Cultivated land quality evaluation method and device, computer equipment and medium

Abstract

The invention belongs to the technical field of cultivated land quality evaluation, and discloses a cultivated land quality evaluation method and device, computer equipment and a medium. The method comprises the following steps: acquiring land grid data of a cultivated land to be evaluated, and performing comparative learning on features of the land grid data by a backbone network to obtain context features; converting the land grid data into three-dimensional tensor data; inputting the three-dimensional tensor data into a CNN module, and extracting potential local features in the three-dimensional tensor data to obtain spatial features; combining the spatial features with the context features to obtain comprehensive features, and inputting the comprehensive features into a full connection layer to obtain a spatial prediction value; extracting topological attribute characteristics in the three-dimensional tensor data by using a GNN module to obtain a topological predicted value; and fusing the space prediction value and the topology prediction value to obtain an evaluation result of the quality of the cultivated land to be evaluated. The accuracy of soil quality evaluation can be improved.

Description

Technical Field

[0001] The present application relates to the technical field of arable land quality assessment, and in particular to an assessment method, device, computer equipment and medium for arable land quality. Background Art

[0002] The quality evaluation of cultivated land resources is crucial to the realization of cultivated land protection. At present, the quality evaluation of cultivated land resources is mostly done manually, and experts in the field are organized to conduct a comprehensive analysis and judgment on the situation of each plot of land, and then the quality of the plot is obtained. At the same time, some mathematical models are established, and some mathematical methods such as the Delphi method and the entropy weight method are used to evaluate the importance of various factors affecting the quality of cultivated land, and then the quality of each plot is scored. In addition, some methods based on deep learning models and end-to-end training paradigms have emerged to evaluate the quality of cultivated land.

[0003] However, these technologies currently have some limitations. The manual judgment process is cumbersome and inefficient, and the evaluation results are easily affected by subjective factors. The method based on mathematical models lacks complex nonlinear modeling capabilities. In the task of arable land quality assessment, the relationship between influencing factors and arable land quality is not linear, so the mathematical model cannot effectively model this relationship, resulting in inaccurate quality evaluation results. The current deep learning-based methods ignore the widespread problem of data imbalance in actual application scenarios. The uneven distribution of data leads to difficulties in training deep learning classification models, because the model often shows bias towards categories with more training data, and has poor evaluation performance for scarce data categories, such as high-quality arable land. Summary of the invention

[0004] To this end, the embodiments of the present application provide a method, apparatus, computer equipment and medium for evaluating the quality of cultivated land, thereby improving the evaluation performance of the quality of cultivated land and obtaining more accurate evaluation results.

[0005] In a first aspect, the present application provides a method for evaluating the quality of arable land.

[0006] This application is achieved through the following technical solutions:

[0007] A method for evaluating the quality of cultivated land is implemented based on a deep learning model SAFE, wherein the deep learning model SAFE includes a backbone network, a CNN module and a GNN module, including:

[0008] Obtaining land grid data of the cultivated land to be evaluated, inputting the land grid data into the backbone network, and the backbone network performing comparative learning on the features of the land grid data to obtain context features of the land grid data;

[0009] Associating the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data;

[0010] Inputting the three-dimensional tensor data into a CNN module, extracting potential local features in the three-dimensional tensor data, and obtaining spatial features;

[0011] Combining the spatial features with the context features to obtain comprehensive features, and inputting the comprehensive features into a fully connected layer to obtain a spatial prediction value of the cultivated land to be evaluated;

[0012] Inputting the three-dimensional tensor data into the GNN module, extracting topological attribute features in the three-dimensional tensor data, predicting the topological attribute features, and obtaining a topological prediction value of the cultivated land to be evaluated;

[0013] The spatial prediction value and the topological prediction value are integrated to obtain an evaluation result of the quality of the cultivated land to be evaluated.

[0014] In a preferred example of the present application, it can be further configured that before the three-dimensional tensor data is input into the CNN module, it also includes:

[0015] The three-dimensional data tensor is divided into each grid x i The center of the image is taken as the center and divided into local matrix slices of size C×C, where the cropping parameter C is set to 30~35.

[0016] In a preferred example of the present application, it can be further configured that the land grid data is input into the backbone network, and the backbone network performs comparative learning on the features of the land grid data, including:

[0017] In the backbone network, the distance between the input sample and the anchor point sample is measured based on cosine similarity, and the loss function of contrastive learning is set as:

[0018]

[0019] Among them, when the input sample and the anchor sample belong to the same category, the loss function is minimized to make the input sample and the anchor sample close:

[0020]

[0021] When the input sample and the anchor sample belong to different categories, the loss function is minimized to separate the input sample and the anchor sample:

[0022]

[0023] in, represents the loss function, represents the cosine similarity, represents the contrastive learning loss function value of the i-th sample, represents the contrastive learning loss function value of the sample with the i-th predicted value of 1, represents the contrastive learning loss function value of the sample whose prediction value is 0. represents the feature vector of the i-th input sample, represents the feature vector of the anchor point sample, Represents the predicted value of the i-th input sample, ∈ represents the threshold that controls whether to perform comparative learning loss calculation on input samples and anchor samples of different categories.

[0024] In a preferred example of the present application, it can be further configured to associate the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data, including:

[0025] Get the longitude and latitude sets of all grids from the geohash attribute of the land grid data;

[0026] Sort the longitudes in the longitude collection and the latitudes in the latitude collection, and add the longitude index and latitude index to each grid;

[0027] Based on the longitude index and latitude index, the attribute vector of the land grid data is mapped to the corresponding grid position to obtain three-dimensional tensor data.

[0028] In a preferred example of the present application, it can be further configured that the comprehensive features are input into the fully connected layer to obtain the spatial prediction value of the cultivated land to be evaluated, including:

[0029]

[0030] in, Indicates that the CNN module extracts spatial features, D2 is the dimension of the extracted spatial features, represents the fully connected layer, represents three-dimensional tensor data, represents the context features extracted by the backbone network, X i Represents the land grid data of the i-th grid.

[0031] In a preferred example of the present application, it can be further configured that the three-dimensional tensor data is input into the GNN module, and the topological attribute features in the three-dimensional tensor data are extracted, including:

[0032] Inputting the three-dimensional tensor data into the GNN module, performing an aggregation operation on adjacent node features based on an aggregation function to obtain an aggregated feature;

[0033] Based on the update function, the aggregated features are nonlinearly transformed to generate updated node representations, and attribute interaction information in the updated node representations is extracted to obtain topological attribute features.

[0034] In a preferred example of the present application, it can be further configured that the topological attribute characteristics are predicted to obtain the topological prediction value of the cultivated land to be evaluated, including:

[0035] A multilayer perceptron is used to predict the topological attribute features, and a ReLU function is used as an activation function.

[0036] In a second aspect, the present application provides a device for evaluating the quality of cultivated land.

[0037] This application is achieved through the following technical solutions:

[0038] A device for evaluating the quality of cultivated land, the device being used to execute the evaluation method described in the first aspect, the device comprising:

[0039] A data acquisition module, used to acquire land grid data of the cultivated land to be evaluated; and used to associate the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data;

[0040] A feature extraction module is used to compare and learn the features of the land grid data through a backbone network to obtain context features of the land grid data; extract potential local features in the three-dimensional tensor data from the three-dimensional tensor data through a CNN module to obtain spatial features, and combine the spatial features with context features to obtain comprehensive features; extract topological attribute features in the three-dimensional tensor data from the three-dimensional tensor data through a GNN module;

[0041] The quality assessment module is used to obtain the spatial prediction value of the cultivated land to be evaluated from the comprehensive characteristics; predict the topological attribute characteristics to obtain the topological prediction value of the cultivated land to be evaluated; and fuse the spatial prediction value and the topological prediction value to obtain the evaluation result of the quality of the cultivated land to be evaluated.

[0042] In a preferred example of the present application, it can be further set as follows:

[0043] On the third aspect, the present application is realized through the following technical solutions:

[0044] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of any one of the above-mentioned methods for evaluating the quality of cultivated land when executing the computer program.

[0045] In a fourth aspect, the present application provides a computer-readable storage medium.

[0046] This application is achieved through the following technical solutions:

[0047] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of any one of the above-mentioned methods for evaluating the quality of cultivated land are implemented.

[0048] In summary, compared with the prior art, the technical solution provided in the embodiment of the present application has at least the following beneficial effects:

[0049] This application uses a deep learning model SAFE based on spatial perception. SAFE uses a convolutional neural network CNN to integrate and extract the influence paradigm of spatial local features in land grid data. CNN effectively encodes the influence fusion paradigm of neighboring areas in each feature dimension, and uses a graph neural network GNN to mine the topological attribute interaction effects between related nodes, capturing the mutual influence relationship between each land grid attribute in the topological structure of the land grid data; in addition, the classification performance of the backbone network under unbalanced data conditions is enhanced through contrastive learning in the backbone network, so that it can minimize the distance between similar data and maximize the distance between different categories to optimize feature embedding, making the land quality assessment results more accurate. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A schematic diagram of a flow chart of a method for evaluating farmland quality provided in one embodiment of the present application;

[0051] Figure 2 A schematic diagram of the structure of a device for evaluating the quality of cultivated land provided in one embodiment of the present application;

[0052] Description of reference numerals:

[0053] Data acquisition module-101, feature extraction module-201, quality assessment module-301. DETAILED DESCRIPTION

[0054] This specific embodiment is merely an explanation of the present application and is not a limitation of the present application. After reading this specification, those skilled in the art may make modifications to the present embodiment without any creative contribution as needed, but such modifications are protected by the patent law as long as they are within the scope of the claims of the present application.

[0055] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0056] In addition, the term "and / or" in this application is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this application, unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.

[0057] In this application, the terms "first", "second", etc. are used to distinguish identical or similar items with substantially the same effects and functions. It should be understood that there is no logical or temporal dependency between "first", "second", and "nth", nor is there any limitation on quantity and execution order.

[0058] The method for evaluating the quality of cultivated land in the present application is based on a deep learning model SAFE (Spatial Awareness Framework for Evaluating Farmland), which is a two-classification model, wherein the deep learning model SAFE includes a backbone network, a CNN module, and a GNN module. The backbone network is a TabTransformer model, which is used to use the mutual influence information of neighboring areas for auxiliary modeling; the convolutional neural network, i.e., the CNN module, is used to extract the potential spatial local feature fusion paradigm of the grid data used to evaluate the quality of cultivated land; the graph neural network, i.e., the GNN module, is used to extract the topological relationship and attribute interaction information in the grid data used to evaluate the quality of cultivated land. At the same time, in order to enhance the processing ability of the deep learning model SAFE for unbalanced data, self-supervised regularization technology is introduced into the backbone network, such as contrastive learning to enhance the discrimination of the backbone network for feature embedding. The deep learning model SAFE can predict whether a land grid is high-quality cultivated land or whether it has the potential to become high-quality cultivated land.

[0059] The following is a further detailed description of the embodiments of the present application in conjunction with the accompanying drawings. Figure 1 As shown, a method for evaluating the quality of cultivated land proposed in this application includes:

[0060] S10: Obtain the land grid data of the cultivated land to be evaluated, input the land grid data into the backbone network, and the backbone network performs comparative learning on the features of the land grid data to obtain the context features of the land grid data.

[0061] Specifically, the cultivated land to be evaluated is divided into several grids of the same size. The land grid data is a tabular data feature matrix, represented by X∈R N×D , where N represents the number of land grids and D is the attribute / factor dimension of each grid.

[0062] Each land grid has complex influencing factor attributes, which can be roughly divided into three categories from the attribute category: physical conditions, environmental impacts, and land types. For example, physical conditions include but are not limited to factors such as soil thickness and soil texture, environmental impacts include but are not limited to altitude and slope, and land types include but are not limited to areas that belong to construction land or areas that do not belong to construction land. From the value of the attribute, it can be divided into two categories: continuous numerical attributes and discrete numerical attributes. For example, soil thickness, altitude, and slope are continuous numerical attributes, while soil texture and soil type are discrete numerical attributes.

[0063] The backbone network is designed to capture the potential features of each land network and analyze the spatial context information therein. The backbone network is flexible and can adopt various architectures, such as MLP framework, DeepFM framework and Transformer framework. Preferably, in this application, in order to convert these sparse attribute vectors into dense potential features, TabTransformer is used as the backbone structure of the deep learning model SAFE. TabTransformer is good at utilizing the contextual features of classification attributes.

[0064] S20: Associating the land grid data with geographic attributes to convert the land grid data into three-dimensional tensor data.

[0065] S30: Input the three-dimensional tensor data into the CNN module, extract the potential local features in the three-dimensional tensor data, and obtain the spatial features.

[0066] S40: Combine the spatial features with the context features to obtain comprehensive features, input the comprehensive features into the fully connected layer, and obtain the spatial prediction value of the cultivated land to be evaluated.

[0067] S50: Input the three-dimensional tensor data into the GNN module, extract the topological attribute features in the three-dimensional tensor data, predict the topological attribute features, and obtain the topological prediction value of the cultivated land to be evaluated.

[0068] S60: The spatial prediction value and the topological prediction value are integrated to obtain an evaluation result of the quality of the cultivated land to be evaluated.

[0069] Among them, the evaluation result is the predicted probability value of high-quality arable land or the potential to become high-quality arable land.

[0070] Through a deep learning model SAFE based on spatial perception, SAFE uses convolutional neural network CNN to integrate and extract the influence paradigm of spatial local features in land grid data. CNN effectively encodes the influence fusion paradigm of neighboring areas within each feature dimension, and uses graph neural network GNN to mine the interaction effects of topological attributes between related nodes, capturing the mutual influence relationship between each land grid attribute in the topological structure of land grid data; in addition, the classification performance of the backbone network under unbalanced data conditions is enhanced through contrastive learning in the backbone network, so that it can minimize the distance between similar data and maximize the distance between different categories to optimize feature embedding, making the land quality assessment results more accurate.

[0071] Specifically, the evaluation results It is calculated by combining the spatial prediction value of the spatial attribute and the topological prediction value of the topological attribute. The formula is expressed as:

[0072] in Expressed as spatial prediction value, Expressed as the topological prediction value, It is expressed as the evaluation result, which represents the probability value that the cultivated land to be evaluated belongs to high-quality land or has the potential to become high-quality land.

[0073] It should be noted that, in view of the challenges brought by unbalanced data sets, this application uses balanced softmax loss to train the classifier. In the model of this application, let n l represents the number of samples in category l∈{0,1}, y i Represents the land grid x i The true category, then the loss function It can be expressed as:

[0074]

[0075] The backbone network TabTransformer is particularly good at utilizing the contextual features of categorical attributes, where the self-attention mechanism is represented as a function This enables it to capture the interaction patterns between influencing factors, thereby enhancing the understanding of the internal dependencies of the input land grid data:

[0076]

[0077] Among them, f MLP (·) represents the multi-layer perceptron function, and the vector and Represent the continuous numerical attributes and categorical attributes of each instance respectively, and the contextual embedding of the categorical attributes is given by Get, where f Emb (·) is an embedded function. The operator Represents a join operation.

[0078] The TabTransformer function with parameter θ1 is defined as:

[0079]

[0080] Where D1 represents the dimension of feature representation. Therefore, the dense latent feature matrix Calculated as

[0081] The land grid data is input into the backbone network, and the backbone network performs comparative learning on the features of the land grid data. First, the binary data is constructed, and the binary data is used to improve the comparative learning ability of the backbone network. For a data set containing N samples {x i ,y i} i＝1...N , forming a training data containing N tuples where i = 1...N, x a is a randomly selected positive sample as an anchor point, Represents x i Does it belong to x a The same category. Among them, the positive sample indicates that the land is or has the potential to become high-quality farmland, and the negative sample indicates that the land is not or has no potential to become high-quality farmland.

[0082] In some embodiments, the distance between the input sample and the anchor point is measured based on cosine similarity in the backbone network, and the loss function of contrastive learning is set to:

[0083]

[0084] Among them, when the input sample and the anchor sample belong to the same category (i.e. When , the input sample and the anchor point sample are close by minimizing the loss function:

[0085]

[0086] When the input sample and the anchor sample belong to different categories (i.e. When , the input sample and the anchor sample are separated by minimizing the loss function:

[0087]

[0088] in, represents the loss function, represents the cosine similarity, represents the contrastive learning loss function value of the i-th sample, represents the contrastive learning loss function value of the sample with the i-th predicted value of 1, represents the contrastive learning loss function value of the sample whose prediction value is 0. represents the feature vector of the i-th input sample, represents the feature vector of the anchor point sample, Represents the predicted value of the i-th input sample, ∈ represents the threshold that controls whether to perform comparative learning loss calculation on input samples and anchor samples of different categories.

[0089] Contrastive learning has shown remarkable effectiveness in solving the problem of data imbalance. By integrating this method into the feature extraction process of the backbone network, the feature mapping function can be optimized. This process makes similar data points cluster together and dissimilar data points disperse, thereby improving the performance of the classifier.

[0090] In some embodiments, land grid data is associated with geographic attributes to convert the land grid data into three-dimensional tensor data, including: obtaining the longitude set and latitude set of all grids from the geohash attribute of the land grid data; sorting the longitudes in the longitude set and the latitudes in the latitude set, and adding a longitude index and a latitude index to each grid; based on the longitude index and the latitude index, mapping the attribute vector of the land grid data to the corresponding grid position to obtain three-dimensional tensor data.

[0091] Specifically, the table format of the original land grid data is not suitable for the training and feature extraction of convolutional neural networks (CNNs) and GNNs. To solve this problem, this application proposes a data conversion strategy that represents the geographic longitude and latitude coordinates of each grid according to its geohash attribute, thereby converting the land grid data into grid data. To illustrate the above process, the data set is represented as: X = {x1, x2, ..., x i , ..., x N},

[0092] Among them, the longitude set is expressed as:

[0093]

[0094] The latitude set is represented as:

[0095]

[0096] The longitude set and the latitude set correspond to the geographic coordinates required to convert the land grid data into a grid format suitable for CNN and GNN processing. After sorting the longitude set and the latitude set independently, the land grid x i Exported longitude and latitude grid index in Here, W and H represent the total number of grid indices in the longitude dimension and the total number of grid indices in the latitude dimension, respectively. Using these indices, the original land grid data is reconstructed into three-dimensional tensor data Map each attribute vector to its corresponding grid position. This mapping relationship can reconstruct the two-dimensional matrix corresponding to each feature dimension.

[0097] Considering the use of the entire three-dimensional tensor data To predict the grid classification of the cultivated land to be evaluated, the amount of calculation is relatively large. At the same time, the influence between geographically distant areas is weakened. In this application, before the three-dimensional tensor data is input into the CNN module, a hyperparameter C is introduced to define the relevant spatial range. Specifically: the three-dimensional data tensor is divided into each grid x i The center of the land grid is taken as the center and divided into local matrix slices of size C×C, where the clipping parameter C is set to 30-35. After the above processing, the land grid data is converted into tensor data Preferably, in the present application, the trimming parameter C is set to 32.

[0098] Using the converted three-dimensional tensor data, the convolutional neural network (CNN) is used to extract and integrate local features through convolution operations. The convolutional neural network (CNN) has higher analysis efficiency when performing fine-grained local analysis and can better capture the spatial hierarchical structure in the data. The extraction process of the backbone network is expressed as: D2 is the dimension of the extracted spatial features.

[0099] In some embodiments, the comprehensive features are input into the fully connected layer to obtain the spatial prediction value of the cultivated land to be evaluated, including:

[0100]

[0101] in, Indicates that the CNN module extracts spatial features, D2 is the dimension of the extracted spatial features, represents the fully connected layer, represents three-dimensional tensor data, Represents the context features extracted by the backbone network.

[0102] Considering that the distribution of high-quality farmland is usually uneven and irregular, graph neural networks (GNNs) are suitable for capturing potential attribute interactions between plots. By leveraging the inherent topological structure of the graph, graph neural networks (GNNs) are able to efficiently learn node attributes through edge updates, providing substantial benefits in attribute interaction and extraction. Compared with traditional grid- and sequence-based structures, graph neural networks (GNNs) have better performance because they can integrate more complex and flexible structural information.

[0103] In some embodiments, three-dimensional tensor data is input into a GNN module, and topological attribute features in the three-dimensional tensor data are extracted, including: inputting the three-dimensional tensor data into the GNN module, performing aggregation operations on adjacent node features based on an aggregation function to obtain aggregated features; performing nonlinear transformation on the aggregated features based on an update function to generate an updated node representation, extracting attribute interaction information in the updated node representation, and obtaining topological attribute features.

[0104] Specifically, the converted three-dimensional data (local matrix slicing), and split the data into N1 blocks, h and w represent the height and width of the block, respectively, and each block is regarded as a block with feature representation v i ∈R D×hw Nodes. The collection of unordered nodes can be further represented as: For edge initialization, we initialize each node v i Connect to its eight nearest neighbor nodes, forming a network from neighbor nodes to node v i Directed edges of In building the graph Finally, the graph convolutional layer facilitates information exchange between nodes.

[0105] Among them, the topological attribute features are predicted to obtain the topological prediction value of the cultivated land to be evaluated, including: using a multi-layer perceptron to predict the topological attribute features and using the ReLU function as the activation function:

[0106]

[0107] in, Indicates v i The set of adjacent nodes. It should be noted that It will be dynamically adjusted in each iteration (each convolutional layer), and the sparse k-nearest neighbor (k-NN) function is used to measure the distance during the iteration. After completing the iterative adjustment process, the node v i The topological property characteristics and MLP function Topological prediction of u irepresents the vertex feature aggregation vector under the maximum pooling operation, represents the update operation of the graph convolutional layer, Represents the activation vector after the graph convolution update layer and activation function.

[0108] In the task of evaluating farmland quality, the overall objective loss function is minimized to complete the optimization of the entire model:

[0109] Wherein γ is a weight coefficient, which can be set to 1 in this application.

[0110] Another embodiment of the present application also provides a device for evaluating the quality of cultivated land, which is used to perform the above-mentioned evaluation method, such as Figure 2 As shown, the evaluation device includes:

[0111] The data acquisition module 101 is used to acquire the land grid data of the cultivated land to be evaluated; and to associate the land grid data with geographical attributes so as to convert the land grid data into three-dimensional tensor data;

[0112] The feature extraction module 201 is used to compare and learn the features of the land grid data through the backbone network to obtain the context features of the land grid data; extract potential local features in the three-dimensional tensor data from the three-dimensional tensor data through the CNN module to obtain spatial features, and combine the spatial features with the context features to obtain comprehensive features; extract topological attribute features in the three-dimensional tensor data from the three-dimensional tensor data through the GNN module;

[0113] The quality assessment module 301 is used to obtain the spatial prediction value of the cultivated land to be evaluated from the comprehensive characteristics; predict the topological attribute characteristics to obtain the topological prediction value of the cultivated land to be evaluated; and fuse the spatial prediction value and the topological prediction value to obtain the evaluation result of the quality of the cultivated land to be evaluated.

[0114] The specific definition of the device for evaluating the quality of cultivated land provided in this embodiment can be found in the embodiment of the method for evaluating the quality of cultivated land in the above text, and will not be repeated here. Each module in the above-mentioned device for evaluating the quality of cultivated land can be implemented in whole or in part by software, hardware, and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0115] An embodiment of the present application provides a computer device, which may include a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the processor executes the steps of the method for evaluating the quality of cultivated land as in any of the above embodiments.

[0116] The working process, working details and technical effects of the computer equipment provided in this embodiment can be found in the embodiment of the method for evaluating the quality of cultivated land mentioned above, and will not be described in detail here.

[0117] The embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method for evaluating the quality of cultivated land in any of the above embodiments are implemented. The computer-readable storage medium refers to a carrier for storing data, which may include but is not limited to a floppy disk, an optical disk, a hard disk, a flash memory, a USB flash drive and / or a memory stick, etc., and the computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0118] The working process, working details and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiment of the method for evaluating arable land quality described above, and will not be described in detail here.

[0119] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0120] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0121] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the system described in the present application is divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A method for evaluating the quality of cultivated land, characterized in that: Based on the deep learning model SAFE, the deep learning model SAFE includes a backbone network, a CNN module and a GNN module, and the method includes: Obtaining land grid data of the cultivated land to be evaluated, inputting the land grid data into the backbone network, and the backbone network performing comparative learning on the features of the land grid data to obtain context features of the land grid data; Associating the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data; Inputting the three-dimensional tensor data into a CNN module, extracting potential local features in the three-dimensional tensor data, and obtaining spatial features; Combining the spatial features with the context features to obtain comprehensive features, and inputting the comprehensive features into a fully connected layer to obtain a spatial prediction value of the cultivated land to be evaluated; Inputting the three-dimensional tensor data into the GNN module, extracting topological attribute features in the three-dimensional tensor data, predicting the topological attribute features, and obtaining a topological prediction value of the cultivated land to be evaluated; The spatial prediction value and the topological prediction value are integrated to obtain an evaluation result of the quality of the cultivated land to be evaluated.

2. The method for evaluating cultivated land quality according to claim 1, characterized in that: Before the three-dimensional tensor data is input into the CNN module, the following steps are also included: The three-dimensional data tensor is divided into each grid x i The center of is taken as the center and divided into local matrix slices of size C×C, where the cropping parameter c is set to 30~35.

3. The method for evaluating cultivated land quality according to claim 1, characterized in that: The land grid data is input into the backbone network, and the backbone network performs comparative learning on the features of the land grid data, including: In the backbone network, the distance between the input sample and the anchor point sample is measured based on cosine similarity, and the loss function of contrastive learning is set as: Among them, when the input sample and the anchor sample belong to the same category, the loss function is minimized to make the input sample and the anchor sample close: When the input sample and the anchor sample belong to different categories, the loss function is minimized to separate the input sample and the anchor sample: in, represents the loss function, represents the cosine similarity, represents the contrastive learning loss function value of the i-th sample, represents the contrastive learning loss function value of the sample with the i-th predicted value of 1, represents the contrastive learning loss function value of the sample whose prediction value is 0. represents the feature vector of the i-th input sample, represents the feature vector of the anchor point sample, Represents the predicted value of the i-th input sample, ∈ represents the threshold that controls whether to perform comparative learning loss calculation on input samples and anchor samples of different categories.

4. The method for evaluating cultivated land quality according to claim 1, characterized in that: Associating the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data includes: Get the longitude and latitude sets of all grids from the geohash attribute of the land grid data; Sort the longitudes in the longitude collection and the latitudes in the latitude collection, and add the longitude index and latitude index to each grid; Based on the longitude index and latitude index, the attribute vector of the land grid data is mapped to the corresponding grid position to obtain three-dimensional tensor data.

5. The method for evaluating cultivated land quality according to claim 1, characterized in that: The comprehensive features are input into the fully connected layer to obtain the spatial prediction value of the cultivated land to be evaluated, including: in, represents the spatial prediction value, Indicates that the CNN module extracts spatial features, D2 is the dimension of the extracted spatial features, represents the fully connected layer, represents three-dimensional tensor data, represents the context features extracted by the backbone network, X i Represents the land grid data of the i-th grid.

6. The method for evaluating cultivated land quality according to claim 1, characterized in that: The three-dimensional tensor data is input into the GNN module, and the topological attribute features in the three-dimensional tensor data are extracted, including: Inputting the three-dimensional tensor data into the GNN module, performing an aggregation operation on adjacent node features based on an aggregation function to obtain an aggregated feature; Based on the update function, the aggregated features are nonlinearly transformed to generate updated node representations, and attribute interaction information in the updated node representations is extracted to obtain topological attribute features.

7. The method for evaluating cultivated land quality according to claim 6, characterized in that: Predict the topological attribute characteristics to obtain the topological prediction value of the cultivated land to be evaluated, including: A multilayer perceptron is used to predict the topological attribute features, and a ReLU function is used as an activation function.

8. A device for evaluating the quality of cultivated land, characterized in that: Used to perform the evaluation method according to any one of claims 1 to 7, comprising: A data acquisition module, used to acquire land grid data of the cultivated land to be evaluated; and used to associate the land grid data with geographic attributes so as to convert the land grid data into three-dimensional tensor data; A feature extraction module is used to compare and learn the features of the land grid data through a backbone network to obtain context features of the land grid data; extract potential local features in the three-dimensional tensor data from the three-dimensional tensor data through a CNN module to obtain spatial features, and combine the spatial features with context features to obtain comprehensive features; extract topological attribute features in the three-dimensional tensor data from the three-dimensional tensor data through a GNN module; The quality assessment module is used to obtain the spatial prediction value of the cultivated land to be evaluated from the comprehensive characteristics; predict the topological attribute characteristics to obtain the topological prediction value of the cultivated land to be evaluated; and fuse the spatial prediction value and the topological prediction value to obtain the evaluation result of the quality of the cultivated land to be evaluated.

9. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images