Land property detection method based on spatio-temporal adaptive fusion and geographical features

Abstract

The application provides a land property detection method based on space-time adaptive fusion and geographical features, and relates to the technical field of image processing, and comprises the following steps: acquiring a sampling image, land attribute data and digital elevation model data of a land block to be detected; extracting image features from the sampling image by adopting a Swin Transformer model combined with a ground type texture adaptive fusion module GTAF; converting the land attribute data into a node and edge form to acquire attribute features by constructing a land block relation graph; and extracting land block terrain features from the digital elevation model data by using a MobileNetV3 network; the image features comprise land block texture features, land block color features and land block structure features; the attribute features comprise land block area features, land block coordinate features and land block type features; the land block terrain features comprise slope features and slope direction features; the image features, the attribute features and the land block terrain features are weighted and fused by using an adaptive mutual enhancement attention module Ada-MBA; and the land property of the land block to be detected is determined based on the weighted and fused features.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, and in particular to a land property detection method based on spatio-temporal adaptive fusion and geographical features. BACKGROUND

[0002] In the field of land resource detection and management, accurately identifying and judging the land properties of plots, such as farmland, forest land, and construction land, is a core task for realizing efficient use, dynamic monitoring, and scientific planning of land resources. Traditional land property discrimination methods mainly rely on manual field investigation and manual comparison in the office. Field investigation usually involves field measurement, sampling, and recording, which is time-consuming, labor-intensive, and inefficient, and is limited by the subjective experience of investigators, which can lead to inconsistencies and errors in the classification results. Office comparison relies on manual analysis of remote sensing images or cadastral data, which requires a lot of manual annotation and data processing work, and is not only costly, but also difficult to meet the needs of large-scale and high-frequency monitoring. In addition, traditional methods often have low accuracy when dealing with complex terrain (such as mountainous areas and plateaus) or dynamic change scenarios (such as land use changes), due to insufficient data resolution or outdated updates, making it difficult to meet the requirements of modern land resource management for refinement and real-time.

[0003] In recent years, with the rapid development of remote sensing technology and artificial intelligence technology, image processing methods based on deep learning have provided new possibilities for land property discrimination and change detection. Convolutional neural networks (CNN) and generative adversarial networks (GAN) have made significant progress in remote sensing image analysis, and can extract texture, color, and structural features of the land surface from high-resolution images to achieve automated land property classification. However, existing methods mostly focus on single modal data, and do not make full use of high-resolution photos collected in the field. SUMMARY

[0004] Therefore, the present application provides a land property detection method based on spatio-temporal adaptive fusion and geographical features.

[0005] The technical solution of the present application is as follows: The present application provides a land property detection method based on spatio-temporal adaptive fusion and geographical features in the first aspect, comprising:

[0006] obtaining a sample image, plot attribute data, and digital elevation model data of a plot to be detected;

[0007] The Swin Transformer model is combined with a ground type texture adaptive fusion module GTAF to extract image features from the sampling images, attribute features are obtained by converting the parcel attribute data into nodes and edges in the form of a parcel relationship graph, and a MobileNetV3 network is used to extract parcel terrain features from the digital elevation model data; the image features include parcel texture features, parcel color features, and parcel structure features, the attribute features include parcel area features, parcel coordinate features, and parcel type features; and the parcel terrain features include slope features and slope direction features.

[0008] An adaptive mutual enhancement attention module Ada-MBA is used to perform weighted fusion on the image features, the attribute features, and the parcel terrain features, and the land property of the parcel to be detected is determined based on the weighted fusion features.

[0009] Based on the above technical solutions, preferably, the sampling images, the parcel attribute data, and the digital elevation model data of the parcel to be detected are obtained, including:

[0010] The sampling images of the parcel to be detected are obtained by an external sampling device at different time nodes and different shooting angles;

[0011] The area data, coordinate data, and type data of the parcel to be detected are obtained from a cadastral database, the area data and the coordinate data are normalized, and the type data is labeled using one-hot encoding to obtain parcel attribute data;

[0012] The digital elevation model data is generated based on the ground elevation data collected by a drone.

[0013] Based on the above technical solutions, preferably, the Swin Transformer model is combined with the ground type texture adaptive fusion module GTAF to extract image features from the sampling images, including:

[0014] The sampling images are converted into RGB three-channel images and are divided into 4 4 blocks to generate a first image feature with a shape of 56 56 96;

[0015] A four-stage structure is used to extract multi-scale features from the first image feature, and a second image feature with a shape of 7 7 768 is output;

[0016] The second image feature is processed by global average pooling and a fully connected layer to generate the parcel texture features, the parcel color features, and the parcel structure features.

[0017] Preferably, based on the above technical solutions, the attribute feature is obtained by converting the plot attribute data into a node and an edge through the construction of a plot relationship graph, comprising:

[0018] Taking the unique identifier of the plot to be detected as a node, obtaining an edge based on spatial adjacency relationship and semantic similarity, and constructing a plot relationship graph;

[0019] Extracting the plot area feature, the plot coordinate feature and the plot type feature from the plot relationship graph by using a GraphSAGE model

[0020] Preferably, based on the above technical solutions, the plot terrain feature is extracted from the digital elevation model data by using a MobileNetV3 network, comprising:

[0021] The digital elevation model data of different scales is normalized and data enhanced, and then input into the MobileNetV3 network after adjusting the channel number to obtain the plot terrain feature.

[0022] Preferably, based on the above technical solutions, the image feature, the attribute feature and the plot terrain feature are weighted and fused by using an adaptive mutual enhancement attention module Ada-MBA, and the land property of the plot to be detected is determined based on the weighted fusion feature, comprising:

[0023] The image feature, the attribute feature and the plot terrain feature are linearly transformed and normalized, and a full connection layer is used to generate respective corresponding attention weights;

[0024] The image feature, the attribute feature and the plot terrain feature are weighted and fused based on the attention weights, and the land property of the plot to be detected is determined based on the weighted fusion feature.

[0025] Preferably, based on the above technical solutions, the image feature, the attribute feature and the plot terrain feature are weighted and fused by using an adaptive mutual enhancement attention module Ada-MBA, and the land property of the plot to be detected is determined based on the weighted fusion feature, comprising:

[0026] Obtaining a first weighted fusion feature at a first time point and a second weighted fusion feature at a second time point;

[0027] Determining the land property change information of the plot to be detected based on the norm difference between the first weighted fusion feature and the second weighted fusion feature.

[0028] Further preferably, the second aspect of the present application provides a land property detection device based on spatio-temporal adaptive fusion and geographical features, comprising: a data acquisition module, a feature extraction module and a feature fusion module; wherein,

[0029] The data acquisition module is configured to acquire a sampling image of a land block to be detected, land block attribute data and digital elevation model data;

[0030] The feature extraction module is configured to extract image features from the sampling image by using a Swin Transformer model combined with a ground class texture adaptive fusion module GTAF, convert the land block attribute data into node and edge forms to obtain attribute features by constructing a land block relationship graph, and extract land block terrain features from the digital elevation model data by using a MobileNetV3 network; the image features include land block texture features, land block color features and land block structure features, the attribute features include land block area features, land block coordinate features and land block type features, and the land block terrain features include slope features and slope direction features;

[0031] The feature fusion module is configured to perform weighted fusion on the image features, the attribute features and the land block terrain features by using an adaptive mutual enhancement attention module Ada-MBA, and determine the land property of the land block to be detected based on the weighted fusion features.

[0032] Further preferably, the third aspect of the present application provides an electronic device comprising a processor and a memory; the memory stores a computer program, wherein the computer program, when executed by the processor, implements the land property detection method based on spatio-temporal adaptive fusion and geographical features of the first aspect.

[0033] Further preferably, the fourth aspect of the present application provides a computer storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the land property detection method based on spatio-temporal adaptive fusion and geographical features of the first aspect.

[0034] The land property detection method based on spatio-temporal adaptive fusion and geographical features of the present application has the following beneficial effects compared with the prior art:

[0035] 1. The Swin Transformer model is used in combination with the ground class texture adaptive fusion module GTAF to extract image features from the sampling image, which can comprehensively capture the texture, color and structural features of the land block. By constructing a land block relationship graph, the land block attribute data is converted into node and edge forms to obtain attribute features, which cover key information such as land block area, coordinates and type. The MobileNetV3 network is used to extract land block terrain features from digital elevation model data, which reduces the consumption of computing resources while ensuring the effectiveness of feature extraction. Comprehensive utilization of multi-dimensional features and adaptive weighted fusion can more comprehensively and accurately describe the features of the land block to be detected, thereby improving the accuracy of land property detection.

[0036] 2. The adaptive mutual enhancement attention module Ada-MBA is used to weight and fuse image features, attribute features and land block terrain features. According to the importance of different features to land property judgment, the weights of each feature are dynamically adjusted, so that the fused features can highlight the key information and improve the accuracy of land property detection. The mutual enhancement attention mechanism of the Ada-MBA module can promote the information interaction and mutual enhancement between different features. By focusing on the correlation between different features, the potential relationship between features can be mined, further improving the quality of the fused features.

[0037] 3. By using multiple feature extraction methods and adaptive fusion strategies, it has strong adaptability to different types of data and land blocks, can maintain good detection performance under different environmental conditions, and enhances the robustness of land property detection, providing a scientific basis for land planning, development and management. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0039] Figure 1 A flowchart of a land property detection method based on spatio-temporal adaptive fusion and geographic features provided by an embodiment of the present application;

[0040] Figure 2 A structural schematic diagram of a land property detection device based on spatio-temporal adaptive fusion and geographic features provided by an embodiment of the present application;

[0041] Figure 3 A structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0043] In some embodiments, as shown in Figure 1 , Figure 1 A flowchart of a land property detection method based on spatio-temporal adaptive fusion and geographic features is provided for the embodiments of the present application. The land property detection method based on spatio-temporal adaptive fusion and geographic features provided by the present application comprises the following steps:

[0044] S110, acquiring a sampling image, land attribute data and digital elevation model data of a land parcel to be detected.

[0045] The sampling image can include a visible light image, a multispectral image or a hyperspectral image, etc. The visible light image can intuitively present the appearance of the land parcel, such as the vegetation coverage, the building distribution, etc. The multispectral and hyperspectral images contain more band information, which can reflect the physical and chemical properties of the land parcel, such as the health status of the vegetation, the composition of the soil, etc. The land attribute data includes the area, shape and coordinates of the land parcel. The area of the land parcel determines the scale of land use, the shape may affect the development and utilization mode of the land, and the coordinates are used to determine the position of the land parcel in the geographical space. The digital elevation model DEM data describes the terrain undulation of the land parcel.

[0046] In some embodiments, S110, acquiring a sampling image, land attribute data and digital elevation model data of a land parcel to be detected, comprises:

[0047] Using an external sampling device to acquire the sampling image of the land parcel to be detected at different time nodes and different shooting angles;

[0048] Acquiring the area data, coordinate data and type data of the land parcel to be detected from a cadastral database, normalizing the area data and coordinate data, and marking the type data using one-hot encoding to obtain the land attribute data;

[0049] Generating the digital elevation model data based on the ground elevation data collected by the unmanned aerial vehicle.

[0050] In this embodiment, image data of the land parcel can be obtained by means of sensors carried by satellites, cameras carried by airplanes or drones, etc. The ownership information and basic attributes of the land parcel can be obtained by consulting relevant archives and data of the land registration department, such as the cadastral database. These data are the legal basis for land management and have high accuracy and authority. Of course, they can also be obtained through field investigation. The three-dimensional coordinate information of the ground can be obtained by scanning the land parcel with a laser radar device carried by a drone, and then a digital elevation model can be generated, or the elevation information of the ground can be extracted by processing and analyzing aerial or satellite images to generate a digital elevation model.

[0051] In S120, the Swin Transformer model combined with the ground class texture adaptive fusion module GTAF is used to extract image features from the sampled images, the land parcel attribute data is converted into node and edge forms to obtain attribute features by constructing a land parcel relationship graph, and the MobileNetV3 network is used to extract land parcel terrain features from the digital elevation model data; the image features include land parcel texture features, land parcel color features and land parcel structure features, the attribute features include land parcel area features, land parcel coordinate features and land parcel type features, and the land parcel terrain features include slope features and slope direction features.

[0052] The Swin Transformer is a visual model based on the Transformer architecture, which can process image information of different scales through a windowed self-attention mechanism, and is suitable for extracting texture, color and structure features in land parcel images. The GTAF module can adaptively fuse different features according to the ground class texture, enhancing the sensitivity and adaptability of the model to specific ground class textures. The MobileNetV3 can reduce the consumption of computing resources while ensuring the effect of feature extraction.

[0053] In some embodiments, in S120, the Swin Transformer model combined with the ground class texture adaptive fusion module GTAF is used to extract image features from the sampled images, including:

[0054] The sampled images are converted into RGB three-channel images and divided into 4 4 blocks of size 56 56 96 to generate a first image feature;

[0055] A four-stage structure is used to extract multi-scale features from the first image feature, and a second image feature with a shape of 7 7 768 is output;

[0056] The second image feature is processed by global average pooling and a fully connected layer to generate land parcel texture features, land parcel color features and land parcel structure features.

[0057] In this embodiment, the image features include plot texture features, plot color features and plot structure features. Taking the plot texture features as an example, the texture features are extracted based on a gray level co-occurrence matrix:

[0058] ;

[0059] wherein, , , are respectively the channel values of the photo red-green-blue channel at position (i, j), with a value range of ([0, 1]), is the gray value at position (i, j), is the quantized gray level, and L is the number of gray levels, is the number of times that the gray levels k and l are adjacent in the horizontal direction, is the normalized probability, is the texture complexity, , is the empirical range of the texture complexity, is the normalized texture complexity.

[0060] The RGB three-channel photo is divided into 4 4-sized blocks, to generate initial features 56 56 96, and then multi-scale features are extracted using a four-stage structure, to output 7 7 768, and image features are generated through global average pooling and a fully connected layer :

[0061] ;

[0062] wherein, is the image feature, is the input image data, is the fully connected layer, is the global average pooling, is the query, key and value matrices, is the dimension of the key, characterizes the multi-head attention mechanism calculation.

[0063] The purpose of the Geo-Texture Adaptive Fusion (GTAF) module is to fuse the initial convolutional features Fconv and the Gaber texture features Fgaber, in combination with the environmental input , T:

[0064] ;

[0065] wherein, is a concatenation feature, , is a global average pooling and a max pooling, , is a channel weight, respectively corresponding to and channels of is a spatial weight, is a fusion feature, is an initial feature of a Swin Transformer, characterizes a mapping operation, characterizes a data form conversion, , respectively characterize a 1x1 convolution and a 3x3 convolution.

[0066] In some embodiments, S120 converts the plot attribute data into a form of nodes and edges by constructing a plot relationship graph to obtain an attribute feature, including:

[0067] Taking a unique identifier of a plot to be detected as a node, obtaining edges based on spatial adjacency relationship and semantic similarity, and constructing a plot relationship graph;

[0068] Using a GraphSAGE model to extract plot area features, plot coordinate features and plot type features from the plot relationship graph.

[0069] In this embodiment, the area and the coordinate are normalized respectively; the area normalization is:

[0070] ;

[0071] wherein, is a normalized area, is a plot area, , is a mean and a standard deviation of the area of the data set.

[0072] The coordinate normalization is:

[0073] ;

[0074] wherein, , is a normalized coordinate, x and y are plot center coordinates, is a coordinate range of plot data.

[0075] A plot relationship graph G=(V, E, X) is constructed, wherein the node V is a unique identifier of a plot entity, the feature X includes normalized area, coordinate and plot type, and the edge E is based on spatial adjacency (distance <d) and semantic similarity.

[0076] .

[0077] wherein, is the Euclidean distance of land i and j, and is the normalized coordinate of land i, is the area relative difference, is the area of land i, is the area of land j.

[0078] A 64-dimensional feature vector Fattr is extracted using GraphSAGE, and the node features are updated by sampling neighbors and mean aggregation:

[0079] ;

[0080] wherein, is the initial feature, is the feature of node v at layer l, is the sampled neighbor set, is the mean of neighbor features, is the weight matrix, is the ReLU activation function, is the attribute feature vector, denotes the concatenation operation, is the square of the feature of node v.

[0081] In some embodiments, S120 extracts landform features of land blocks from digital elevation model data using a MobileNetV3 network, including:

[0082] The digital elevation model data of different scales is normalized and data-augmented, and then input to the MobileNetV3 network after adjusting the number of channels to obtain landform features of land blocks.

[0083] In this embodiment, the normalized and data-augmented digital elevation model (DEM) data is input into the MobileNetV3 network. The light and efficient structure of the network can effectively extract deep terrain feature information contained in the DEM data, thereby assisting in identifying spatial features such as slope change, slope distribution, and terrain undulation of land blocks.

[0084] The slope and aspect can be represented as:

[0085] ;

[0086] wherein, slope is the slope, aspect is the aspect, z is the elevation value of the dem data, , is the gradient of the height along the x, y direction, is the conversion factor of the radian turning angle.

[0087] S130, the adaptive mutual enhancement attention module Ada-MBA is used to weight and fuse the image features, attribute features and landform features of the plot, and the land property of the to-be-detected plot is determined based on the weighted fusion features.

[0088] The Ada-MBA module can dynamically adjust the weights of the features according to the importance of the input features through the adaptive mechanism, and promote the information interaction and mutual enhancement between different features through the mutual enhancement attention mechanism, helping to capture the potential relationship between the features, thereby improving the quality of the fused features.

[0089] In some embodiments, S130, the adaptive mutual enhancement attention module Ada-MBA is used to weight and fuse the image features, attribute features and landform features of the plot, and the land property of the to-be-detected plot is determined based on the weighted fusion features, including:

[0090] The image features, attribute features and landform features are linearly transformed and normalized, and the full connection layer is used to generate the corresponding attention weights;

[0091] The image features, attribute features and landform features are weighted and fused based on the attention weights, and the land property of the to-be-detected plot is determined based on the weighted fusion features.

[0092] In this embodiment, the adaptive mutual enhancement attention module is used to fuse the image features, attribute features and landform features of the plot:

[0093] ;

[0094] wherein, is the environmental feature, is the first layer weight and bias, and Z is the intermediate feature, is the second layer weight and bias, is the branch weight, is the fusion feature, is the landform feature of the plot, , is the adjustment coefficient.

[0095] The classification principle is: ;

[0096] wherein, FC is the full connection layer, and P is the type probability distribution.

[0097] In some embodiments, S130, the image features, attribute features and landform features are weighted and fused by using an adaptive mutual enhancement attention module Ada-MBA, the land property of the to-be-detected land plot is determined based on the weighted and fused features, including:

[0098] The first weighted and fused features at the first time point and the second weighted and fused features at the second time point are obtained;

[0099] The land property change information of the to-be-detected land plot is determined based on the norm difference of the first weighted and fused features and the second weighted and fused features.

[0100] The norm difference can quantify the change degree between the features at two time points, even a very small change can be detected, so that subtle changes in land property can be found in time, and strong support is provided for land dynamic monitoring.

[0101] In one example, the fused features of two time points and are compared, and the L2 norm difference is calculated.

[0102] ;

[0103] wherein, 、 is the fused feature of time point 、 , is the feature L2 distance, is the change threshold, Change is the change flag, 1 represents change, and 0 represents no change.

[0104] In some embodiments, referring to Figure 2 , Figure 2 is a structural schematic diagram of a land property detection device based on spatio-temporal adaptive fusion and geographic features provided by an embodiment of the present application. The present application provides a land property detection device 200 based on spatio-temporal adaptive fusion and geographic features, comprising: a data acquisition module 210, a feature extraction module 220 and a feature fusion module 230; wherein,

[0105] The data acquisition module 210 is configured to acquire sampling images, land attribute data and digital elevation model data of the to-be-detected land plot;

[0106] The feature extraction module 220 is configured to extract image features from the sampling images by using a Swin Transformer model combined with a ground type texture adaptive fusion module GTAF, convert ground block attribute data into node and edge forms to obtain attribute features by constructing a ground block relationship graph, and extract ground block terrain features from digital elevation model data by using a MobileNetV3 network; the image features include ground block texture features, ground block color features, and ground block structure features; the attribute features include ground block area features, ground block coordinate features, and ground block type features; and the ground block terrain features include slope features and slope direction features.

[0107] The feature fusion module 230 is configured to perform weighted fusion on the image features, the attribute features, and the ground block terrain features by using an adaptive mutual enhancement attention module Ada-MBA, and determine the land property degree of the ground block to be detected based on the weighted fusion features.

[0108] In some embodiments, the data acquisition module 210 is specifically configured to:

[0109] acquire sampling images of the ground block to be detected at different time nodes and from different shooting angles by using external sampling equipment;

[0110] acquire area data, coordinate data, and type data of the ground block to be detected from a cadastral database, normalize the area data and the coordinate data, and mark the type data by using one-hot encoding to obtain ground block attribute data;

[0111] generate digital elevation model data based on surface elevation data collected by a drone.

[0112] In some embodiments, the feature extraction module 220 is specifically configured to:

[0113] convert the sampling images into RGB three-channel images, and divide the images into 4 4 size blocks to generate first image features with a shape of 56 56 96;

[0114] extract multi-scale features from the first image features by using a four-stage structure, and output second image features with a shape of 7 7 *768;

[0115] perform global average pooling and full connection layer processing on the second image features to generate ground block texture features, ground block color features, and ground block structure features.

[0116] In some embodiments, the feature extraction module 220 is specifically configured to:

[0117] A unique identifier of the to-be-detected plot is taken as a node, edges are obtained based on spatial adjacency relationship and semantic similarity, and a plot relationship graph is constructed;

[0118] The plot area feature, the plot coordinate feature and the plot type feature are extracted from the plot relationship graph by using a GraphSAGE model

[0119] In some embodiments, the feature extraction module 220 is specifically configured to:

[0120] The digital elevation model data of different scales is normalized and subjected to data enhancement processing, and is input into a MobileNetV3 network after adjusting the number of channels to obtain the plot terrain feature.

[0121] In some embodiments, the feature fusion module 230 is specifically configured to:

[0122] The image feature, the attribute feature and the plot terrain feature are subjected to linear transformation and normalization processing, and respective corresponding attention weights are generated by using a fully connected layer;

[0123] The image feature, the attribute feature and the plot terrain feature are weightedly fused based on the attention weights, and the land property of the to-be-detected plot is determined based on the weightedly fused feature.

[0124] In some embodiments, the feature fusion module 230 is specifically configured to:

[0125] The first weightedly fused feature at the first time point and the second weightedly fused feature at the second time point are obtained;

[0126] The land property change information of the to-be-detected plot is determined based on the norm difference between the first weightedly fused feature and the second weightedly fused feature.

[0127] It should be noted that the land property detection device based on spatio-temporal adaptive fusion and geographical features provided in the embodiments of the present application and the land property detection method based on spatio-temporal adaptive fusion and geographical features provided in the embodiments of the present application are based on the same application concept, and therefore the specific implementation of this embodiment can be referred to the implementation of the aforementioned land property detection method based on spatio-temporal adaptive fusion and geographical features, and the repeated parts will not be described herein.

[0128] In some embodiments, please refer to Figure 3 , Figure 3 A structural schematic diagram of an electronic device provided in the embodiments of the present application. The electronic device 300 provided in the embodiments of the present application includes a processor 310 and a memory 320; the memory 320 stores a computer program, wherein the computer program is implemented when executed by the processor to realize the aforementioned land property detection method based on spatio-temporal adaptive fusion and geographical features.

[0129] In particular, the processor 310 can include a general purpose microprocessor, a set of instructions, and / or a related chip set, and / or a special purpose microprocessor, such as an application specific integrated circuit (ASIC), etc. The processor 310 can also include on-board memory for cache use. The processor 310 can be a single processing unit or multiple processing units for executing different actions of the method processes according to embodiments of the present application.

[0130] The memory 320 may, for example, be any medium capable of containing, storing, communicating, propagating or transporting instructions. For example, the memory 320 can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared or semiconductor system, apparatus, device or propagation medium. Specific examples of the memory 320 include a magnetic storage device, such as a hard disk drive (HDD); an optical storage device, such as a compact disc (CD-ROM); also a random access memory (RAM) or a flash memory; and / or a wired / wireless communication link.

[0131] The present application also provides a computer readable medium having stored thereon a computer program, which, when executed by a processor, implements the above-mentioned land property detection method based on spatio-temporal adaptive fusion and geographical features. The computer readable medium can be included in the device / apparatus / system described in the above embodiments; or can exist separately without being assembled into the device / apparatus / system. The above computer readable medium carries one or more programs, which, when executed, implement the method according to embodiments of the present application.

[0132] According to embodiments of the present application, the computer readable medium can be a computer readable signal medium or a computer readable storage medium or any combination thereof. The computer readable storage medium can be, for example but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present application, the computer readable storage medium can be any tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. In the present application, the computer readable signal medium can include a computer readable program code that can be transmitted or propagated over a computer readable medium, in baseband or as part of a carrier wave. Such a computer readable program code can take many forms, including but not limited to, an electromagnetic signal, an optical signal, or any suitable combination of the foregoing. The computer readable signal medium can also be any computer readable medium that can be transitory, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer readable medium can be transmitted or propagated using any suitable medium, including but not limited to wireless, wired, optical fiber cable, RF, and the like, or any suitable combination of the foregoing.

[0133] It is to be understood that the features mentioned herein can be combined in various ways without departing from the application. Although the application has been described with reference to specific examples, it will be appreciated that the application can be practiced otherwise than is specifically described. For example, the order of various steps can be changed, or replaced or eliminated altogether. Each document, patent, patent application or submission cited herein is specifically and individually incorporated herein by reference in its entirety. It is expressly not admitted that any of the documents, patents, patent applications or submissions cited herein are directly or indirectly comprised in, or prior art to, any element of the present application.

Claims

1. A method for detecting land properties based on spatio-temporal adaptive fusion and geographical features, characterized in that, The method comprises the following steps: obtaining a sampling image, land attribute data and digital elevation model data of a land block to be detected; extracting image features from the sampling image by using a Swin Transformer model combined with a ground type texture adaptive fusion module GTAF, converting the land attribute data into node and edge forms to obtain attribute features by constructing a land block relationship graph, and extracting land block terrain features from the digital elevation model data by using a MobileNetV3 network; the image features include land block texture features, land block color features and land block structure features, the attribute features include land block area features, land block coordinate features and land block type features, and the land block terrain features include slope features and slope direction features; performing linear transformation and normalization processing on the image features, the attribute features and the land block terrain features, and generating respective corresponding attention weights by using a fully connected layer; performing weighted fusion on the image features, the attribute features and the land block terrain features based on the attention weights, obtaining first weighted fusion features at a first time point and second weighted fusion features at a second time point, determining land property change information of the land block to be detected based on norm differences of the first weighted fusion features and the second weighted fusion features; the weighted fusion of the image features, the attribute features and the land block terrain features based on the attention weights is as follows: ； wherein, is an environmental feature, is a first layer weight and bias, Z is an intermediate feature, is a second layer weight and bias, is a branch weight, is a fused feature, is a parcel terrain feature, , is a regulation coefficient.

2. The method for detecting land properties based on spatio-temporal adaptive fusion and geographical features according to claim 1, wherein, The method comprises the following steps: obtaining a sampling image, land attribute data and digital elevation model data of a land block to be detected; obtaining the sampling image of the land block to be detected at different time nodes and different shooting angles by using an external sampling device; obtaining area data, coordinate data and type data of the land block to be detected from a cadastral database, performing normalization on the area data and the coordinate data, and marking the type data by using one-hot encoding to obtain land attribute data; 3. The method of claim 1, wherein the method is based on spatio-temporal adaptive fusion and geographical features for land property detection. generating digital elevation model data based on ground elevation data collected by a drone. transforming the sampled image into an RGB three-channel image and segmenting into 4 4-sized blocks, generating a first image feature of shape 56 56 96 extracting multi-scale features from the first image features using a four-stage structure, outputting a shape of 7 7 768 second image features; The method comprises the following steps:

4. The method for detecting land properties based on spatio-temporal adaptive fusion and geographical features according to claim 1, wherein, processing the second image features by global average pooling and a fully connected layer to generate the land block texture features, the land block color features and the land block structure features. The method comprises the following steps: constructing a land block relationship graph by taking a unique identifier of the land block to be detected as a node and obtaining edges based on spatial adjacency relationship and semantic similarity; 5.The land property detection method based on spatio-temporal adaptive fusion and geographical features according to claim 1, wherein, extracting the land block area features, the land block coordinate features and the land block type features from the land block relationship graph by using a GraphSAGE model. The method comprises the following steps:

6. A device for detecting land properties based on spatio-temporal adaptive fusion and geographical features, characterized in that, performing normalization and data enhancement processing on the digital elevation model data of different scales, adjusting the number of channels, and inputting the digital elevation model data into a MobileNetV3 network to obtain land block terrain features. The method comprises the following steps: The data acquisition module, the feature extraction module and the feature fusion module; wherein The data acquisition module is configured to acquire sampling images, land attribute data and digital elevation model data of the land to be detected. The feature extraction module is configured to extract image features from the sampling images by using a Swin Transformer model combined with a ground texture adaptive fusion module GTAF, convert the land attribute data into node and edge forms to obtain attribute features by constructing a land relationship graph, and extract land terrain features from the digital elevation model data by using a MobileNetV3 network; the image features include land texture features, land color features and land structure features, the attribute features include land area features, land coordinate features and land type features, and the land terrain features include slope features and slope direction features; The feature fusion module is configured to perform linear transformation and normalization processing on the image features, the attribute features and the land terrain features, generate respective corresponding attention weights by using a fully connected layer, perform weighted fusion on the image features, the attribute features and the land terrain features based on the attention weights, acquire first weighted fusion features at a first time point and second weighted fusion features at a second time point, and determine land property change information of the land to be detected based on norm differences of the first weighted fusion features and the second weighted fusion features; the weighted fusion of the image features, the attribute features and the land terrain features based on the attention weights is as follows: ； wherein, is an environmental feature, is a first layer weight and bias, Z is an intermediate feature, is a second layer weight and bias, is a branch weight, is a fused feature, is a parcel terrain feature, , is a regulation coefficient.

7. An electronic device comprising a processor and a memory; said memory having stored a computer program, wherein, The computer program, when executed by the processor, implements the land property detection method based on spatio-temporal adaptive fusion and geographical features in any one of claims 1 to 5.

8. A computer storage medium, characterized in that The computer program, when executed by the processor, implements the land property detection method based on spatio-temporal adaptive fusion and geographical features in any one of claims 1 to 5.

Images