Farmland identification method, device, equipment and computer-readable storage medium

Abstract

The present invention relates to a method, device, equipment and computer-readable storage medium for identifying farmland, and the method comprises: obtaining a current radar remote sensing image data set and a current optical remote sensing image data set corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area; performing optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, performing radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, and determining a current feature set of the study area based on the current optical feature information and the current radar feature information; and obtaining a farmland identification result of the study area based on the current feature set and a preset farmland identification model, wherein the farmland identification result includes a highland barley farmland area and a rapeseed farmland area. The present application has the effect of realizing farmland identification of artificial grasslands with similar spectral features and texture features.

Description

Technical Field

[0001] The present invention relates to the technical field of crop classification, and in particular to a method, device, equipment and computer-readable storage medium for identifying farmland and pasture. Background Art

[0002] As the demand for modernization and precision in agricultural production continues to increase, quickly and accurately obtaining information on crop planting area and type is crucial for agricultural management and decision-making.

[0003] The current crop classification is to achieve high-precision classification and detection of crops and other targets in remote sensing images through Shenjiang network technology.

[0004] However, there are still some challenges in identifying special landforms in specific areas, such as farmland and pasture in the Qinghai-Tibet Plateau. Due to the unique geographical environment and complex climatic conditions, farmland and pasture types are diverse and complex in distribution, including artificial grasslands such as highland barley and rapeseed. These special crop types often show similar spectral and texture characteristics in remote sensing images, making their classification and identification particularly difficult. Therefore, there is an urgent need for a farmland identification method that can identify artificial grasslands such as highland barley and rapeseed. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method, device, equipment and computer-readable storage medium for identifying farmland and pasture, aiming to solve at least one of the above-mentioned technical problems.

[0006] The technical solution of the present invention to solve the above technical problems is as follows:

[0007] In the first aspect, the present application provides a method for identifying agricultural and pastoral land, which adopts the following technical solutions:

[0008] A method for identifying agricultural and pasturage land, comprising:

[0009] Acquire a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area;

[0010] Performing optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI and wet;

[0011] Performing radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index;

[0012] Based on the current optical feature information, the current radar feature information and the preset farmland identification model, the farmland identification result of the study area is obtained, and the farmland identification result includes highland barley farmland area and rapeseed farmland area. The preset farmland identification model is obtained by training samples of multiple surface land use areas of different types, and the multiple surface land use areas include at least highland barley farmland area and rapeseed farmland area. The samples are generated based on historical radar remote sensing image datasets, historical optical remote sensing image datasets and ground survey data of the study area in a set time period.

[0013] The beneficial effect of the present invention is that by combining the optical remote sensing image data set and the radar remote sensing image data set, the method can obtain multi-source data of the study area, thereby more comprehensively reflecting the surface characteristics. In particular, in the study of mountainous areas, the combination of optical data and radar data can complement each other's missing information.

[0014] Specific optical characteristic indices, such as NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, wet, etc., as well as radar characteristic information such as surface roughness index and soil moisture index, are used to distinguish different types of surface cover. In particular, indices such as NDYI, GRVI, NRFI, and PSRI show great potential for distinguishing rapeseed from other vegetation, thereby improving the separability between highland barley and rapeseed from other vegetation. In the feature extraction process, the introduction of terrain characteristics (such as terrain index time series information) further enhances the accuracy of classification.

[0015] The agricultural and pastoral land identification model trained based on multi-source data and samples of various surface land use areas has strong generalization ability and classification accuracy, which means that the model can not only accurately identify highland barley and rapeseed agricultural and pastoral land areas, but also better classify other types of surface cover.

[0016] Based on the above technical solution, the present invention can also be improved as follows.

[0017] Furthermore, the method for establishing the preset farmland identification model includes:

[0018] Acquire a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period, wherein the ground survey data includes at least one crop planted in the study area and the longitude and latitude information corresponding to the crop, wherein the crop includes at least highland barley and rapeseed;

[0019] Determine the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset;

[0020] Based on the phenological characteristics and the ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to the various first land uses, wherein the sample sets include multiple radar image sub-datasets and multiple optical image sub-datasets;

[0021] Based on a preset sample generation rule, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for a second time to obtain sample sets corresponding to the plurality of second land uses;

[0022] The first land use and the second land use are both regarded as surface land use areas, and radar feature recognition is performed on a plurality of radar image sub-datasets corresponding to each surface land use area to obtain radar feature information corresponding to the study area;

[0023] Performing optical feature recognition on a plurality of optical image sub-datasets corresponding to each surface land use area to obtain optical feature information corresponding to the study area;

[0024] generating a feature set of the study area based on the optical feature information and the radar feature information;

[0025] The feature set is divided into a training set and a validation set, and the training set is input into a pre-established classification model for training until a farmland identification model that meets set conditions is obtained.

[0026] The beneficial effect of adopting the above further scheme is: considering that natural vegetation and artificial vegetation are similar in spectral characteristics in most cases, the harmonic analysis method is first used to extract the phenological information of crops, and then the required vegetation index is determined based on the phenological information. By considering the phenological characteristics of crops, the model can more accurately capture the spectral changes of different crops during the growth cycle, thereby improving the recognition accuracy of crops, especially highland barley and rapeseed. Combined with the backscatter coefficient of radar data, the model can capture information that may be missing in optical data, such as surface features under cloud cover, thereby further enhancing classification accuracy.

[0027] The first and second screening processes ensure the diversity and representativeness of the sample set, which helps the model learn more comprehensive surface features. Not only samples of the first land use (areas for growing barley and rapeseed) but also samples of the second land use (such as forest land, water area, urban land, rural settlements, other construction land, sandy land, Gobi, saline-alkali land, bare land, bare rock texture, etc.) need to be considered to improve the generalization ability of the model.

[0028] Furthermore, determining the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset includes:

[0029] Based on the historical optical remote sensing image data set, a time series of multiple target indices is calculated, wherein the multiple target indices are NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, wet, TGSI, BSI, NDBSI, NDWI, MNDWI, DEM, Slope and Aspect;

[0030] Based on a preset harmonic analysis algorithm and the time series of each of the target indexes, a phenological characteristic curve corresponding to each of the target indexes is generated;

[0031] The phenological characteristic curve of each target index is fitted to determine the phenological characteristic information corresponding to each crop in the study area, wherein the phenological characteristic information includes the growth period.

[0032] The beneficial effect of adopting the above further scheme is that by calculating the time series of multiple target indices (such as NDYI, SAVI, VARI, etc.), the changes in the spectral characteristics of crops at different growth stages can be captured. These indices reflect information such as vegetation coverage and growth conditions of crops, providing a rich data basis for accurately extracting phenological characteristics. The generated phenological characteristic curve can intuitively show the growth changes of crops in a year, including key stages such as the beginning of growth, peak growth, and end of growth. These phenological characteristic information is of great significance for distinguishing different types of crops and improving classification accuracy.

[0033] Furthermore, based on the phenological characteristics and ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to a plurality of first land uses, including:

[0034] Based on the phenological characteristics, determining the first radar image data and the first optical image data corresponding to each crop in the growth period;

[0035] Based on the ground survey data, the first radar image data and the first optical image data corresponding to each crop in the growth period are clipped to obtain the second radar image data and the second optical image data corresponding to each crop;

[0036] For each of the crops, the second radar image data and the second optical image data corresponding to the crop are divided according to a preset time period to obtain sample sets corresponding to the multiple first land uses.

[0037] The beneficial effects of adopting the above further scheme are: determining the specific time period of each crop during the growth period through phenological characteristics, and screening out the corresponding radar image data and optical image data accordingly, thereby ensuring the pertinence and relevance of the data; combining with ground survey data, the screened radar image data and optical image data are cropped to obtain the image data corresponding to each crop. This cropping method based on the actual geographical location can remove redundant information and retain areas directly related to crop growth, thereby enhancing the accuracy of the data. The image data corresponding to the crops are divided according to the preset time period to obtain sample sets corresponding to multiple first land uses. This division method not only takes into account the different stages of crop growth, but also covers the growth of crops under different geographical locations and climatic conditions, thereby enriching the diversity of the sample set.

[0038] Furthermore, based on the preset sample generation rules, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the second time to obtain sample sets corresponding to the various second land uses, including:

[0039] Obtain land cover information of the study area in different years;

[0040] Carry out a patch analysis on the land use information of each year to determine the center point coordinates of each land use type patch corresponding to each year;

[0041] Determine the target point based on the coordinates of the center point of each land use type map corresponding to each year;

[0042] Based on the target point, the historical radar remote sensing image data set and the historical optical remote sensing image data set are screened for the second time to obtain sample sets corresponding to the various second land uses.

[0043] The beneficial effect of adopting the above further scheme is that by determining the center point coordinates of the land use type map of each year, the representative areas of different land use types can be accurately located. The sample set clipped based on these center point coordinates will directly reflect the characteristics of these land use types, thereby enhancing the pertinence of the data.

[0044] Furthermore, the processing of the pre-established classification model includes:

[0045] Extracting features from the image in the feature set by using a convolutional neural network to generate feature maps with different spatial resolutions and abstraction levels;

[0046] The feature map is passed through three independent one-dimensional convolutional layers to generate a query matrix, a key matrix and a value matrix respectively, wherein the query matrix represents the role of each position or feature point in the feature map as a query in the attention calculation, and the query role is the reference feature in the study area; the key matrix represents the role of each position or feature point in the feature map as a key point in the attention calculation, and the key point is the target feature in the study area related to the reference feature; and the value matrix represents the associated information of each key point;

[0047] Based on the generated query matrix, the key matrix and the value matrix, an attention weight distribution is obtained, where the attention weight distribution represents the correlation between different positions or feature points in the feature map;

[0048] Determine the weight of the element corresponding to each feature map based on the attention weight distribution to obtain an attention result;

[0049] The attention result and the original image are feature fused, and the size is adjusted through a cascade upsampler to output a feature image with the same full resolution as the original image, so as to classify the feature image through a classifier to obtain a classification result.

[0050] The beneficial effects of adopting the above further scheme are:

[0051] By extracting features using a convolutional neural network (CNN), the model can automatically learn feature maps of different spatial resolutions and abstraction levels in the image. This feature extraction method is not only efficient, but also can capture multi-scale information in the image, which is particularly important for processing complex remote sensing image data.

[0052] The attention weight distribution is generated by using the query, key and value mechanism, and the weight of each element corresponding to the feature map is determined accordingly. This attention mechanism enables the model to focus on the key areas in the image. By reducing the attention to non-key areas, the model can improve the accuracy and efficiency of classification.

[0053] Feature fusion of the attention results and the original feature maps helps enhance the model's ability to identify important features in the image. By combining the attention results and the original features, the model can make full use of the information in the image and improve the reliability of classification.

[0054] By resizing through cascaded upsamplers, the model is able to output full-resolution images. This full-resolution output allows the model to provide more refined classification results while preserving image details.

[0055] Further, after obtaining the farmland identification result of the study area based on the current optical feature information, the current radar feature information and the preset farmland identification model, the method further includes:

[0056] Obtaining crop planting area information of the study area in a set time period;

[0057] Performing a first evaluation on the farmland identification result based on the crop sowing area information to obtain first evaluation information;

[0058] Inputting the verification set into a preset farmland identification model to obtain a verification sample classification result;

[0059] Constructing a confusion matrix based on the verification sample classification results and the farmland identification results;

[0060] Performing a second evaluation on the farmland identification result based on the mixed matrix to obtain second evaluation information;

[0061] The farmland identification model is evaluated based on the first evaluation information and the second evaluation information.

[0062] The beneficial effect of adopting the above further scheme is that by obtaining the crop planting area information of the study area in the set time period and comparing it with the farmland identification results, the accuracy of the identification results can be first evaluated based on the actual data. This evaluation method based on actual data is more intuitive and reliable. By inputting the verification set into the preset farmland identification model, the verification sample classification results are obtained, and compared with the farmland identification results, the performance of the model on unknown data can be evaluated.

[0063] In the second aspect, the present application provides a farmland identification device, which adopts the following technical solution:

[0064] A farmland identification device, comprising:

[0065] An acquisition module, used to acquire a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area;

[0066] An optical feature recognition module is used to perform optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI and wet;

[0067] A radar feature recognition module is used to perform radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index;

[0068] An identification module is used to obtain an agricultural and pastoral land identification result of the study area based on the current optical feature information, the current radar feature information and a preset agricultural and pastoral land identification model, wherein the agricultural and pastoral land identification result includes a highland barley agricultural and pastoral land area and a rapeseed agricultural and pastoral land area. The preset agricultural and pastoral land identification model is obtained by training samples of various surface land use areas of different types, wherein the various surface land use areas include at least a highland barley agricultural and pastoral land area and a rapeseed agricultural and pastoral land area, and the samples are generated based on a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and a ground survey data of the study area in a set time period.

[0069] In a third aspect, the present application provides an electronic device, which adopts the following technical solution:

[0070] An electronic device comprises a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and execute the farmland identification method described in any one of the first aspects.

[0071] In a fourth aspect, the present application provides a computer-readable storage medium, which adopts the following technical solution:

[0072] A computer-readable storage medium stores a computer program that can be loaded by a processor and execute the farmland identification method described in any one of the first aspects.

[0073] Additional aspects and advantages of the present application will be partially given in the following description, which will become apparent from the following description, or will be understood through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] Figure 1 A schematic flow chart of a method for identifying agricultural and pastoral land provided by an embodiment of the present invention;

[0075] Figure 2 A schematic diagram of the structure of a farmland identification device provided by an embodiment of the present invention;

[0076] Figure 3 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0077] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0078] In addition, the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally means that the associated objects before and after are in an "or" relationship.

[0079] An embodiment of the present application provides a method for identifying farmland and pasture, which can be executed by an electronic device, which can be a server or a mobile terminal device, wherein the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services; the mobile terminal device can be a laptop computer, a desktop computer, etc., but is not limited to this.

[0080] The following is a further detailed description of the embodiments of the present application in conjunction with the accompanying drawings. Figure 1 As shown, a method for identifying agricultural and pastoral land includes steps S1 to S5:

[0081] Step S1, obtaining a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area;

[0082] In the present implementation method, the study area is a region with complex terrain and a large amount of natural grassland and shrubs, such as the Qinghai-Tibet Plateau, where barley and rapeseed are mixedly planted, and barley and rapeseed are easily confused in classification.

[0083] The current radar remote sensing image dataset can be a radar remote sensing image dataset of any year. The radar remote sensing image dataset includes Sentinel-1 satellite remote sensing data. Sentinel-1 satellite remote sensing images need to be processed by radiation correction, geometric correction and terrain correction. Vertical launch vertical reception (VV) and vertical launch horizontal reception (VH) backscatter data are selected as auxiliary information. Radar remote sensing image datasets can be downloaded through open access platforms such as ESACopernicus and Google earth engine.

[0084] The current optical remote sensing image dataset can be any year's optical remote sensing image dataset, including Sentinel-2A and Sentinel-2B satellite remote sensing data, which need to be atmospherically corrected and radiated. Its multispectral data is used, with 13 bands from visible light to near infrared, of which 4 bands have a spatial resolution of 10m and 6 bands have a spatial resolution of 20m. The optical remote sensing image dataset can be downloaded through open access platforms such as NASA Earthdata, ESACopernicus, and Google earth engine.

[0085] Step S2, performing optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, and wet;

[0086] In the implementation manner of the present application, the vegetation index time series information, the soil index time series information, the water body index time series information and the terrain index time series information are shown in Table 1.

[0087] Table 1:

[0088]

[0089] The data of each target index in Table 1 above at different time points in the same area are combined into a time series to analyze the changes in vegetation cover over time, the time series changes in soil properties, the distribution and changes of water bodies, and the impact of terrain on elements such as vegetation, soil and water bodies and their changes over time.

[0090] In Table 1, NDVI is a widely recognized vegetation sensitivity index that can be used to detect the growing period and can effectively distinguish between vegetation and non-vegetation areas; BSI can effectively reflect the mineral content in the soil and is used to monitor the bare soil period; NDWI can monitor the moisture content of soil and vegetation, thereby distinguishing between bare soil, weak vegetation areas and heavy vegetation areas, and can be used for drought detection and judgment of vegetation moisture changes; NDYI, GRVI, NRFI, and PSRI have great potential for distinguishing rapeseed from other vegetation, which improves the separability of highland barley rapeseed from other vegetation, among which GRVI is sensitive to the flowering period of rapeseed.

[0091] The current optical remote sensing image set is divided according to preset time intervals. For example, the current optical remote sensing image is divided into optical remote sensing image subsets corresponding to each monthly time series according to the month. The vegetation index, terrain index and water body index of each subset are calculated to obtain vegetation index time series information, soil index time series information, water body index time series information and terrain index time series information.

[0092] Step S3, performing radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index;

[0093] In an implementation manner of the present application, the backscattering coefficients of VV and VH polarizations of each image in the current radar remote sensing image data set are extracted to obtain the surface roughness index and the soil moisture index.

[0094] The current radar remote sensing image data set is divided according to preset time intervals. For example, the current radar remote sensing image is divided into radar remote sensing image subsets corresponding to each monthly time series according to months, and the time series information of the surface roughness index and the time series information of the soil moisture index are obtained.

[0095] Step S4, based on the current optical feature information, the current radar feature information and the preset farmland identification model, obtain the farmland identification result of the study area, the farmland identification result includes highland barley farmland area and rapeseed farmland area, the preset farmland identification model is obtained by training samples of multiple surface land use areas of different types, the multiple surface land use areas include at least highland barley farmland area and rapeseed farmland area, the samples are generated based on historical radar remote sensing image data sets, historical optical remote sensing image data sets and ground survey data of the study area in a set time period.

[0096] In the implementation mode of the present application, the current optical feature information and the current radar feature information are fused in time and space to obtain the current feature set, ensuring that the feature information from two different sources matches in space and time. The current feature set is then input into the farmland identification model, and the farmland classification model is used to realize farmland identification in any year in areas with complex terrain and a large amount of natural grassland and shrubs, so as to obtain the classification results of the corresponding highland barley area and rapeseed area.

[0097] The classification model used in the farmland identification method provided by the embodiment of the present invention is obtained based on historical radar remote sensing image data sets, historical optical remote sensing image data sets and sample training. The model can not only accurately identify highland barley and rapeseed farmland areas, but also better classify other types of surface cover. By combining the optical remote sensing image data set and the radar remote sensing image data set, the method can obtain multi-source data of the study area, thereby more comprehensively reflecting the surface characteristics. In particular, in the study of mountainous areas, the combination of optical data and radar data can complement each other's information loss.

[0098] Specific optical characteristic indices, such as NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, wet, etc., and radar characteristic information, such as surface roughness index and soil moisture index, were used to distinguish different types of land cover. In particular, indices such as NDYI, GRVI, NRFI, and PSRI showed great potential for distinguishing rapeseed from other vegetation, thereby improving the separability between highland barley and rapeseed and other vegetation.

[0099] To facilitate understanding of the farmland identification model provided in the above embodiment, an embodiment of the present invention provides a method for establishing a farmland identification model, see the following steps S51 to S58:

[0100] Step S51, obtaining a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period, wherein the ground survey data includes at least one crop planted in the study area and the latitude and longitude information corresponding to the crop, and the crop includes at least highland barley and rapeseed;

[0101] In the embodiment of the present application, historical radar remote sensing image data sets and historical optical remote sensing image data sets of multiple years are obtained through platforms such as ESA Copernicus and Google Earth Engine. The longitude and latitude information of highland barley and rapeseed are recorded by using GPS equipment or other positioning tools, and relevant crop planting information is collected.

[0102] Step S52, determining the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset;

[0103] In this implementation manner, specifically, step S52 includes:

[0104] Based on the historical optical remote sensing image data set, a time series of multiple target indices is calculated, where the multiple target indices are NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, wet, TGSI, BSI, NDBSI, NDWI, MNDWI, DEM, Slope, and Aspect;

[0105] Based on a preset harmonic analysis algorithm and the time series of each of the target indexes, a phenological characteristic curve corresponding to each of the target indexes is generated;

[0106] The phenological characteristic curve of each target index is fitted to determine the phenological characteristic information corresponding to each crop in the study area, wherein the phenological characteristic information includes the growth period.

[0107] In the above implementation, for each optical remote sensing image in the historical optical remote sensing image data set, the value of each target index is calculated, and then the calculated target index values ​​are arranged in chronological order to form a time series, which will cover the time range of the entire historical data set.

[0108] Afterwards, a harmonic analysis algorithm is applied to the time series of each target index, where the preset harmonic analysis algorithm is the Fourier sequence algorithm, which is as follows:

[0109]

[0110] Wherein, A0 represents the harmonic margin; Aj represents the amplitude of each harmonic; θj represents the initial phase of each harmonic; i=1,2,...,n, where n represents the number of points of the fitting data; m represents the harmonic order; and kj represents the harmonic frequency.

[0111] By fitting the phenological characteristic curve of each target index, key information such as growth start date, growth end date, growth rate, etc. can be extracted from the phenological characteristic curves of highland barley and rapeseed in the study area.

[0112] Step S53, based on the phenological characteristics and the ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to the various first land uses, wherein the sample sets include multiple radar image sub-datasets and multiple optical image sub-datasets;

[0113] In the implementation manner of the present application, specifically, step S53 includes:

[0114] Firstly, based on the phenological characteristics, the first radar image data and the first optical image data corresponding to each crop in the growth period are determined in the historical radar remote sensing image data set and the historical optical remote sensing image data set;

[0115] Then, based on the ground survey data, the first radar image data and the first optical image data corresponding to each crop in the growth period are clipped to obtain the second radar image data and the second optical image data corresponding to each crop;

[0116] Finally, for each of the crops, the second radar image data and the second optical image data corresponding to the crop are divided according to a preset time period to obtain sample sets corresponding to the multiple first land uses.

[0117] In the implementation mode of the present application, the first land is agricultural and pasturage land such as highland barley and rapeseed. The retrieved image data is time-matched with the growth period of the crops to ensure that these data can cover the entire growth period of the crops. The latitude and longitude information of the crop planting area provided in the ground survey data is used to accurately locate the corresponding first radar image data and the first optical image data. Using geographic information system (GIS) software or remote sensing image processing software, the image data after positioning is clipped to obtain the second radar image data and the second optical image data corresponding to each crop. These clipped image data should only contain the planting area of ​​the corresponding crop to remove the influence of other irrelevant areas. For each crop, the corresponding second radar image data and the second optical image data are divided according to the set time period to obtain a sample set corresponding to each of the multiple first lands (i.e., the planting area of ​​the crops in different time periods). The set time period is one month.

[0118] Step S54, based on a preset sample generation rule, performing a second screening on the historical radar remote sensing image dataset and the historical optical remote sensing image dataset to obtain sample sets corresponding to the plurality of second land uses;

[0119] In the implementation manner of the present application, step S54 specifically includes the following sub-steps:

[0120] Obtain land cover information of the study area in different years;

[0121] Carry out a patch analysis on the land use information of each year to determine the center point coordinates of each land use type patch corresponding to each year;

[0122] Determine the target point based on the coordinates of the center point of each land use type map corresponding to each year;

[0123] A second screening is performed on the historical radar remote sensing image dataset and the historical optical remote sensing image dataset based on the target point to obtain sample sets corresponding to the various second land uses.

[0124] In the implementation mode of the present application, the second land use is forest land, water area, urban land, rural residential area, other construction land, sandy land, Gobi, saline-alkali land, bare land, bare rock texture, etc. The center point coordinates of each land use type map corresponding to each year are used to filter the point information whose land cover attributes have not changed for many years in the center point coordinates, so that the selected target points are more representative.

[0125] Step S55, taking both the first land and the second land as surface land use areas, and performing radar feature recognition on the plurality of radar image sub-datasets corresponding to each surface land use area, to obtain radar feature information corresponding to the study area;

[0126] Step S56, performing optical feature recognition on a plurality of optical image sub-datasets corresponding to each surface land use area to obtain optical feature information corresponding to the study area;

[0127] In the implementation manner of the present application, radar feature recognition is performed on the multiple radar image sub-datasets corresponding to each surface land use area in the manner of the above-mentioned step S3, and optical feature recognition is performed on the multiple optical image sub-datasets corresponding to each surface land use area in the manner of the above-mentioned step S2, which will not be repeated here.

[0128] Step S57, generating a feature set of the study area based on the optical feature information and the radar feature information;

[0129] Step S58, dividing the feature set into a training set and a validation set, and inputting the training set into a pre-established classification model for training until a farmland identification model that meets the set conditions is obtained.

[0130] In the implementation manner of the present application, after extracting the optical feature information and radar information of the study area, it also includes: screening outlier sample points based on the optical feature information and radar information, and removing outlier sample points.

[0131] Borura, RelifF, Recursive Feature Elimination (REF), Sparrow search Algorithm (SSA) and other algorithms are used to screen the optimal features. All the screened feature points are divided into training set and validation set in a ratio of 7:3 as input data for subsequent classification models.

[0132] The following example uses the method of obtaining data on the Google Earth Engine platform and calculating feature information. The USGS Landsat 5 Level 2 Collection 2 Tier 1 dataset (including surface reflectance data and surface temperature data corrected by the LaSRC-Land Surface Reflectance Code algorithm), USGS Landsat 8 Level 2 Collection 2 Tier 1 dataset (including surface reflectance data and surface temperature data corrected by the LaSRC-Land Surface Reflectance Code algorithm), Harmonized Sentinel-2 MSI: MultiSpectral Instrument Level-2A dataset, Copernicus DEM GLO-30: Global 30m Digital Elevation Model dataset, MOD13Q1.061 Terra Vegetation Indices 16-DayGlobal 250m, and Sentinel-1SAR GRD: C-band Synthetic Aperture Radar Ground Detected, log scaling dataset are selected for relevant calculations. The feature extraction process is as follows:

[0133] For the first site, first, all optical remote sensing datasets during the growth period of highland barley and rapeseed vegetation in the study area were screened and cropped;

[0134] Afterwards, the pixel quality attribute bands (QA_PIXEL, QA60) in the optical remote sensing dataset images were removed from cirrus, cloud, cloud shadow, and snow pixels, and the median synthesis method was used to synthesize the monthly surface reflectance data and the surface reflectance data during the vegetation peak period;

[0135] Afterwards, the STARFM, ESTARFM, and FSDAF algorithms were used to fuse Landsat and MODIS data, and Sentinel-2 and MODIS datasets to obtain NDVI30m and 10m fused datasets for every 8-day long time series;

[0136] Afterwards, the vegetation index, soil index, and water index characteristics in Table 1 were calculated based on the monthly surface reflectance data and the surface reflectance data during the vegetation peak period, and the terrain index characteristics in Table 1 were calculated using the DEM dataset;

[0137] Then, the Sentinel-1 dataset was used to obtain the surface roughness and soil moisture index characteristics in the study area.

[0138] Finally, the above characteristics were organized into multiple two-dimensional data sets, which were the feature sets of the first land use, providing data support for the classification of highland barley and rapeseed in subsequent research areas.

[0139] For the second land use, firstly, the LUCC vector data of each year in the study area were extracted, the center point coordinates of each patch of the LUCC vector data were extracted, and the point information whose LUCC attributes have not changed for many years in the center point coordinates was screened. The information of the corresponding position of each feature in Table 1 was extracted using the screened sample point data to construct the feature set of the second land use.

[0140] In the above step S58, the processing process of the pre-established classification model includes:

[0141] Extracting features from the image in the feature set by using a convolutional neural network to generate feature maps with different spatial resolutions and abstraction levels;

[0142] The feature map is passed through three independent one-dimensional convolutional layers to generate query matrix, key matrix and value matrix respectively. The query matrix represents the role of each position or feature point in the feature map as a query in the attention calculation, and the query role is the reference feature in the study area. The key matrix represents the role of each position or feature point in the feature map as a key point in the attention calculation, and the key point is the target feature related to the reference feature in the study area. The value matrix represents the associated information of each key point. The position refers to the coordinate point of the image in the spatial dimension, and the feature point refers to the specific feature value at the corresponding position. The feature value may be an abstract representation of the color, texture, shape and other information of the image.

[0143] Based on the generated query matrix, the key matrix and the value matrix, an attention weight distribution is obtained, where the attention weight distribution represents the correlation between different positions or feature points in the feature map;

[0144] Determine the weight of the element corresponding to each feature map based on the attention weight distribution to obtain an attention result;

[0145] The attention result and the original image are feature fused, and the size is adjusted through a cascade upsampler to output a feature image with the same full resolution as the original image, so as to classify the feature image through a classifier to obtain a classification result.

[0146] In the implementation of the present application, the CNN-Transformer hybrid model trains the feature set, where CNN is first used as a feature extractor to generate a feature map for the input. Then, the 1×1 patch extracted from the CNN feature map is embedded through Patch embedding, and the position-encoded patch embedding is used as the input of the Transformer encoder. The encoder consists of L layers, each layer contains a multi-head self-attention mechanism (MSA) and a multi-layer perceptron (MLP), and each layer output is updated through layer normalization (LN) and residual connection.

[0147] Among them, the output of the first layer can be written as:

[0148] z l ′ =MSA(LN(z l-1 ))+z l-1 ;

[0149] z l =Mlp(LN(z l ′ ))+z l ′ ;

[0150] Where LN(·) is the layer normalization operator, z L is the encoded image representation.

[0151] Through the cascade upsampler, the encoded hidden feature sequence Reshape After the shape, CUP is instantiated by cascading multiple upsampling blocks to achieve to the full resolution of H×W, where each block consists of a 2× upsampling operator, a 3×3 convolutional layer, and a ReLU layer in sequence.

[0152] As another implementation of the embodiment of the present application, after obtaining the farmland identification result of the study area based on the current feature set and the preset farmland identification model, it also includes steps Sa to Sf:

[0153] Step Sa, obtaining crop planting area information of the study area in a set time period;

[0154] In the implementation mode of the present application, information on the sown area of ​​crops in the study area within a set time period (such as a year, a quarter or a specific month) is obtained from reliable official data sources or professional agricultural statistics agencies. This information usually exists in the form of maps, tables or databases, and contains the sown area and geographical location of various crops.

[0155] Step Sb, performing a first evaluation on the farmland identification result based on the crop sowing area information to obtain first evaluation information;

[0156] In the implementation mode of the present application, the agricultural and pastoral land plots in the identification results are spatially matched and compared with the actual sowing area data, and evaluation indicators such as recognition accuracy and recall rate are calculated.

[0157] Step Sc, inputting the verification set into a preset farmland identification model to obtain a verification sample classification result;

[0158] Step Sd, constructing a confusion matrix based on the verification sample classification results and the farmland identification results;

[0159] In the implementation mode of the present application, the confusion matrix is ​​used to show the correspondence between the model prediction results and the actual labels, including the loss function of the model, the overall accuracy of the verification sample, the Kappa of the verification sample, the producer accuracy of the verification sample, and the user accuracy of the verification sample.

[0160] Step Se, performing a second evaluation on the farmland identification result based on the mixed matrix to obtain second evaluation information;

[0161] Step Sf: evaluating the farmland identification model based on the first evaluation information and the second evaluation information.

[0162] By obtaining the crop planting area information of the study area in the set time period and comparing it with the farmland identification results, the accuracy of the identification results can be evaluated based on the actual data. This evaluation method based on actual data is more intuitive and reliable. By inputting the verification set into the preset farmland identification model, the verification sample classification results are obtained, and compared with the farmland identification results, the performance of the model on unknown data can be evaluated.

[0163] Figure 2 This is a structural block diagram of a farmland identification device 200 according to an embodiment of the present application.

[0164] like Figure 2 As shown, a farmland identification device 200 mainly includes:

[0165] An acquisition module 201 is used to acquire a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area;

[0166] The optical feature recognition module 202 is used to perform optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, and wet;

[0167] A radar feature recognition module 203 is used to perform radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index;

[0168] The identification module 204 is used to obtain the farmland identification result of the study area based on the current optical feature information, the current radar feature information and the preset farmland identification model, the farmland identification result includes highland barley farmland area and rapeseed farmland area, the preset farmland identification model is obtained by training samples of multiple surface land use areas of different types, the multiple surface land use areas include at least highland barley farmland area and rapeseed farmland area, and the samples are generated based on historical radar remote sensing image data sets, historical optical remote sensing image data sets and ground survey data of the study area in a set time period.

[0169] In one example, the module in any of the above devices can be one or more integrated circuits configured to implement the above methods, such as: one or more application specific integrated circuits (ASICs), or, one or more digital signal processors (DSPs), or, one or more field programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0170] For another example, when the modules in the device can be implemented in the form of a processing element scheduling program, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor that can call a program. For another example, these modules can be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0171] Figure 3 This is a structural block diagram of an electronic device 300 according to an embodiment of the present application.

[0172] like Figure 3 As shown, the electronic device 300 includes a processor 301 and a memory 302 , and may further include one or more of an information input / information output (I / O) interface 303 , a communication component 304 , and a communication bus 305 .

[0173] The electronic device 300 can be implemented by one or more application specific integrated circuits (ASIC), digital signal processors (DSP), digital signal processing devices (DSPD), programmable logic devices (PLD), field programmable gate arrays (FPGA), controllers, microcontrollers, microprocessors or other electronic components to execute the farmland identification method given in the above embodiment.

[0174] The computer-readable storage medium provided in the embodiment of the present application is introduced below. The computer-readable storage medium described below and the farmland identification method described above can be referenced to each other.

[0175] The present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned agricultural and pastoral land identification method are implemented.

[0176] The computer-readable storage medium may include: a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and other media that can store program codes.

[0177] The terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article, or apparatus.

[0178] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used. Those skilled in the art should understand that the scope of application involved in the present application is not limited to the technical solution formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the aforementioned application concept. For example, the above features are replaced with (but not limited to) technical features with similar functions applied in the present application.

Claims

1. A method for identifying agricultural and pastoral land, characterized in that: include: Acquire a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area; Performing optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes normalized yellowness index, soil adjusted vegetation index, visible light atmospheric impedance vegetation index, ratio vegetation index, green-red ratio index, carotenoid I, normalized rapeseed flowering index, plant senescence reflectance index and humidity index; Performing radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index; Based on the current optical feature information, the current radar feature information and a preset farmland identification model, an farmland identification result of the study area is obtained, the farmland identification result includes a highland barley farmland area and a rapeseed farmland area, the preset farmland identification model is obtained by training based on samples of different types of multiple surface land use areas, the multiple surface land use areas at least include a highland barley farmland area and a rapeseed farmland area, and the samples are generated based on a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period; The method for establishing the preset farmland identification model includes: Acquire a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period, wherein the ground survey data includes at least one crop planted in the study area and the longitude and latitude information corresponding to the crop, wherein the crop includes at least highland barley and rapeseed; Determine the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset; Based on the phenological characteristics and the ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to the various first land uses, wherein the sample sets include multiple radar image sub-datasets and multiple optical image sub-datasets; Based on a preset sample generation rule, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for a second time to obtain sample sets corresponding to the plurality of second land uses; The first land use and the second land use are both regarded as surface land use areas, and radar feature recognition is performed on a plurality of radar image sub-datasets corresponding to each surface land use area to obtain radar feature information corresponding to the study area; Performing optical feature recognition on a plurality of optical image sub-datasets corresponding to each surface land use area to obtain optical feature information corresponding to the study area; generating a feature set of the study area based on the optical feature information and the radar feature information; The feature set is divided into a training set and a validation set, and the training set is input into a pre-established classification model for training until a farmland identification model that meets set conditions is obtained.

2. A method for identifying agricultural and pastoral land according to claim 1, characterized in that: Determining the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset includes: Based on the historical optical remote sensing image data set, a time series of multiple target indices is calculated, wherein the multiple target indices are NDYI, SAVI, VARI, RVI, GR, CRI, NRFI, PSRI, wet, TGSI, BSI, NDBSI, NDWI, MNDWI, DEM, Slope and Aspect; Based on a preset harmonic analysis algorithm and the time series of each of the target indexes, a phenological characteristic curve corresponding to each of the target indexes is generated; The phenological characteristic curve of each target index is fitted to determine the growing period corresponding to each crop in the study area, and the growing period corresponding to each crop is used as phenological characteristic information.

3. A method for identifying agricultural and pastoral land according to claim 1, characterized in that: Based on the phenological characteristics and ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to multiple first land uses, including: Based on the phenological characteristics, determining the first radar image data and the first optical image data corresponding to each crop in the growth period; Based on the ground survey data, the first radar image data and the first optical image data corresponding to each crop in the growth period are clipped to obtain the second radar image data and the second optical image data corresponding to each crop; For each of the crops, the second radar image data and the second optical image data corresponding to the crop are divided according to a preset time period to obtain sample sets corresponding to the multiple first land uses.

4. A method for identifying agricultural and pastoral land according to claim 3, characterized in that: Based on the preset sample generation rules, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the second time to obtain sample sets corresponding to various second land uses, including: Obtain land cover information of the study area in different years; Carry out a patch analysis on the land use information of each year to determine the center point coordinates of each land use type patch corresponding to each year; Determine the target point based on the coordinates of the center point of each land use type map corresponding to each year; Based on the target point, the historical radar remote sensing image data set and the historical optical remote sensing image data set are screened for the second time to obtain sample sets corresponding to the various second land uses.

5. The method for identifying agricultural and pastoral land according to claim 1, characterized in that: The processing of the pre-established classification model includes: Extracting features from the image in the feature set by using a convolutional neural network to generate feature maps with different spatial resolutions and abstraction levels; The feature map is passed through three independent one-dimensional convolutional layers to generate a query matrix, a key matrix and a value matrix respectively, wherein the query matrix represents the role of each position or feature point in the feature map as a query in the attention calculation, and the query role is the reference feature in the study area; the key matrix represents the role of each position or feature point in the feature map as a key point in the attention calculation, and the key point is the target feature in the study area related to the reference feature; and the value matrix represents the associated information of each key point; Based on the generated query matrix, the key matrix and the value matrix, an attention weight distribution is obtained, where the attention weight distribution represents the correlation between different positions or feature points in the feature map; Determine the weight of the element corresponding to each feature map based on the attention weight distribution to obtain an attention result; The attention result and the original image are feature fused, and the size is adjusted through a cascade upsampler to output a feature image with the same full resolution as the original image, so as to classify the feature image through a classifier to obtain a classification result.

6. A method for identifying agricultural and pastoral land according to claim 1, characterized in that: After obtaining the farmland identification result of the study area based on the current optical feature information, the current radar feature information and the preset farmland identification model, the method further includes: Obtaining crop planting area information of the study area in a set time period; Performing a first evaluation on the farmland identification result based on the crop sowing area information to obtain first evaluation information; Inputting the verification set into a preset farmland identification model to obtain a verification sample classification result; Constructing a confusion matrix based on the verification sample classification results and the farmland identification results; Performing a second evaluation on the farmland identification result based on the confusion matrix to obtain second evaluation information; The farmland identification model is evaluated based on the first evaluation information and the second evaluation information.

7. A device for identifying agricultural and pastoral land, characterized in that: include: An acquisition module, used to acquire a current radar remote sensing image dataset and a current optical remote sensing image dataset corresponding to a study area, wherein at least highland barley and rapeseed are planted in the study area; An optical feature recognition module is used to perform optical feature recognition on the current optical remote sensing image data set to obtain current optical feature information of the study area, wherein the current optical feature information includes vegetation index time series information, soil index time series information, water index time series information and terrain index time series information, and the vegetation index includes normalized yellowness index, soil adjusted vegetation index, visible light atmospheric impedance vegetation index, ratio vegetation index, green-red ratio index, carotenoid I, normalized rapeseed flowering index, plant senescence reflectance index and humidity index; A radar feature recognition module is used to perform radar feature recognition on the current radar remote sensing image data set to obtain current radar feature information of the study area, wherein the current radar feature information includes a surface roughness index and a soil moisture index; an identification module, for obtaining an identification result of farmland in the study area based on the current optical feature information, the current radar feature information and a preset farmland identification model, wherein the identification result of farmland includes a highland barley farmland area and a rapeseed farmland area, the preset farmland identification model is obtained by training samples of a variety of surface land use areas of different types, the variety of surface land use areas at least include a highland barley farmland area and a rapeseed farmland area, and the samples are generated based on a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period; The method for establishing the preset farmland identification model includes: Acquire a historical radar remote sensing image dataset, a historical optical remote sensing image dataset and ground survey data of the study area in a set time period, wherein the ground survey data includes at least one crop planted in the study area and the longitude and latitude information corresponding to the crop, wherein the crop includes at least highland barley and rapeseed; Determine the phenological characteristics of crops in the study area based on the historical optical remote sensing image dataset; Based on the phenological characteristics and the ground survey data, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for the first time to obtain sample sets corresponding to the various first land uses, wherein the sample sets include multiple radar image sub-datasets and multiple optical image sub-datasets; Based on a preset sample generation rule, the historical radar remote sensing image dataset and the historical optical remote sensing image dataset are screened for a second time to obtain sample sets corresponding to the plurality of second land uses; The first land use and the second land use are both regarded as surface land use areas, and radar feature recognition is performed on a plurality of radar image sub-datasets corresponding to each surface land use area to obtain radar feature information corresponding to the study area; Performing optical feature recognition on a plurality of optical image sub-datasets corresponding to each surface land use area to obtain optical feature information corresponding to the study area; generating a feature set of the study area based on the optical feature information and the radar feature information; The feature set is divided into a training set and a validation set, and the training set is input into a pre-established classification model for training until a farmland identification model that meets set conditions is obtained.

8. An electronic device, characterized in that: comprising a processor coupled to a memory; The processor is configured to execute a computer program stored in the memory, so that the electronic device executes the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: The method comprises a computer program or an instruction, which, when executed on a computer, causes the computer to execute the method according to any one of claims 1 to 6.

Images