Grassland category interpretation model training method and device, grassland category interpretation method and device, electronic equipment and storage medium

Through the fusion processing and labeling of digital orthophotographs and digital elevation models, the grassland category interpretation model is trained, and the objectivity and accuracy of grassland category interpretation in the existing technology is solved, and automatic interpretation and efficient identification are realized .

CN119048797APending Publication Date: 2024-11-29Xinjiang Uygur Autonomous Region Natural Resources Planning Research Institute +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410999677.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-24
Publication Date
2024-11-29

AI Technical Summary

Technical Problem

In the identification and interpretation of natural grassland categories, the existing technology has problems such as cumbersome operational processes, high professionalism, low standards consistency and strong subjectivity of results, and it is difficult to ensure the objectivity, efficiency and accuracy of grassland category interpretation.

Method used

By fusing the digital orthophotograph and the digital elevation model, the non-target-related land-type map pattern is removed, and the land-type tag of the target land-type map pattern is determined and marked. Based on this, the initial grassland category interpretation model is trained and trained. Good grassland category interpretation model for automatic interpretation of grassland categories.

Benefits of technology

Improve the accuracy and reliability of grassland category interpretation, realize automatic interpretation of grassland category, avoid the problems of low efficiency and inaccurate results caused by manual interpretation and field practice, and ensure the objective of grassland category interpretation efficiency, efficiency and accuracy.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119048797A_ABST
    Figure CN119048797A_ABST
Patent Text Reader

Abstract

The invention discloses a grassland category interpretation model training method, a grassland category interpretation method, a grassland category interpretation device, electronic equipment and a storage medium, and relates to the technical fields of remote sensing image processing, deep learning and the like, and the method comprises the steps: carrying out the fusion processing of a digital orthoimage and a digital elevation model of a target region, and obtaining a first fusion image; performing mask processing on the first fusion image based on a preset target land class pattern spot set to remove non-target related land class pattern spots in the first fusion image to obtain a second fusion image; land class labels of all the first target land class pattern spots in the second fusion image are determined, label labeling is carried out, and a third fusion image is obtained; and training the initial grassland category interpretation model based on the third fusion image to obtain a trained grassland category interpretation model, the grassland category interpretation model being used for interpreting a grassland category. According to the method, objectivity, high efficiency and accuracy of grassland category interpretation can be guaranteed.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to technical fields such as remote sensing image processing and deep learning, and in particular to a grassland category interpretation model training method, a grassland category interpretation method, a device, an electronic device and a storage medium. Background Art

[0002] With the rapid development of remote sensing technology, accurate identification and interpretation of natural grassland categories have become important tasks in the fields of ecological monitoring, grassland resource management and environmental protection. Grassland category distribution information can provide basic data support for natural grassland survey and monitoring, digital protection and precise governance.

[0003] At present, the extraction and interpretation of natural grassland categories and distribution are mainly achieved through manual visual interpretation and field surveys. This has the problems of complicated operating procedures, high professional requirements, low standard consistency, and highly subjective results. It is difficult to ensure the objectivity, efficiency and accuracy of grassland category interpretation. Summary of the invention

[0004] The main purpose of the present invention is to provide a grassland category interpretation model training method, a grassland category interpretation method, a device, an electronic device and a storage medium, aiming to ensure the objectivity, efficiency and accuracy of grassland category interpretation.

[0005] To achieve the above object, the present invention proposes a grassland category interpretation model training method, the method comprising:

[0006] Fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image;

[0007] The first fused image is masked based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to the grassland category;

[0008] Determine the land class label of each first target land class patch in the second fused image, and perform label marking to obtain a third fused image;

[0009] The initial grassland category interpretation model is trained based on the third fused image to obtain a trained grassland category interpretation model, and the grassland category interpretation model is used to interpret grassland categories.

[0010] In one embodiment, the step of fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image includes:

[0011] Resampling the digital orthophoto and digital elevation model of the target area respectively to obtain an intermediate digital orthophoto and an intermediate digital elevation model with the same feature dimensions;

[0012] The intermediate digital orthophoto and the intermediate digital elevation model are fused to obtain a first fused image.

[0013] In one embodiment, before the step of performing mask processing on the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image and obtaining the second fused image, the step further includes:

[0014] Query the land type codes related to grassland categories from the preset survey data;

[0015] Obtain the target land class patch set corresponding to the land class code from the land class patches in the survey data.

[0016] In one embodiment, before determining the land class labels of each first target land class patch in the second fused image and labeling the labels, the step of obtaining the third fused image further includes:

[0017] An interpretation sample library is constructed, in which vector data corresponding to the second fused image is stored, the vector data including each second target land class patch, a land class label of each second target land class patch, and geographic location information of each second target land class patch.

[0018] In one embodiment, the steps of determining the land class labels of each first target land class patch in the second fused image and labeling the labels to obtain the third fused image include:

[0019] Determining the geographical location information of each first target land type patch in the second fused image;

[0020] Determine the correspondence between each first target land class spot and each second target land class spot according to the geographical location information of each first target land class spot and the geographical location information of each second target land class spot;

[0021] Based on the corresponding relationship, each first target land class spot in the second fused image is annotated with a land class label according to the land class label of each second target land class spot to obtain a third fused image.

[0022] In one embodiment, the step of training the initial grassland category interpretation model based on the third fused image to obtain the trained grassland category interpretation model includes:

[0023] The third fused image is cropped to obtain sample data, wherein the sample data includes a plurality of first cropped images and a land class label of a first target land class patch in each first cropped image;

[0024] Inputting the plurality of first cropped images into the initial grassland category interpretation model to obtain training interpretation results corresponding to the plurality of first cropped images;

[0025] Determine the loss value according to the training interpretation results corresponding to each of the plurality of first cropped images and the land class label of the first target land class patch in each of the first cropped images;

[0026] The parameters of the initial grassland category interpretation model are updated according to the loss value to obtain a trained grassland category interpretation model.

[0027] In addition, to achieve the above purpose, the present invention also proposes a grassland category interpretation method, which includes:

[0028] fusing the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image;

[0029] Performing cropping processing on the fourth fused image to obtain second cropped images;

[0030] Inputting each second cropped image into a preset grassland category interpretation model to obtain respective interpretation results of each second cropped image, wherein the grassland category interpretation model is trained according to the grassland category interpretation model training method described in the above embodiment;

[0031] The interpretation results are spliced ​​together to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0032] In addition, to achieve the above-mentioned purpose, the present invention also proposes a grassland category interpretation model training device, which includes:

[0033] A first fusion processing module is used to fuse the digital orthophoto and the digital elevation model of the target area to obtain a first fused image;

[0034] A mask processing module is used to perform mask processing on the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to the grassland category;

[0035] A label marking module is used to determine the land class label of each first target land class patch in the second fused image, and perform label marking to obtain a third fused image;

[0036] The model training module is used to train the initial grassland category interpretation model based on the third fused image to obtain a trained grassland category interpretation model, and the grassland category interpretation model is used to interpret grassland categories.

[0037] In addition, to achieve the above-mentioned purpose, the present invention also proposes a grassland category interpretation device, which comprises:

[0038] The second fusion processing module is used to fuse the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image;

[0039] A cropping module, used for cropping the fourth fused image to obtain second cropped images;

[0040] An interpretation module, used for inputting each second cropped image into a preset grassland category interpretation model to obtain an interpretation result of each second cropped image, wherein the grassland category interpretation model is trained according to the grassland category interpretation model training device described in the above embodiment;

[0041] The splicing module is used to splice the interpretation results to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0042] In addition, to achieve the above-mentioned purpose, the present invention also proposes an electronic device, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the grassland category interpretation model training method and / or the grassland category interpretation method as described above.

[0043] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, which is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, the steps of the grassland category interpretation model training method and / or the grassland category interpretation method as described above are implemented.

[0044] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer program product, which includes a computer program, and when the computer program is executed by a processor, it implements the steps of the grassland category interpretation model training method and / or grassland category interpretation method as described above.

[0045] The present invention proposes a grassland category interpretation model training method, which obtains a first fused image containing spectral information and terrain characteristics of the target area by fusing a digital orthophoto and a digital elevation model of a target area; then, the first fused image is masked to remove land class patches irrelevant to the grassland category in the first fused image, thereby obtaining a second fused image; on this basis, the land class labels of each first target land class patch in the second fused image are determined and labeled, thereby obtaining a third fused image; finally, the initial grassland category interpretation model is trained based on the third fused image, thereby obtaining a trained grassland category interpretation model for interpreting grassland categories.

[0046] In this way, through the fusion processing of digital orthophotos and digital elevation models, rich spectral features and terrain characteristics are provided for interpretation. By introducing spectral information and terrain factors into the training process of grassland category interpretation model, the model can fully learn the spectral characteristics and terrain characteristics of various types of grasslands during training, thereby improving the training precision and accuracy of the model.

[0047] The present invention also proposes a grassland category interpretation method, which applies the grassland category interpretation model trained by the above-mentioned grassland category interpretation model training method. First, the digital orthophoto and the digital elevation model of the area to be interpreted are fused to obtain a fourth fused image. Then, the fourth fused image is cropped to obtain second cropped images. Then, each second cropped image is input into the grassland category interpretation model to obtain the interpretation result of each second cropped image. Finally, each interpretation result is spliced ​​to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0048] In this way, by using the trained grassland category interpretation model to interpret the fourth fused image, the automatic interpretation of natural grassland categories can be achieved, and the accuracy and reliability of the interpretation results can be improved, thereby avoiding the problems of low interpretation efficiency and inaccurate interpretation results caused by manual visual interpretation and field adjustments, and ensuring the objectivity, efficiency and accuracy of grassland category interpretation. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0050] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0051] Figure 1 A schematic diagram of a process flow provided for Embodiment 1 of the grassland category interpretation model training method of the present invention;

[0052] Figure 2 A schematic diagram of the fusion processing process provided in Example 2 of the grassland category interpretation model training method of the present invention;

[0053] Figure 3 A schematic diagram of vector data provided for the second embodiment of the grassland category interpretation model training method of the present invention;

[0054] Figure 4A schematic diagram of geographical location information of a second target land type patch provided in Example 2 of the grassland classification interpretation model training method of the present invention;

[0055] Figure 5 A schematic diagram of geographical location information of a first target land type patch provided in Example 2 of the grassland type interpretation model training method of the present invention;

[0056] Figure 6 A schematic diagram of the network structure of an initial grassland category interpretation model provided in Example 2 of the grassland category interpretation model training method of the present invention;

[0057] Figure 7 A schematic diagram of the effect of the training interpretation results provided in Example 2 of the grassland category interpretation model training method of the present invention;

[0058] Figure 8 A schematic diagram of the implementation process of the second embodiment of the grassland category interpretation model training method of the present invention;

[0059] Fig. 9 This is a schematic diagram of the module structure of a grassland category interpretation model training device according to an embodiment of the present invention;

[0060] Fig.10 It is a schematic diagram of the device structure of the hardware operating environment involved in the grassland category interpretation model training method and the grassland category interpretation method in the embodiment of the present invention.

[0061] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings in conjunction with the embodiments. DETAILED DESCRIPTION

[0062] It should be understood that the specific embodiments described herein are only used to explain the technical solutions of the present invention and are not used to limit the present invention.

[0063] In order to better understand the technical solution of the present invention, a detailed description will be given below in conjunction with the accompanying drawings and specific implementation methods.

[0064] The embodiment of the present invention provides a grassland category interpretation model training method. Figure 1 4 is a flow chart of the first embodiment of the grassland classification interpretation model training method of the present invention.

[0065] Reference Figure 1 In this embodiment, the grassland category interpretation model training method includes steps S10 to S40:

[0066] Step S10, fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image.

[0067] It should be noted that Digital Orthophoto Map (DOM) is based on digitized aerospace photographs or remote sensing images (monochrome / color), which are scanned, processed, and subjected to radiometric correction, differential correction, and mosaicking pixel by pixel. The image data is clipped according to the range of the topographic map and carries a plane map of topographic element information such as symbols, line drawings, annotations, kilometer grids, and outline (inside and outside) decoration. DOM can provide rich surface information, which is conducive to extracting a variety of grassland and non-grassland feature samples from DOM for training interpretation models.

[0068] In addition, it should be noted that the Digital Elevation Model (DEM) is a digital simulation of the ground terrain (i.e., a digital expression of the terrain surface morphology) achieved through limited terrain elevation data. It is a physical ground model that represents the undulating morphology of the ground elevation in the form of a set of ordered numerical arrays. DEM data can provide elevation information of the terrain, which helps to identify and distinguish different terrain features in grassland classification, such as slopes, valleys, hills, etc. These features are very important for accurate identification and classification of grassland categories. For example, some types of grassland may only be distributed in areas with specific altitudes or slopes, while other types of grassland may be more common in other terrain features.

[0069] In this embodiment, for the grassland classification task, considering that there is an obvious vertical zoning relationship between different grassland categories and altitudes, this embodiment fuses the digital orthophoto and digital elevation model of the target area to obtain a first fused image, and uses the first fused image as a base map to produce sample data for training the grassland category interpretation model, which is conducive to improving the pertinence and reliability of the sample data, thereby facilitating improving the classification accuracy and classification precision of the model. In this embodiment of the present invention, the target area is any area with a large distribution of grasslands.

[0070] Step S20, masking the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to grassland categories.

[0071] The target land class patch set is a set of land class patches related to the grassland category, and the non-target related land class patches are land class patches in the first fused image that are not related to the grassland category. By using the preset target land class patch set to mask the first fused image, pixels that are not related to the grassland category in the first fused image can be eliminated, and only pixels related to the grassland category can be retained, thereby obtaining a masked image, that is, a second fused image, which is image data stored in a raster format. This is conducive to quickly and effectively obtaining sample data for training grassland category interpretation models.

[0072] In a feasible embodiment, before step S20, steps A10 to A20 may be further included:

[0073] Step A10, querying the land type code related to the grassland category from the preset survey data.

[0074] It should be noted that the survey data is, for example, the basic survey data of the third national land survey, which includes land type distribution information of multiple regions, including land type name, land type category, land type code, land type distribution, etc. All land type codes related to grassland categories can be queried from the survey data of the third national land survey, and the land type codes correspond to the land types one by one. Of course, in actual application scenarios, the survey data can also be other data or data platforms that can query the land type distribution information of each region, which is not specifically limited in this embodiment.

[0075] The land type codes related to the grassland categories can be queried from the survey data. The land type codes correspond to the land types one by one. The land type codes related to the grassland categories can include codes of multiple grassland categories such as forest land, grassland and other lands.

[0076] Step A20, obtaining a target land type patch set corresponding to the land type code from the land type patches in the survey data.

[0077] It should be noted that the land class spots in the survey data include land class spot information corresponding to multiple regions, such as land class codes, land class names and corresponding land class spots. Based on the determined land class codes related to grassland categories, target land class spots related to these land class codes are queried from the land class spots, such as land class spots corresponding to multiple grassland categories such as woodland, grassland and other land, and these land class spots are integrated into a target land class spot set.

[0078] Step S30, determining the land class label of each first target land class patch in the second fused image, and marking the label to obtain a third fused image.

[0079] For each identified first target land type patch in the second fused image, determine its corresponding land type label, which is a specific description of the land cover type to which the patch belongs, such as temperate desert type, temperate steppe desert type, alpine desert type, lowland meadow type, mountain meadow type, alpine meadow type, etc.

[0080] After the land class label is determined, the label information is annotated on the corresponding land class patch, and the accurate correspondence between the label and the patch is ensured, so as to obtain a third fused image containing the land class label annotation.

[0081] Step S40: training the initial grassland category interpretation model based on the third fused image to obtain a trained grassland category interpretation model, where the grassland category interpretation model is used to interpret grassland categories.

[0082] Based on the third fused image, sample data for training an initial grassland category interpretation model is generated, and the initial grassland category interpretation model is trained using the sample data to obtain a trained grassland category interpretation model, which is used to interpret grassland categories. For example, the grassland category interpretation model is used to obtain the grassland category of each land type patch in the image based on remote sensing (multispectral) image interpretation.

[0083] It should be noted that the initial grassland category interpretation model can be, for example, a convolutional neural network model (CNN) or a residual neural network model. Of course, based on different application requirements, other types of models can also be used to train grassland category interpretation models, such as a Transformer model using an attention mechanism, etc. This embodiment does not make specific limitations on this.

[0084] The embodiment of the present invention obtains a first fused image containing spectral information and terrain characteristics of the target area by fusing a digital orthophoto and a digital elevation model of the target area; then, by performing mask processing on the first fused image, land class patches irrelevant to grassland categories in the first fused image are removed to obtain a second fused image; on this basis, the land class labels of each first target land class patch in the second fused image are determined and labeled to obtain a third fused image; finally, the initial grassland category interpretation model is trained based on the third fused image to obtain a trained grassland category interpretation model for interpreting grassland categories.

[0085] In this way, through the fusion processing of digital orthophotos and digital elevation models, rich spectral characteristics and terrain properties are provided for interpretation. By introducing spectral information and terrain factors into the training process of grassland category interpretation model, the model can fully learn the spectral characteristics and terrain characteristics of various types of grasslands during training, thereby improving the training precision and accuracy of the model.

[0086] Based on the first embodiment of the present invention, in the second embodiment of the present invention, the same or similar contents as those in the above-mentioned embodiment 1 can be referred to the above introduction, and will not be described in detail later. On this basis, step S10 can include steps S101 to S102:

[0087] Step S101 , resampling the digital orthophoto and digital elevation model of the target area respectively to obtain an intermediate digital orthophoto and an intermediate digital elevation model with the same feature dimensions.

[0088] Step S102: fusing the intermediate digital orthophoto and the intermediate digital elevation model to obtain a first fused image.

[0089] The fusion processing of the DOM and DEM data of the target area specifically refers to first resampling the DOM and the corresponding DEM data until the two have the same characteristic dimension, and then fusing the resampled intermediate digital orthophoto with the intermediate digital elevation model to obtain a first fused image.

[0090] For example, in a feasible embodiment, the specific fusion process is as follows: Figure 2 As shown, using tensor R c1 ×h1×w1 The features of the digital orthophoto DOM of the target area, where c1, h1, and w1 represent the number of channels, height, and width of the DOM, respectively. c2×h2×w2 Indicates the characteristics of the DEM data corresponding to the target area, where c2, h2, and w2 represent the number of channels, height, and width of the DEM data, respectively. For example, c1 is a 4-channel number, representing the red band R, green band G, blue band B, and near-infrared band NIR, respectively. For example, c2 is a single channel number, representing the elevation value of the terrain within the target area.

[0091] First, the DEM data is resampled to convert R c2×h2×w2 Convert to R c2×h1×w1 After that, R c2×h1×w1 and R c1×h1×w1 Perform fusion processing to obtain the first fused image R c3×h1×w1 , where c3=c1+c2.

[0092] In a feasible embodiment, step S30 may further include step B10:

[0093] Step B10, constructing an interpretation sample library, in which vector data corresponding to the second fused image is stored, the vector data including each second target land class patch, the land class label of each second target land class patch and the geographical location information of each second target land class patch.

[0094] The interpretation sample library stores vector data corresponding to the masked fused image (i.e., the second fused image). The area corresponding to the vector data is consistent with the area corresponding to the second fused image, i.e., the vector data is vector data for the target area, and the second fused image is raster data for the target area.

[0095] The vector data contains multiple target land class spots represented in vector form. The target land class spots are obtained by manually outlining the DOM corresponding to the target area and marking the land class category in advance. Each target land class spot has a corresponding land class label. The land class label is used to indicate the land class category of the corresponding target land class spot.

[0096] Vector data can be divided into three basic vector types: point, line (or arc) and polygon data. Based on the basic types, composite objects can also be formed, such as multiple points, multiple lines or multiple surfaces and composite objects composed of points, lines and surfaces.

[0097] like Figure 3 As shown, the vector data includes a plurality of second target land type spots represented by surfaces, such as second target land type spots 11 and second target land type spots 12. Each second target land type spot has corresponding geographical location information. Taking the second target land type spot 11 as an example, Figure 4 As shown, starting from the leftmost side of the second target land class spot 11, the coordinates of the vertices of the second target land class spot 11 are (x1, y1), (x2, y2), ..., (x8, y8). Based on the coordinates of the vertices of the second target land class spot 11, the geographic location information of the second target land class spot 11 in the vector data can be determined. Similarly, for each second target land class spot in the vector data, the geographic location information of the second target land class spot can also be determined. In addition, each second target land class spot has a corresponding land class label.

[0098] In a feasible embodiment, step S30 may include steps S301 to S303:

[0099] Step S301, determining the geographical location information of each first target land type patch in the second fused image;

[0100] The masked fused image (i.e., the second fused image) is composed of a matrix of pixels (or pixels) organized in rows and columns (or grids). In the second fused image, the geographic location information of each pixel can be determined based on the row and column of each pixel in the grid matrix. Then, based on the geographic location information of each pixel, the geographic location information of each first target land class spot can be determined.

[0101] For example, each raster data generally has corresponding geographic attributes, including coordinate system, reference coordinate position, pixel size, row count, and column count. The geographic location information of any pixel can be determined using the above geographic attributes.

[0102] like Figure 5 As shown in FIG. 1 , the pixel located at the upper left corner of the second fused image is used as the reference position, and its coordinates are (x0, y0). For any pixel (x s ,y s )(like Figure 5 The position of the asterisk shown in the figure can be used to determine the number of rows and columns (rows in the x direction and columns in the y direction) from the reference position. Then, based on the number of rows and columns between the two, the geographic location information of the reference position, and the actual width and height represented by each pixel, the geographic location information of the pixel can be determined.

[0103] Step S302: determining the correspondence between each first target land class spot and each second target land class spot according to the geographical location information of each first target land class spot and the geographical location information of each second target land class spot.

[0104] Since the vector data and the second fused image are respectively vector representations and raster representations of the same area, the correspondence between each first target land class spot and each second target land class spot can be determined based on the geographic location information of each second target land class spot in the vector data and the geographic location information of each first target land class spot in the second fused image. If the geographic location information of a first target land class spot is the same as the geographic location information of a second target land class spot, the first target land class spot and the second target land class spot are determined to be the same target land class spot, otherwise, the first target land class spot and the second target land class spot are determined to be two different target land class spots.

[0105] Step S303 , based on the corresponding relationship, each first target land class patch in the second fused image is labeled with a land class label according to the land class label of each second target land class patch, so as to obtain a third fused image.

[0106] The land class labels of the first target land class spots in the second fused image are determined according to the correspondence between the first target land class spots and the second target land class spots. Then, the determined labels are correspondingly marked on the first land class spots in the second fused image to obtain a third fused image carrying the land class labels.

[0107] In a feasible embodiment, step S40 may include steps S401 to S404:

[0108] Step S401 , cropping the third fused image to obtain sample data, wherein the sample data includes a plurality of first cropped images and a land class label of a first target land class patch in each first cropped image.

[0109] The masked fused image carrying the land class label (i.e., the third fused image) is cropped to obtain sample data for training the grassland category interpretation model. The sample data includes multiple first cropped images and the land class labels of the first target land class patches in each first cropped image. The size of the first cropped image can be set according to actual needs and is not specifically limited in this embodiment.

[0110] For example, in a feasible embodiment, in the process of cropping the third fused image carrying the land class label and 5 bands (red band R, green band G, blue band B, near infrared band NIR and elevation data), a sliding window strategy can be used to crop the sample, and the cropping window is, for example, 64*64 pixels in size, and the sliding step is 32 pixels in size. After the image cropping is completed, the obtained multiple first cropped images are used as sample data, each of which includes at least one first target land class spot, and the land class label corresponding to the at least one first target land class spot is used as the real land class label of the first cropped image.

[0111] In addition, in order to further enrich the sample data volume, data enhancement processing can be performed on each first cropped image. The data enhancement methods include, for example, geometric space transformation (such as rotation, horizontal flipping, diagonal flipping, left-right flipping, scaling, etc.), color transformation (such as linear stretching, Gaussian noise, blurring, etc.), scale transformation, etc.

[0112] Step S402: input the plurality of first cropped images into an initial grassland category interpretation model to obtain training interpretation results corresponding to the plurality of first cropped images.

[0113] The initial grassland category interpretation model is, for example, a convolutional neural network model, whose network structure is as follows: Figure 6 As shown, the model includes 1 3*3 convolutional layer, 1 batch normalization BN (Batch Normalization) layer, 1 Relu activation layer, the first residual block, the second residual block, 1 1*1 convolutional layer, 1 Relu activation layer and 1 1*1 convolutional layer. Figure 6 The 1*1 convolution layer and 1 Relu activation layer corresponding to the asterisk-marked part are optional network layer structures. In some embodiments, the network layer corresponding to the asterisk-marked part can also be deleted.

[0114] The first residual block and the second residual block include 1 3*3 convolution layer, 1 batch normalization BN layer, 1 Relu activation layer, 1 3*3 convolution layer, 1 batch normalization BN layer, and 1 Relu activation layer according to the data flow direction.

[0115] In this embodiment, the batch normalization BN layer is a normalization method used in neural networks to accelerate the training process and improve the performance of the model. The input of the BN layer is a four-dimensional tensor, usually expressed as (number of samples, number of channels, height, width). The output of the BN layer is also a four-dimensional tensor with the same shape as the input. The BN layer suppresses the internal covariate shift problem in the training process by normalizing each channel of the input tensor, thereby accelerating the training of the model.

[0116] The convolution layer is used to extract features from the image. The ReLU activation function is a commonly used nonlinear activation function that can enhance the nonlinear expression ability of the model. The input and output shapes of the ReLU activation layer are the same as those of the convolution layer and the BN layer.

[0117] In the process of interpreting model training, the input sample data (including multiple first cropped images) are convolved using a 3*3 convolutional layer, and the output result after convolution is processed by a batch normalization BN layer and a Relu activation layer. After that, the output result is input into the first residual block and the second residual block for processing. Finally, after being processed by a 1*1 convolutional layer, the training interpretation result corresponding to the sample data (including the training interpretation results corresponding to each of the multiple first cropped images) is obtained. The training interpretation result corresponding to each first cropped image is used to indicate the land category prediction of each pixel in the first cropped image.

[0118] The effect of the training interpretation result corresponding to a first cropped image is as follows Figure 7 As shown, in Figure 7 In the figure, each small grid represents a pixel, and the number marked in each pixel (such as F, W, U in the figure) represents the land class corresponding to the pixel. If two pixels have the same number, it means that the two pixels belong to the same land class. For example, the pixel numbered U belongs to the same land class, and the pixel numbered F belongs to another land class. Based on the training interpretation results corresponding to each first cropped image, the category prediction of each first target land class patch in the first cropped image can be determined.

[0119] Step S403 , determining a loss value according to the training interpretation results corresponding to each of the plurality of first cropped images and the land class label of the first target land class patch in each of the first cropped images.

[0120] Step S404, updating the parameters of the initial grassland category interpretation model according to the loss value to obtain a trained grassland category interpretation model.

[0121] The loss value is calculated using the training interpretation results corresponding to each of the multiple first cropped images and the land class labels of the first target land class patches in each first cropped image, and the loss value is used to adjust and update the parameters of the model until the model converges or reaches a preset number of training iterations. Thus, the training of the initial grassland category interpretation model is completed, and a trained grassland category interpretation model is obtained.

[0122] The loss functions used in the model training process include, for example, the cross entropy loss function and the Dice loss function. Among them, the Dice loss function is used to reduce the interference caused by sample imbalance. In some embodiments, only the cross entropy loss function can be used to calculate the loss generated in the model training process.

[0123] In addition, in a feasible embodiment, in actual application, the trained grassland category interpretation model can be verified. The evaluation indicators of model verification use overall accuracy (OA), precision (P), recall (R), F1 score (F), etc. The specific calculation method is as follows:

[0124]

[0125]

[0126] In the above formula, TP, TN, FP and FN respectively represent the number of positive samples correctly identified, the number of negative samples correctly identified, the number of negative samples identified as positive samples, and the number of positive samples identified as negative samples, which can be calculated by the confusion matrix.

[0127] During the model verification process, the above-mentioned evaluation indicators are used to verify the training effect of the interpretation model. In an example, if OA, P, F and R reach the preset evaluation thresholds, such as OA>90%, P>90%, R>90%, F>0.90, respectively, it is considered that the accuracy of the interpretation model meets the expected requirements. Of course, the evaluation thresholds of the above-mentioned evaluation indicators can also be set according to actual conditions, and this embodiment does not make specific limitations on this. The grassland category interpretation model after training and verification can be used to interpret remote sensing images, thereby achieving efficient and accurate interpretation of different grassland categories in remote sensing images.

[0128] Based on the first embodiment of the present invention, in the second embodiment of the present invention, the same or similar contents as those of the above-mentioned embodiment 1 can be referred to the above introduction and will not be described in detail later.

[0129] The embodiment of the present invention proposes a grassland category interpretation method, which includes steps C10 to C40:

[0130] Step C10, fusing the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image.

[0131] After the trained grassland category interpretation model is obtained, the digital orthophoto and digital elevation model data of the area to be interpreted are fused to obtain a fourth fused image.

[0132] Step C20, cropping the fourth fused image to obtain second cropped images.

[0133] The fourth fused image is cropped to obtain second cropped images, and the size of the second cropped images is determined based on the processable image size of the model. For example, in this embodiment, the fourth fused image can be cropped with a size of 512*512 pixels and a step length of 400 to obtain second cropped images.

[0134] Step C30 , inputting each second cropped image into a grassland classification interpretation model to obtain respective interpretation results of each second cropped image.

[0135] The cropped second cropped images are input into the grassland category interpretation model, and the second cropped images are classified using the trained grassland category interpretation model to obtain respective interpretation results of the second cropped images. The grassland category interpretation model is trained according to the grassland category interpretation model training method described in the above embodiment.

[0136] Step C40, splicing the interpretation results to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0137] The interpretation results of each second cropped image are spliced ​​together to obtain a natural grassland category distribution map corresponding to the area to be interpreted.

[0138] In addition, after obtaining the grassland category distribution map corresponding to the remote sensing image to be interpreted, it can also be post-processed, and the post-processing includes at least one of the following operations:

[0139] (1) Dilation and / or erosion are performed on each target land type in the grassland distribution map to further correct the shape of each target land type.

[0140] (2) Remove small patches in the grassland type distribution map. For example, remove small patches in the grassland type distribution map whose connected domain area is lower than a preset threshold.

[0141] (3) The grassland distribution map is vectorized and polygons are simplified using the Douglas-Peucker algorithm to obtain a grassland vector distribution map. Thus, each grassland type can be represented in a vectorized form.

[0142] In summary, in the embodiment of the present invention, by introducing prior data (i.e., survey data), spectral information, and terrain factors into the training process of the interpretation model during the training stage of the grassland category interpretation model, the model can combine prior knowledge to eliminate interference items during the training process, and fully learn the spectral characteristics and terrain characteristics of various types of grasslands, thereby improving the training precision and accuracy of the model. In the grassland category interpretation stage, the interpretation model obtained by training is used to interpret the fourth fused image, which can realize the automatic interpretation of natural grassland categories and improve the accuracy and reliability of the interpretation results, thereby avoiding the problems of low interpretation efficiency and inaccurate interpretation results caused by visual interpretation, and ensuring the objectivity, efficiency, and accuracy of grassland category interpretation.

[0143] At the same time, through continuous data accumulation and model optimization in the application, it is beneficial to significantly reduce the manpower and material costs required for grassland field surveys and internal data processing in the future, and realize economical and efficient interpretation and extraction of natural grassland categories.

[0144] For example, in order to help understand the implementation process of the grassland category interpretation model training method and grassland category interpretation method obtained by combining this embodiment with the above-mentioned embodiment 1, please refer to Figure 8 , Figure 8 A brief process diagram is provided, specifically:

[0145] The process implementation includes three stages: sample preparation, model training, and image interpretation and post-processing. The sample preparation process includes the following steps 1 to 4:

[0146] Step 1: The digital orthophoto DOM and the digital elevation model DEM corresponding to the target area are fused by band superposition to obtain a fused image.

[0147] Step 2, based on the survey data query, obtain the target land class code related to the grassland category, query from the land class patch DLTB to obtain the land class patch set related to the target land class code (i.e., the target land class patches in the figure), and mask the fused image based on the land class patch set related to the target land class code to remove the land class patches in the fused image that are not related to the natural grassland category, and obtain the masked fused image, i.e., the second fused image mentioned above, which is raster data.

[0148] Step 3, using the vector data stored in the pre-constructed interpretation sample library to mark the labels corresponding to each target land type patch in the second fused image, to obtain a masked fused image carrying the labels (ie, the third fused image).

[0149] Step 4: crop the masked fused image carrying the label (ie, the third fused image) to obtain sample data for training the interpretation model.

[0150] The model training process is to train the interpretation model based on the obtained sample data to obtain the trained interpretation model, that is, the grassland category interpretation model.

[0151] The image interpretation and post-processing process includes two parts: image interpretation and post-interpretation processing. The image interpretation part is to interpret the fourth fused image of the area to be interpreted using the grassland category interpretation model to obtain a natural grassland category distribution map corresponding to the area to be interpreted. The post-interpretation processing part is used to process the natural grassland category distribution map according to actual application requirements and application scenarios. It should be noted that the image interpretation and post-processing process is the same or similar to the above embodiment, and can refer to the above introduction, which will not be repeated here.

[0152] It should be noted that the above examples are only used to understand the present invention and do not constitute a limitation on the grassland category interpretation model training method and grassland category interpretation method of the present invention. More forms of simple transformations based on this technical concept are all within the protection scope of the present invention.

[0153] The present invention also provides a grassland category interpretation model training device, please refer to Fig. 9 , the grassland category interpretation model training device includes:

[0154] The first fusion processing module 10 is used to fuse the digital orthophoto and the digital elevation model of the target area to obtain a first fused image;

[0155] The mask processing module 20 is used to perform mask processing on the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to the grassland category;

[0156] The label marking module 30 is used to determine the land class label of each first target land class patch in the second fused image, and mark the label to obtain a third fused image;

[0157] The model training module 40 is used to train the initial grassland category interpretation model based on the third fused image to obtain a trained grassland category interpretation model, and the grassland category interpretation model is used to interpret grassland categories.

[0158] Optionally, the first fusion processing module 10 is further configured to:

[0159] Resampling the digital orthophoto and digital elevation model of the target area respectively to obtain an intermediate digital orthophoto and an intermediate digital elevation model with the same feature dimensions;

[0160] The intermediate digital orthophoto and the intermediate digital elevation model are fused to obtain a first fused image.

[0161] Optionally, the grassland category interpretation model training device further includes a land type map set acquisition module (not shown), and the land type map set acquisition module is used to:

[0162] Query the land type codes related to grassland categories from the preset survey data;

[0163] Obtain the target land class patch set corresponding to the land class code from the land class patches in the survey data.

[0164] Optionally, the grassland category interpretation model training device further includes a sample library construction module (not shown), and the sample library construction module is used to:

[0165] An interpretation sample library is constructed, in which vector data corresponding to the second fused image is stored, the vector data including each second target land class patch, a land class label of each second target land class patch, and geographic location information of each second target land class patch.

[0166] Optionally, the label marking module 30 is further used for:

[0167] Determine the geographical location information of each first target land type patch in the second fused image;

[0168] Determine the correspondence between each first target land class spot and each second target land class spot according to the geographical location information of each first target land class spot and the geographical location information of each second target land class spot;

[0169] Based on the corresponding relationship, each first target land class spot in the second fused image is annotated with a land class label according to the land class label of each second target land class spot to obtain a third fused image.

[0170] Optionally, the model training module 40 is further used to:

[0171] The third fused image is cropped to obtain sample data, wherein the sample data includes a plurality of first cropped images and a land class label of a first target land class patch in each first cropped image;

[0172] Inputting the plurality of first cropped images into the initial grassland category interpretation model to obtain training interpretation results corresponding to the plurality of first cropped images;

[0173] Determine the loss value according to the training interpretation results corresponding to each of the plurality of first cropped images and the land class label of the first target land class patch in each of the first cropped images;

[0174] The parameters of the initial grassland category interpretation model are updated according to the loss value to obtain a trained grassland category interpretation model.

[0175] The grassland category interpretation model training device provided in the embodiment of the present invention adopts the grassland category interpretation model training method in the above embodiment to improve the training accuracy of the model. Compared with the prior art, the grassland category interpretation model training device provided in the embodiment of the present invention has the same beneficial effects as the grassland category interpretation model training method provided in the above embodiment, and other technical features in the grassland category interpretation model training device are the same as the features disclosed in the above embodiment method, which will not be described in detail here.

[0176] The embodiment of the present invention further provides a grassland category interpretation device, the grassland category interpretation device comprising:

[0177] The second fusion processing module is used to fuse the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image;

[0178] A cropping module, used for cropping the fourth fused image to obtain second cropped images;

[0179] An interpretation module, used for inputting each second cropped image into a grassland category interpretation model to obtain respective interpretation results of each second cropped image; wherein the grassland category interpretation model is trained according to the grassland category interpretation model training device described in the above embodiment;

[0180] The splicing module is used to splice the interpretation results to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0181] The grassland category interpretation device provided in the embodiment of the present invention adopts the grassland category interpretation method in the above embodiment, which can avoid the problems of low interpretation efficiency and inaccurate interpretation results caused by manual visual interpretation and field survey, and ensure the objectivity, efficiency and accuracy of grassland category interpretation. Compared with the prior art, the beneficial effects of the grassland category interpretation device provided in the embodiment of the present invention are the same as the beneficial effects of the grassland category interpretation method provided in the above embodiment, and other technical features in the grassland category interpretation device are the same as the features disclosed in the above embodiment method, which will not be repeated here.

[0182] An embodiment of the present invention provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the grassland category interpretation model training method and / or the grassland category interpretation method in the above-mentioned embodiment.

[0183] Reference below Fig.10, which shows a schematic diagram of the structure of an electronic device suitable for implementing an embodiment of the present invention. The electronic device in the embodiment of the present invention may include, but is not limited to, mobile terminals such as mobile phones, notebook computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Descriptions), PMPs (Portable Media Players), vehicle-mounted terminals (such as vehicle-mounted navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Fig.10 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0184] like Fig.10 As shown, the electronic device may include a processing device 1001 (e.g., a central processing unit, a graphics processor, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM: Read Only Memory) 1002 or a program loaded from a storage device 1003 to a random access memory (RAM: Random Access Memory) 1004. In RAM1004, various programs and data required for the operation of the electronic device are also stored. The processing device 1001, ROM1002, and RAM1004 are connected to each other through a bus 1005. An input / output (I / O) interface 1006 is also connected to the bus. Generally, the following devices can be connected to the I / O interface 1006: an input device 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, etc.; an output device 1008 including, for example, a liquid crystal display (LCD: Liquid Crystal Display), a speaker, a vibrator, etc.; a storage device 1003 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 1009. The communication device 1009 can allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although the figure shows an electronic device with various devices, it should be understood that it is not required to implement or have all the devices shown. More or fewer devices can be implemented or provided alternatively.

[0185] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through a communication device, or installed from the storage device 1003, or installed from the ROM 1002. When the computer program is executed by the processing device 1001, the above-mentioned functions defined in the method of the embodiment disclosed in the present invention are executed.

[0186] Compared with the prior art, the beneficial effects of the electronic device provided by the embodiment of the present invention are the same as the beneficial effects of the grassland category interpretation model training method and / or grassland category interpretation method provided by the above-mentioned embodiments, and the other technical features in the electronic device are the same as the features disclosed in the method of the previous embodiment, which will not be repeated here.

[0187] It should be understood that the various parts disclosed in the embodiments of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the description of the above embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0188] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art who is familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0189] An embodiment of the present invention provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, and the computer-readable program instructions are used to execute the grassland category interpretation model training method and / or the grassland category interpretation method in the above-mentioned embodiment.

[0190] The computer-readable storage medium provided in the embodiment of the present invention may be, for example, a USB flash drive, but is not limited to electrical, magnetic, optical, electromagnetic, infrared, or semiconductor devices, devices or components, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM: Random Access Memory), a read-only memory (ROM: Read Only Memory), an erasable programmable read-only memory (EPROM: Erasable Programmable Read Only Memory or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM: CD-Read Only Memory), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution device, device or component. The program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (Radio Frequency: Radio Frequency), etc., or any suitable combination thereof.

[0191] The computer-readable storage medium may be included in the electronic device, or may exist independently without being installed in the electronic device.

[0192] The computer-readable storage medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device performs the following operations: fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image; masking the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are not related to the grassland category; determining the land class labels of each first target land class patch in the second fused image, and annotating the labels to obtain a third fused image; and based on the third fused image, performing masking on the first fused image to remove the non-target related land class patches in the first fused image to obtain a second fused image. The method comprises the following steps: training an initial grassland category interpretation model to obtain a trained grassland category interpretation model, wherein the grassland category interpretation model is used to interpret grassland categories; and fusing the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image; cropping the fourth fused image to obtain second cropped images; inputting each second cropped image into a preset grassland category interpretation model to obtain an interpretation result of each second cropped image, wherein the grassland category interpretation model is trained according to the grassland category interpretation model training method described in the above embodiment; and splicing each interpretation result to obtain a grassland category distribution map corresponding to the area to be interpreted.

[0193] Computer program code for performing the operations of embodiments of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0194] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the device, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flow chart, and the combination of the boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based device that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0195] The modules involved in the embodiments of the present invention may be implemented by software or hardware, wherein the name of the module does not limit the unit itself in some cases.

[0196] The readable storage medium provided by the embodiment of the present invention is a computer-readable storage medium, which stores computer-readable program instructions (i.e., computer programs) for executing the above-mentioned grassland category interpretation model training method and / or grassland category interpretation method. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided by the embodiment of the present invention are the same as the beneficial effects of the grassland category interpretation model training method and / or grassland category interpretation method provided by the above-mentioned embodiment, which will not be elaborated here.

[0197] An embodiment of the present invention also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the grassland category interpretation model training method and / or grassland category interpretation method as described above.

[0198] Compared with the prior art, the beneficial effects of the computer program product provided by the embodiments of the present invention are the same as the beneficial effects of the grassland category interpretation model training method and / or grassland category interpretation method provided by the above embodiments, which will not be elaborated here.

[0199] The above are only some embodiments of the present invention, and are not intended to limit the patent scope of the present invention. All equivalent structural changes made using the contents of the present invention's specification and drawings under the technical concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. A grassland category interpretation model training method, characterized in that: The method comprises: Fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image; Performing mask processing on the first fused image based on a preset target land class patch set to remove non-target related land class patches in the first fused image to obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to grassland categories; Determine the land class label of each first target land class patch in the second fused image, and perform label marking to obtain a third fused image; The initial grassland category interpretation model is trained based on the third fused image to obtain a trained grassland category interpretation model, and the grassland category interpretation model is used to interpret grassland categories.

2. The method according to claim 1, characterized in that The step of fusing the digital orthophoto and the digital elevation model of the target area to obtain a first fused image comprises: Resampling the digital orthophoto and digital elevation model of the target area respectively to obtain an intermediate digital orthophoto and an intermediate digital elevation model with the same feature dimensions; The intermediate digital orthophoto and the intermediate digital elevation model are fused to obtain a first fused image.

3. The method according to claim 1, characterized in that Before the step of performing mask processing on the first fused image based on the preset target land class patch set to remove non-target related land class patches in the first fused image to obtain the second fused image, the step further includes: Query the land type codes related to grassland categories from the preset survey data; A target land type patch set corresponding to the land type code is obtained from the land type patches in the survey data.

4. The method according to claim 1, characterized in that Before the step of determining the land class labels of each first target land class patch in the second fused image and labeling the labels to obtain the third fused image, the step further includes: An interpretation sample library is constructed, in which vector data corresponding to the second fused image is stored, the vector data including each second target land class patch, a land class label of each second target land class patch, and geographic location information of each second target land class patch.

5. The method according to claim 4, characterized in that The step of determining the land class labels of each first target land class patch in the second fused image and labeling the labels to obtain the third fused image includes: Determining geographic location information of each first target land type patch in the second fused image; Determine the correspondence between each of the first target land class spots and each of the second target land class spots according to the geographical location information of each of the first target land class spots and the geographical location information of each of the second target land class spots; Based on the corresponding relationship, each of the first target land class spots in the second fused image is annotated with a land class label according to the land class label of each of the second target land class spots to obtain a third fused image.

6. The method according to claim 5, characterized in that The step of training the initial grassland category interpretation model based on the third fused image to obtain the trained grassland category interpretation model comprises: Performing cropping processing on the third fused image to obtain sample data, wherein the sample data includes a plurality of first cropped images and a land class label of a first target land class patch in each of the first cropped images; Inputting the plurality of first cropped images into an initial grassland category interpretation model to obtain training interpretation results corresponding to each of the plurality of first cropped images; Determine a loss value according to the training interpretation results corresponding to each of the plurality of first cropped images and the land class label of the first target land class patch in each of the first cropped images; The parameters of the initial grassland category interpretation model are updated according to the loss value to obtain a trained grassland category interpretation model.

7. A grassland category interpretation method, characterized in that: The method comprises: fusing the digital orthophoto and the digital elevation model of the area to be interpreted to obtain a fourth fused image; Performing cropping processing on the fourth fused image to obtain second cropped images; Inputting each of the second cropped images into a grassland category interpretation model to obtain respective interpretation results of each of the second cropped images, wherein the grassland category interpretation model is trained according to any one of claims 1 to 6; The interpretation results are spliced ​​together to obtain a grassland category distribution map corresponding to the area to be interpreted.

8. A grassland category interpretation model training device, characterized in that: The device comprises: A first fusion processing module is used to fuse the digital orthophoto and the digital elevation model of the target area to obtain a first fused image; A mask processing module, configured to perform mask processing on the first fused image based on a preset target land class patch set, so as to remove non-target related land class patches in the first fused image, and obtain a second fused image, wherein the non-target related land class patches are land class patches that are irrelevant to grassland categories; a label marking module, used to determine the land class label of each first target land class patch in the second fused image, and perform label marking to obtain a third fused image; The model training module is used to train the initial grassland category interpretation model based on the third fused image to obtain a trained grassland category interpretation model, and the grassland category interpretation model is used to interpret grassland categories.

9. An electronic device, characterized in that: The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program is configured to implement the steps of the grassland category interpretation model training method as described in any one of claims 1 to 6, and / or to implement the steps of the grassland category interpretation method as described in claim 7.

10. A storage medium, characterized in that: The storage medium is 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 grassland category interpretation model training method as described in any one of claims 1 to 6 are implemented, and / or the steps of the grassland category interpretation method as described in claim 7 are implemented.