionally, in the above example A, as Figure 5As shown, from the time perspective, the high importance values are mainly concentrated in April-May (immature crops) and September-October (harvested crops), indicating that the classification effect of farmland shelterbelt is better in this period; the low values are mainly concentrated in December-January (ice and snow cover) and June-July (farmland shelterbelt and crops grow well, and it is difficult to distinguish), indicating that the classification effect is poor in this period. Therefore, it can be determined that the target months include April, May, September and October, which can improve the operation efficiency compared to analyzing the remote sensing images of all months of the year through the classification model.

[0158] Optionally, in the above example A, as shown in Figure 5 As shown from the feature perspective, the high importance features are mainly gcvi, lswi, msr705, ndvi, red, DSM slope and slope, and the low importance features are mainly the texture features calculated according to ndvi. Because the importance of the texture features is far below the average level, the texture features are removed, which can improve the operation efficiency compared to analyzing all image feature categories in Table 1 through the classification model.

[0159] Optionally, in the above example A, in order to obtain the data set for training the random forest model, 300 sample points of each of farmland, farmland shelterbelt, woodland, building land and water body can be manually sampled from Google Maps in the above data source based on Google Maps to obtain the ground object classification labels corresponding to each sample point. Based on the coordinate positions of each sample point, the image feature samples corresponding to each preset image feature category in Table 1 can be extracted from Sentinel-2, SRTM DEM and ALOS DSM in the above data source. Then for each sample point, the ground object classification label and the image feature samples corresponding to each preset image feature category can be obtained, and the ground object classification label and the image feature samples are summarized to obtain the data set for training the classification model. Among them, 70% of the data set can be used as the training set and 30% of the data set can be used as the test set.

[0160] Based on the above data set, the random forest model can be supervised trained to obtain the above target random forest classification model.

[0161] Optionally, according to the farmland shelterbelt image extraction method provided by the application, the plurality of preset image feature categories are filtered based on the feature importance values corresponding to each target period of each preset image feature category, and the plurality of target image feature categories are determined, comprising:

[0162] For each preset image feature category, based on the feature importance values corresponding to each target time period, the feature importance average value corresponding to the preset image feature category is determined by calculating the average value.

[0163] Based on the first feature importance threshold and the feature importance average value corresponding to each preset image feature category, the plurality of preset image feature categories are screened to determine the plurality of target image feature categories.

[0164] Specifically, for each preset image feature category, based on the feature importance values corresponding to each target time period, the feature importance average value corresponding to each preset image feature category is determined by calculating the average value. In this way, a more comprehensive evaluation can be obtained, reflecting the average importance of each feature category in all target time periods. A first feature importance threshold can be set as a judgment standard. If the feature importance average value of a certain preset image feature category exceeds or equals the threshold, this feature category is considered an important target image feature category. Through such a screening process, the feature categories that are more important to the target time period can be selected, thereby focusing more on the analysis and research of these feature categories. This helps to improve the efficiency and accuracy of data analysis.

[0165] Optionally, according to the image extraction method of the farmland shelter forest provided by the present application, the months in a year are screened based on the feature importance values corresponding to each target time period to determine the plurality of target months, comprising:

[0166] For each month, based on the second feature importance threshold and the feature importance values corresponding to each target time period, the target category number of the preset image feature categories whose feature importance values are greater than the second feature importance threshold is counted to determine the target category number corresponding to the month;

[0167] Based on the category number threshold and the target category number corresponding to each month, the months in a year are screened to determine the plurality of target months.

[0168] Specifically, a second feature importance threshold can be set as a judgment standard. If the feature importance value of a certain preset image feature category in a certain month is greater than or equal to the threshold, this feature category is considered a more important target category.

[0169] By counting the number of target categories greater than the second feature importance threshold, it can be analyzed how many important target categories are in each month. A category number threshold can be set as a screening criterion. If the number of target categories in a month is greater than or equal to the category number threshold, then this month is considered to be an important target month. Through such a screening process, the most important months for target analysis and research can be selected, so as to focus more on data analysis and research of these months. This helps to improve the efficiency and accuracy of analysis.

[0170] Optionally, according to the image extraction method of the farmland shelterbelt provided by the application, after the first farmland shelterbelt image is extracted from the remote sensing image of the target area based on the classification result, the method further comprises:

[0171] Based on the known ground object type data, the farmland shelterbelt image is processed by a mask removal method to obtain a second farmland shelterbelt image.

[0172] The connected pixels in the second farmland shelterbelt image whose pixel number is less than a pixel number threshold are filtered out, and the holes in the second farmland shelterbelt image are eliminated to obtain a third farmland shelterbelt image.

[0173] Specifically, Figure 6 is a fourth flowchart of the image extraction method of the farmland shelterbelt provided by the application, as Figure 6 shown, the extraction method comprises steps 601 to 605.

[0174] Step 601: Based on a plurality of target image feature categories and a plurality of target months, feature extraction is performed on a time series remote sensing image set of a target area to obtain a time series feature set corresponding to each target image feature category.

[0175] Step 602: Input the time series feature set corresponding to each target image feature category into a target random forest classification model to obtain a classification result output by the target random forest classification model.

[0176] Step 603: Based on the classification result, a first farmland shelterbelt image is extracted from a remote sensing image of a target area.

[0177] Step 604: Based on known ground object type data, a farmland shelterbelt image is processed by a mask removal method to obtain a second farmland shelterbelt image.

[0178] Specifically, the forest and building land in the classification result can be removed by mask based on the known ground object type data, so as to exclude the case of incorrectly identifying the forest and building land as farmland shelterbelt, and improve the accuracy of the farmland shelterbelt image.

[0179] In step 605, connected pixels with a pixel number less than a pixel number threshold in the second shelterbelt image are filtered out, and holes in the second shelterbelt image are eliminated to obtain a third shelterbelt image.

[0180] Optionally, in the above example A, there are many connected pixels with a small pixel number based on the classification result of the pixels, and these connected pixels are not shelterbelts. All connected pixels can be detected first, and then a pixel number threshold 5 pixels is set to filter out connected pixels with a pixel number less than the pixel number threshold. The classification of the pixels can also cause some abnormal holes in the shelterbelt. The holes are filled by using a closing operation. The closing operation is first dilated and then eroded, and the two operations use the same structural element, which can eliminate small holes and will not change the shape of the object obviously.

[0181] Optionally, Figure 7 is a shelterbelt distribution map provided by the present application. In the above example A, connected pixels with a pixel number less than a pixel number threshold in the second shelterbelt image are filtered out, and holes in the second shelterbelt image are eliminated to obtain a third shelterbelt image as shown in Figure 7 .

[0182] Optionally, according to the method for extracting a shelterbelt image provided by the present application, after the connected pixels with a pixel number less than a pixel number threshold in the second shelterbelt image are filtered out and the holes in the second shelterbelt image are eliminated to obtain a third shelterbelt image, the method further includes:

[0183] performing land parcel segmentation processing based on the remote sensing image of the target area to obtain a land parcel segmentation result image;

[0184] performing image fusion based on the land parcel segmentation result image and the third shelterbelt image to obtain a land parcel surrounding image, the land parcel surrounding image being used to represent a state in which a land parcel is surrounded by a shelterbelt;

[0185] performing evaluation on the shelterbelt of the target area based on the land parcel surrounding image to obtain at least one shelterbelt parameter;

[0186] The at least one shelterbelt parameter includes one or more of the following parameters: land parcel surrounding degree, shelterbelt angle, and effective rate of main-harm wind prevention.

[0187] Specifically, after the land parcel surrounding image is obtained, the total length of the shelterbelt around the land parcel and the perimeter of the land parcel can be obtained by analyzing the land parcel surrounding image, and then a ratio between the total length of the shelterbelt around the land parcel and the perimeter of the land parcel can be calculated to determine the land parcel surrounding degree.

[0188] Optionally, Figure 8is a schematic diagram of the plot segmentation result provided by the present application, in the above example A, based on the remote sensing image of the target area, the plot segmentation processing can be carried out, and the plot segmentation result image as shown in Figure 8

[0189] Optionally, Figure 9 is a schematic diagram of the plot surrounding situation provided by the present application, in the above example A, based on the plot segmentation result image and the third farmland shelter forest image, image fusion is carried out, and the plot surrounding image as shown in Figure 9

[0190] Specifically, after obtaining the plot surrounding image, the angle between the shelter forest direction and the vertical direction of the main harmful wind can be calculated based on the plot surrounding image to determine the forest belt deflection angle. If the forest belt deflection angle is less than 45°, it indicates that the windproof effect of the farmland shelter forest is better.

[0191] Specifically, after obtaining the plot surrounding image, the projection length of the shelter forest in the vertical direction of the main harmful wind can be calculated based on the plot surrounding image, and then the ratio between the projection length and the actual length of the shelter forest can be calculated to determine the effective rate against the main harmful wind.

[0192] The extraction method of the farmland shelter forest image provided by the present application can extract the time sequence feature set corresponding to each target image feature category from the multi-temporal data by performing feature extraction on the time sequence remote sensing image set of the target area based on multiple target image feature categories and multiple target months, and then input the time sequence feature set corresponding to each target image feature category into the target random forest classification model, classify each pixel by the target random forest classification model, obtain the classification result, and the classification result can represent the ground object classification corresponding to each pixel in the remote sensing image of the target area. Then the farmland shelter forest image can be extracted from the remote sensing image of the target area. By performing feature extraction on the multi-temporal data, the extracted features can fully represent the seasonal variation and growth dynamics of the farmland shelter forest, which helps the classification model to identify the ground object classification of the pixel, and improves the accuracy of extracting the farmland shelter forest information from the remote sensing image.

[0193] The extraction device of the farmland shelter forest image provided by the present application is described below. The extraction device of the farmland shelter forest image described below can be correspondingly referred to the extraction method of the farmland shelter forest image described above.

[0194] Figure 10 is a structural schematic diagram of the extraction device of the farmland shelter forest image provided by the present application, as shown in Figure 10 The extraction device comprises a feature extraction module 1001, a ground object classification module 1002 and an image extraction module 1003, wherein: ​​

[0195] The feature extraction module 1001 is configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, and the image feature category is a spectral feature category, a terrain feature category or a texture feature category.

[0196] The feature extraction module 1001 is configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, and the image feature category is a spectral feature category, a terrain feature category or a texture feature category.

[0197] The image extraction module 1003 is configured to extract a first farmland shelterbelt image from the remote sensing image of the target region based on the classification result.

[0198] Figure 11 is the entity structure schematic diagram of the electronic equipment provided by the application, as Figure 11 shown, the electronic equipment can include: processor 1110, communications interface (Communications Interface) 1120, memory 1130 and communication bus 1140, wherein, the processor 1110, the communications interface 1120, the memory 1130 pass through the communication bus 1140 complete mutual communication.Processor 1110 can call the logic instruction in memory 1130, to execute the extraction method of farmland shelterbelt image, the method includes:

[0199] The feature extraction module 1001 is configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, and the image feature category is a spectral feature category, a terrain feature category or a texture feature category.

[0200] The feature extraction module 1001 is configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, and the image feature category is a spectral feature category, a terrain feature category or a texture feature category.

[0201] The image extraction module 1003 is configured to extract a first farmland shelterbelt image from the remote sensing image of the target region based on the classification result.

[0202] In addition, the logic instructions in the memory 1130 described above can be implemented in the form of a software function unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes various media that can store program codes, such as a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, etc.

[0203] The device embodiments described above are only illustrative, wherein the units illustrated as separate components can or can not be physically separated, and the components illustrated as units can or can not be physical units, i.e., can be located in one place or distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0204] From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be realized by means of software and necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions essentially or the parts that contribute to the prior art can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes several instructions for making a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some parts of the embodiments.

[0205] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. An agricultural field shelterbelt image extraction method, characterized by, The method comprises the following steps: performing feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, to obtain a time series feature set corresponding to each target image feature category, wherein the image feature category is a spectral feature category, a terrain feature category or a texture feature category; inputting the time series feature set corresponding to each target image feature category into a target random forest classification model, to obtain a classification result output by the target random forest classification model, wherein the classification result is used to represent a ground object classification corresponding to each pixel in a remote sensing image of the target region; extracting a first farmland shelterbelt image from the remote sensing image of the target region based on the classification result; after the first farmland shelterbelt image is extracted from the remote sensing image of the target region based on the classification result, the method further comprises the following steps: processing the farmland shelterbelt image by a mask removal method based on known ground object type data, to obtain a second farmland shelterbelt image; filtering out connected pixels with a pixel number less than a pixel number threshold in the second farmland shelterbelt image, and eliminating holes in the second farmland shelterbelt image, to obtain a third farmland shelterbelt image; after the connected pixels with the pixel number less than the pixel number threshold in the second farmland shelterbelt image are filtered out, and the holes in the second farmland shelterbelt image are eliminated, to obtain the third farmland shelterbelt image, the method further comprises the following steps: performing land parcel segmentation processing based on the remote sensing image of the target region, to obtain a land parcel segmentation result image; performing image fusion based on the land parcel segmentation result image and the third farmland shelterbelt image, to obtain a land parcel surrounding image, wherein the land parcel surrounding image is used to represent a state in which a land parcel is surrounded by a shelterbelt; evaluating the farmland shelterbelt of the target region based on the land parcel surrounding image, to obtain at least one farmland shelterbelt parameter; the at least one farmland shelterbelt parameter comprises one or more of the following parameters: a land parcel surrounding degree, a shelterbelt angle and an effective rate of preventing a main harmful wind; the land parcel surrounding degree is determined by analyzing the land parcel surrounding image, obtaining a total length of farmland shelterbelts around a land parcel and a perimeter of the land parcel, and then calculating a ratio between the total length of the farmland shelterbelts around the land parcel and the perimeter of the land parcel; the shelterbelt angle is determined based on the land parcel surrounding image, by calculating an included angle between a direction of the shelterbelt and a vertical direction of the main harmful wind; the effective rate of preventing the main harmful wind is determined based on the land parcel surrounding image, by calculating a projection length of the shelterbelt in the vertical direction of the main harmful wind, and then calculating a ratio between the projection length and an actual length of the shelterbelt.

2. The method of claim 1, wherein the image of the shelterbelt is extracted by using a method of image processing. before the feature extraction on the time series remote sensing image set of the target region based on the plurality of target image feature categories and the plurality of target months, to obtain the time series feature set corresponding to each target image feature category, the method further comprises the following step: performing multiple image fusion on the remote sensing image of the target region in time sequence based on a preset time step, to obtain a plurality of fused remote sensing images corresponding to a plurality of target time periods, wherein one image fusion is used to process a plurality of remote sensing images within one preset time step, and a time length of the target time period is equal to the preset time step. Determine the time series remote sensing image set based on the post-fusion remote sensing images corresponding to the target time periods.

3. The method of claim 2, wherein the image of the shelterbelt is extracted by using a method of extracting an image of a shelterbelt from a digital orthoimage. The image fusion comprises: Based on the preset time step, a plurality of to-be-fused remote sensing images in a target time period are obtained; Based on the plurality of to-be-fused remote sensing images, a plurality of pixel remote sensing data corresponding to each pixel are obtained; For each pixel, the target pixel remote sensing data with the highest normalized difference vegetation index (NDVI) is selected from the plurality of pixel remote sensing data corresponding to the pixel; Based on the target pixel remote sensing data corresponding to each pixel, a post-fusion remote sensing image corresponding to the target time period is generated.

4. The method of claim 2, wherein the image of the shelterbelt is extracted by using a method of extracting an image of a shelterbelt from a digital orthoimage. Before performing feature extraction on the time series remote sensing image set of the target region based on a plurality of target image feature categories and a plurality of target months, and obtaining a time series feature set corresponding to each target image feature category, the method further comprises: Based on a plurality of preset image feature categories and the time series remote sensing image set, the feature importance of each preset image feature category in each target time period is obtained by random forest evaluation; Based on the feature importance values of each preset image feature category in each target time period, the plurality of preset image feature categories are screened to determine the plurality of target image feature categories; Based on the feature importance values of each preset image feature category in each target time period, the months in a year are screened to determine the plurality of target months.

5. The method of claim 4, wherein the image of the shelterbelt is extracted by using a method of extracting an image of a shelterbelt from a digital orthoimage. The screening of the plurality of preset image feature categories based on the feature importance values of each preset image feature category in each target time period to determine the plurality of target image feature categories comprises: For each preset image feature category, the average value of the feature importance of the preset image feature category corresponding to each target time period is determined by calculating the average value based on the feature importance values of the preset image feature category in each target time period; Based on the first feature importance threshold and the average value of the feature importance of each preset image feature category, the plurality of preset image feature categories are screened to determine the plurality of target image feature categories.

6. The method of claim 4, wherein the image of the shelterbelt is extracted by using a method of extracting an image of a shelterbelt from a digital orthoimage. The screening of the months in a year based on the feature importance values of each preset image feature category in each target time period to determine the plurality of target months comprises: For each month, the target category number of the preset image feature categories whose feature importance values are greater than the second feature importance threshold is determined based on the second feature importance threshold and the feature importance values of each preset image feature category in each target time period, and the target category number corresponding to the month is determined; Based on the category number threshold and the target category number corresponding to each month, the months in a year are screened to determine the plurality of target months.

7. A device for extracting images of farmland shelterbelts, characterized in that, The method comprises: a feature extraction module configured to perform feature extraction on a time series remote sensing image set of a target region based on a plurality of target image feature categories and a plurality of target months, and obtain a time series feature set corresponding to each target image feature category, wherein the image feature category is a spectral feature category, a terrain feature category, or a texture feature category; The ground object classification module is configured to input a time series feature set corresponding to each target image feature category into a target random forest classification model, and obtain a classification result output by the target random forest classification model, where the classification result is used to represent a ground object classification corresponding to each pixel in a remote sensing image of a target region. The image extraction module is configured to extract a first farmland shelterbelt image from the remote sensing image of the target region based on the classification result. After the first farmland shelterbelt image is extracted from the remote sensing image of the target region based on the classification result, the device is further configured to: process the farmland shelterbelt image by a mask removal method based on known ground object type data, to obtain a second farmland shelterbelt image; filter out connected pixels with a pixel number less than a pixel number threshold in the second farmland shelterbelt image, and eliminate holes in the second farmland shelterbelt image, to obtain a third farmland shelterbelt image; After the connected pixels with the pixel number less than the pixel number threshold in the second farmland shelterbelt image are filtered out, and the holes in the second farmland shelterbelt image are eliminated, to obtain the third farmland shelterbelt image, the device is further configured to: perform land parcel segmentation processing based on the remote sensing image of the target region, to obtain a land parcel segmentation result image; perform image fusion based on the land parcel segmentation result image and the third farmland shelterbelt image, to obtain a land parcel surrounding image, where the land parcel surrounding image is used to represent a state in which a land parcel is surrounded by a shelterbelt; evaluate the farmland shelterbelt of the target region based on the land parcel surrounding image, to obtain at least one farmland shelterbelt parameter; The at least one farmland shelterbelt parameter includes one or more of the following parameters: a land parcel surrounding degree, a shelterbelt skew angle, and a main-harm-wind protection efficiency. The land parcel surrounding degree is determined by analyzing the land parcel surrounding image, obtaining a total length of farmland shelterbelts around a land parcel and a perimeter of the land parcel, and then calculating a ratio between the total length of the farmland shelterbelts around the land parcel and the perimeter of the land parcel. The shelterbelt skew angle is determined based on the land parcel surrounding image, by calculating an included angle between a shelterbelt direction and a main-harm-wind vertical direction. The main-harm-wind protection efficiency is determined based on the land parcel surrounding image, by calculating a projection length of a shelterbelt in a main-harm-wind vertical direction, and then calculating a ratio between the projection length and an actual length of the shelterbelt.

8. An electronic device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The processor implements the program to implement the farmland shelterbelt image extraction method according to any one of claims 1 to 6.

Images