Visual display method and device for land assessment data, equipment and medium

Abstract

The invention discloses a visual display method, device and equipment for land assessment data and a medium, and the method comprises the steps: constructing a multi-modal feature database of land resources, the multi-modal feature database being a database constructed by using land value monitoring indexes and land heterogeneous data; training a preset model by using the multi-modal feature database to obtain a land evaluation model, and calling the land evaluation model to perform evaluation processing to obtain land evaluation data; and constructing a land value thermodynamic diagram by adopting the land evaluation data, and visually displaying the land value thermodynamic diagram. According to the method, the model is trained and evaluated through the database, the evaluation precision can be improved, meanwhile, visualization is performed through the thermodynamic diagram, various data is prevented from being displayed at the same time, and then the display effect can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing and visualization, and particularly relates to a land assessment data visualization display method, device, equipment and medium. BACKGROUND

[0002] With the acceleration of urbanization, rational utilization and dynamic monitoring of land resources have become one of the core problems of natural resource management. Therefore, before land resource development and utilization, relevant data of land resources need to be determined, land value assessment is carried out based on various land resource data, and finally land resource management, development and utilization are carried out according to the assessment results.

[0003] In order to facilitate users to comprehensively view relevant data of various land resources and assessment results, the commonly used method at present is to manually conduct market investigation and value assessment, and then to summarize various data on a terminal, so as to display various investigation data and assessment results on the terminal in a static manner for user reference.

[0004] However, the commonly used method has the following technical problems: manual investigation and assessment have human bias, the assessment accuracy is low, and various data types are required. Through static data display, the data is numerous and disordered, the display effect is poor, and the display of incorrect data also affects subsequent management, development and utilization. SUMMARY

[0005] The present application provides a land assessment data visualization display method, device, equipment and medium, which can solve the technical problems of incorrect data display, numerous data display and poor display effect in the prior art.

[0006] The first aspect of the embodiment of the present application provides a land assessment data visualization display method, which comprises:

[0007] A multi-modal feature database of land resources is constructed, and the multi-modal feature database is a database constructed by using land value monitoring indicators and land heterogeneous data;

[0008] A land assessment model is obtained by training a preset model by using the multi-modal feature database, and the land assessment model is called for assessment processing to obtain land assessment data;

[0009] A land value heat map is constructed by using the land assessment data, and the land value heat map is visualized and displayed.

[0010] In combination with the first aspect, in an implementation mode, the multi-modal feature database of land resources is constructed, comprising:

[0011] Determine land value monitoring indicators and land heterogeneous data, and construct an initial database using the land value monitoring indicators and the land heterogeneous data;

[0012] Data distillation is performed on the initial database to obtain a multi-modal feature database.

[0013] In combination with the first aspect, in an implementation manner, the determining land value monitoring indicators and land heterogeneous data, and constructing an initial database using the land value monitoring indicators and the land heterogeneous data, comprises:

[0014] Determine user-predefined land value monitoring indicators, and obtain land heterogeneous data;

[0015] Preprocess the land heterogeneous data to obtain heterogeneous processing data;

[0016] Perform coordinate conversion, multi-scale decomposition and merging on the heterogeneous processing data in sequence to obtain static feature data and dynamic feature data;

[0017] Construct an initial database using the land value monitoring indicators, the static feature data and the dynamic feature data.

[0018] In combination with the first aspect, in an implementation manner, the data distillation performed on the initial database to obtain a multi-modal feature database, comprises:

[0019] Calculate an importance score value of each data of the initial database based on a random forest;

[0020] According to a dynamic distillation strategy, perform data distillation on the initial database according to the importance score value to obtain a multi-modal feature database.

[0021] In combination with the first aspect, in an implementation manner, the preset model is a spatio-temporal convolutional neural network architecture;

[0022] The spatio-temporal convolutional neural network architecture comprises a spatial stream network and a temporal stream network, and dynamically fuses output features of the spatial stream network and the temporal stream network through a cross-attention mechanism;

[0023] Training of the preset model adopts multi-task learning and regularization constraint.

[0024] In combination with the first aspect, in an implementation manner, after the step of training a preset model using the multi-modal feature database to obtain a land assessment model, the method further comprises:

[0025] The parameter of the land evaluation model is updated based on an online learning mechanism, and a standard deviation of a prediction result of the land evaluation model is calculated based on a Monte Carlo technique, so as to generate a land value prediction confidence interval by using the standard deviation.

[0026] In combination with the first aspect, in an implementation manner, the land value heat map is constructed by using the land evaluation data, and the land value heat map is visually displayed, including:

[0027] The non-spatial attribute data of the land evaluation data is converted into a geographic coordinate system by inverse geocoding, and the land evaluation data is normalized to obtain normalized data;

[0028] The continuous spatial grid surface is generated by using the normalized data, and the color gradient of the continuous spatial grid surface is adjusted by using a color band tool to obtain the land value heat map;

[0029] After the format conversion of the land value heat map, the land value heat map is visually displayed.

[0030] The second aspect of the embodiment of the present application provides a land evaluation data visual display device, the device includes:

[0031] The construction module is configured to construct a multi-modal feature database of land resources, and the multi-modal feature database is a database constructed by using land value monitoring indicators and land heterogeneous data;

[0032] The evaluation module is configured to train a preset model by using the multi-modal feature database to obtain a land evaluation model, and to call the land evaluation model to perform evaluation processing to obtain land evaluation data;

[0033] The display module is configured to construct a land value heat map by using the land evaluation data, and to visually display the land value heat map.

[0034] Compared with the prior art, the land evaluation data visual display method, device, equipment and medium provided by the embodiment of the present application have the beneficial effects that the multi-modal feature database of land resources can be constructed, the preset model is trained by using the multi-modal feature database to obtain the land evaluation model, the land evaluation model is called to perform evaluation processing to obtain the land evaluation data, the land value heat map is constructed by using the land evaluation data, and the land value heat map is visually displayed, the evaluation accuracy can be improved by training the model by using the database and performing evaluation, and the display effect can be improved by visualizing the heat map and avoiding displaying too much data at the same time. BRIEF DESCRIPTION OF DRAWINGS

[0035] Figure 1is a flowchart of a land evaluation data visual display method provided by an embodiment of the present application;

[0036] Figure 2 is an operation flowchart of a land evaluation data visual display method provided by an embodiment of the present application;

[0037] Figure 3 is a structural schematic diagram of a land evaluation data visual display device provided by an embodiment of the present application;

[0038] Figure 4 is a structural schematic diagram of a land evaluation data visual display system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0039] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in 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 in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0040] With the acceleration of urbanization, rational utilization and dynamic monitoring of land resources have become one of the core problems of natural resource management. Therefore, before land resource development and utilization, relevant data of land resources need to be determined, land value assessment is performed based on various land resource data, and finally land resource management, development and utilization are performed according to the assessment results.

[0041] In order to facilitate users to comprehensively view relevant data of various land resources and assessment results, the commonly used method is to perform market investigation and value assessment by artificial, and then to summarize various data on a terminal, and to display various investigation data and assessment results by the terminal in a static manner for users to refer.

[0042] However, the commonly used method has the following technical problems: artificial investigation and assessment have human bias, the assessment accuracy is low, and various data types are required. The data is displayed in a static manner, the data is numerous and disordered, the display effect is poor, and the display of incorrect data also affects subsequent management, development and utilization.

[0043] In order to solve the above problems, the land evaluation data visual display method, device, equipment and medium provided by the embodiments of the present application will be described and explained in detail as follows.

[0044] To solve the technical problems of incorrect data displayed by the prior art, numerous data displayed and poor display effect, refer to Figure 1Fig. 1 shows a flowchart of a method for visualizing land assessment data according to an embodiment of the present application.

[0045] In an embodiment, the method is applicable to a computer device, which can be a natural resource land price dynamic monitoring system.

[0046] The method for visualizing land assessment data can include, for example:

[0047] S11, constructing a multi-modal feature database of land resources, the multi-modal feature database being constructed using land value monitoring indicators and land heterogeneous data.

[0048] The multi-modal feature database can include multiple data, and different land value monitoring indicators can be set for different data. By constructing the multi-modal feature database of land resources, the value assessment can be performed using each data in the database, avoiding the deviation of manual operation, and thus improving the accuracy of the assessment, so that accurate data can be displayed.

[0049] The method for constructing a multi-modal feature database of land resources can include, for example, the following sub-steps:

[0050] S111, determining land value monitoring indicators and land heterogeneous data, and constructing an initial database using the land value monitoring indicators and the land heterogeneous data.

[0051] S112, data distillation is performed on the initial database to obtain a multi-modal feature database.

[0052] In an embodiment, a preset land value monitoring indicator can be determined, and land heterogeneous data can be collected. Then, a plurality of land heterogeneous data and a plurality of land value monitoring indicators can be combined to obtain an initial database.

[0053] Since the initial database contains multiple data, the values of each data can be biased or missing. In order to improve the accuracy of the data, data distillation can be performed on the initial database. Data distillation can be a data filtering process, which can filter valuable data and reduce the use of incorrect or missing data for evaluation processing, so as to improve the accuracy of the evaluation.

[0054] In an optional embodiment, the method for determining the land heterogeneous data corresponding to the preset land value monitoring indicator and constructing an initial database using the land heterogeneous data can include the following sub-steps:

[0055] S1111, determining a land value monitoring indicator defined by a user in advance, and obtaining land heterogeneous data.

[0056] S1112, pre-process the land heterogeneous data to obtain heterogeneous processing data.

[0057] S1113, sequentially perform coordinate conversion, multi-scale decomposition and merging on the heterogeneous processing data to obtain static feature data and dynamic feature data.

[0058] S1114, construct an initial database using the land value monitoring index, the static feature data and the dynamic feature data.

[0059] In an embodiment, the user-predefined land value monitoring index is determined. Optionally, the land value monitoring index can include the following indexes:

[0060] (1) Spatial feature index: land use efficiency (volume rate, building density), three-dimensional spatial attribute (building boundary integrity, height distribution), plot geometry, land development state;

[0061] (2) Time series dynamic index: land change frequency, value fluctuation characteristics, development state persistence, idle period;

[0062] (3) Social and economic index: population heat distribution intensity, enterprise registration and operation state, project investment progress, land market price fluctuation correlation;

[0063] (4) Remote sensing derived index: multi-temporal spectral characteristics, land cover change index, night light intensity gradient.

[0064] In an embodiment, multi-source land heterogeneous data can be collected and pre-processed to obtain heterogeneous processing data. In specific operation, the following multi-modal land heterogeneous data can be collected, and the multi-modal land heterogeneous data can be data cleaned, corrected and standardized, unified and normalized data format, converted different unit data to standardized numerical range, to obtain heterogeneous processing data. In an embodiment, the land heterogeneous data can include the following data:

[0065] (1) Land ownership data: obtain land supply time, planning purpose, completion deadline and other structured attributes from real estate registration system;

[0066] (2) Remote sensing image data: integrate sub-meter high-resolution remote sensing image, multi-spectral time series data, extract plot texture, spectrum and spatial distribution characteristics;

[0067] (3) Three-dimensional building data: analyze building boundary vector and height information based on RSBuilding model to generate plot-level building form characteristics;

[0068] (4) Socio-economic data: access to mobile signaling population heat data, land market price fluctuation sequence, enterprise business registration information, construction project approval database, and quantify plot development activity;

[0069] (5) Basic price data: extract plot price data from land market transaction records and integrate with the aforementioned data to estimate value and make judgments.

[0070] Standardize all multi-modal data, spatially align data with spatial attributes, and sequence correct asynchronous time series data.

[0071] Then, the heterogeneous processing data is sequentially subjected to coordinate conversion, multi-scale decomposition and merging to obtain several land feature data.

[0072] Among them, coordinate conversion can be achieved by spatial registration algorithm and projection conversion technology to unify all multi-source data with spatial attributes to the same geographic coordinate system. SURF algorithm is used to extract remote sensing image and vector boundary for feature point matching. Double cubic convolution interpolation is used for low resolution data (such as population heat grid) to ensure spatial alignment with high resolution image. If non-spatial attribute data (such as the "operation status" field in enterprise registration information) does not contain spatial coordinates itself, it needs to be logically associated indirectly aligned through a unique spatial identifier (such as plot ID), without the need for direct coordinate conversion; for heterogeneous processing data with spatial attributes, unify to the same geographic coordinate system, and indirectly align non-spatial attribute data through logical association.

[0073] Multi-scale decomposition processing can be to use Daubechies wavelet basis function to perform multi-scale decomposition on asynchronous time series data (remote sensing image data, socio-economic data, dynamic monitoring data), extract high frequency (details) and low frequency (trend) components, and retain the local mutation characteristics of time series data (such as price jumps caused by policy regulation). The interpolation reconstruction formula is as follows:

[0074]

[0075] Among them, φ and ψ are scale function and wavelet function respectively, and c and d are decomposition coefficients. Static data (land ownership data, three-dimensional building data) is aligned with dynamic data through time slicing (such as associated with the same time label), and does not need to be decomposed.

[0076] It should be noted that heterogeneous processing data can include remote sensing image data, socio-economic data, dynamic monitoring data, only asynchronous time series data needs to be subjected to multi-scale decomposition, and static data does not need to be decomposed.

[0077] The merging processing can be to merge the plot attributes (area, use), building three-dimensional parameters (volume rate, shape index) into a static feature vector, and to perform exponential weighting on the historical price sequence based on a time decay factor, and the formula is as follows:

[0078] w t =e -λ(T-t) ,λ∈[0.1,0.5];

[0079] Wherein, T is the current time, t is the historical time point, and λ is the decay coefficient.

[0080] Finally, an initial database serving as a land dynamic monitoring sample can be generated based on the above data, and the merged data can include the following data:

[0081] (1) Static feature data: store plot inherent attributes (area, planning use), building three-dimensional parameters (average height, contour complexity), and property status (land supply period, default identification);

[0082] (2) Dynamic feature data: record time series change indicators, including development progress time series (building area growth curve based on RSBuilding), population heat fluctuation sequence, and land market price index;

[0083] Finally, different labels can be added to each land feature data, and the labels can be state classification (development stage, such as planning / under construction / completed / overdue and not completed / partially completed / complete idle, etc.) and value level (potential classification, such as high / medium / low potential) of the plot, supporting subsequent supervised model training.

[0084] In an optional embodiment, the data distillation of the initial database to obtain a multi-modal feature database can include the following sub-steps:

[0085] S1121, calculating the importance score value of each data of the initial database based on a random forest.

[0086] S1122, performing data distillation on the initial database according to the importance score value according to a dynamic distillation strategy to obtain a multi-modal feature database.

[0087] In order to screen important data, the importance of feature data can be evaluated, specifically, the correlation of multi-modal feature data with land value target variables can be quantified, high contribution degree features are screened, and redundant features are removed. Combined with random forest and mutual information algorithm, the feature importance is evaluated from the aspects of model dependency and statistical independence.

[0088] In an operation mode, the importance score of each data can be calculated based on a random forest algorithm. Specifically, the feature importance can be measured by the cumulative reduction of Gini impurity. For feature data X j , the importance score I RF (X j ) is defined as the average of the reduction of Gini impurity when the feature splits in all decision trees:

[0089]

[0090] where S t (X j ) is the set of split nodes based on feature data X j in the t-th tree, ΔGini(s) is the reduction of Gini impurity of node s, and N trees is the total number of decision trees.

[0091] Then, the mutual information can be calculated to measure the nonlinear statistical dependence between feature data X j and target variable Y, and the calculation formula is:

[0092]

[0093] The joint probability distribution p(x j , y) and the marginal distribution p(x j ), p(y) can be estimated by kernel density estimation or histogram method.

[0094] After normalization and weighted fusion of the scores of the two methods, the importance score I total (X j ) is obtained, which can be specifically shown as follows:

[0095] I total (X j ) = aI RF (X j ) + (1-a)I MI (X j ; Y);

[0096] where a ∈ [0, 1] is a weight coefficient (empirical value is 0.6). The features with a score higher than a preset threshold (such as Top 30%) are screened to form a feature subset after preliminary distillation

[0097] Then, the initial database can be distilled according to the importance score value according to a dynamic distillation strategy to obtain a multi-modal feature database.

[0098] The dynamic distillation strategy can be a dynamic characteristic of time series data and spatial heterogeneity, optimize the selection process of feature data, and avoid the loss of regional key features caused by global screening.

[0099] In an operation mode, the dynamic distillation strategy can be a sliding window time series dynamic distillation. For time series feature data (N is the number of features, and T is the time step), the importance score value is dynamically calculated by using a sliding window mechanism. The window length W and the sliding step S are defined, and the feature score is recalculated in the window [t-W+1, t] Dynamic adjustment of the reserved feature set:

[0100]

[0101] where θ threshold ∈(0, 1) is a dynamic threshold proportion coefficient (such as 0.7).

[0102] In yet another operation mode, the dynamic distillation strategy can be a geographically weighted regression spatial local distillation. For regions with significant spatial heterogeneity (such as urban core and suburbs), a geographically weighted regression model can be introduced to calculate the local feature importance. The GWR model expression is:

[0103]

[0104] where (u i , v i ) is the spatial coordinates of the ground i, β j (u i , v i ) is the spatial variable coefficient, reflecting the local influence of feature data X j on the target variable Y. The local feature importance is defined as the spatial average of the absolute value of the coefficient:

[0105]

[0106] Combining the global and local importance score values, a multi-modal feature database can be obtained. Specifically, the final reserved feature data set in the multi-modal feature database can be as follows:

[0107] F final = F dynamic ∪{X j |I GWR (X j )≥γ·max(I GWR )};

[0108] where γ ∈ (0, 1) is a local screening threshold.

[0109] Through data distillation, redundant information compression and key feature enhancement of the multi-modal feature database can be realized.

[0110] S12, training a preset model by using the multi-modal feature database to obtain a land assessment model, and calling the land assessment model to perform assessment processing to obtain land assessment data.

[0111] In the multi-modal feature database, the data of the multi-modal feature database can be used to train a preset model to obtain a land assessment model. After obtaining the land assessment model, the land assessment model can be called to perform assessment processing to obtain land assessment data.

[0112] In an optional embodiment, the preset model is a spatio-temporal convolutional neural network architecture.

[0113] The spatio-temporal convolutional neural network architecture includes a spatial stream network and a temporal stream network, and dynamically fuses the output features of the spatial stream network and the temporal stream network through a cross-attention mechanism.

[0114] The training of the preset model adopts multi-task learning and regularization constraint.

[0115] The spatial stream network can use a 3D convolutional layer (Kernel Size: 3x3x3) to process multi-temporal remote sensing images and three-dimensional building data, and capture spatial structure features (such as building distribution patterns and spatial gradients of land use efficiency) inside the land parcel.

[0116] The temporal stream network can model land price time series data and population heat change sequences based on dilated causal convolution and gated recurrent unit (GRU), and extract long-term dependencies and periodic fluctuation patterns.

[0117] The output features of the dual-stream network are dynamically fused through a cross-attention mechanism, and the formula is:

[0118]

[0119] Wherein, Q spatial , K temporal , V temporal are the query vector of the spatial stream feature, the key vector of the temporal stream feature, and the value vector, respectively, and d is the feature dimension.

[0120] In an embodiment, the multi-task learning and regularization constraint includes multi-objective joint optimization and sparsity regularization constraint. The multi-objective joint optimization can design a multi-task loss function, which fuses feature reconstruction error (mean square error), land value classification loss (cross-entropy), and spatial distribution consistency loss (block neighborhood similarity constraint based on KL divergence). The gradient weighting strategy is used to balance the optimization direction of different tasks to avoid single target overfitting.

[0121] Sparse regularization can impose L1-norm sparsity constraint on the encoder hidden layers, forcing the network to activate a small number of key neurons in the feature compression process, suppressing the interference of irrelevant noise features. The Monte Carlo Dropout technique is used to randomly mask part of the neuron connections, simulating a random feature missing scenario, and enhancing the model's robustness to incomplete data.

[0122] It should be noted that during the model training phase, strategy optimization can be performed, for example, a phased training strategy can be used for pre-training. Specifically, a layer-by-layer greedy unsupervised pre-training strategy can be used to initialize the encoder and decoder of the stacked autoencoder. Based on large-scale unlabeled feature data (such as historical remote sensing images, unannotated land parcel attributes), by minimizing the input feature reconstruction error, the network is forced to learn the potential distribution of land value related features, providing high-discriminability feature representation basis vectors for subsequent supervised training.

[0123] After pre-training is completed, a supervised signal (such as land value grade label, development state classification label) is introduced to jointly optimize the reconstruction loss of the autoencoder and the cross-entropy loss of the Softmax classifier in an end-to-end manner. Through the back propagation algorithm, the network weights are dynamically adjusted to force the latent features to meet the dual constraints of data compression fidelity and land value prediction accuracy, improving the feature discriminability.

[0124] During the training phase, a dynamic weight adjustment mechanism can be used for weight adjustment. For spatio-temporal attention weighting, a multi-head self-attention mechanism can be used to dynamically calculate the weight distribution of features in the spatial dimension (such as land parcel geographic location correlation) and the temporal dimension (such as development stage continuity). Through the Sigmoid function, an attention mask is generated to strengthen high-contribution features and suppress redundant features, allowing the model to adaptively focus on the driving factors of land value evolution.

[0125] Adversarial training enhancement can introduce a generative adversarial network to construct adversarial samples and add small perturbations (such as simulating remote sensing image noise, land price abnormal fluctuations) to the input features. Through adversarial training strategy, the model is forced to learn perturbation-invariant features, improving the generalization ability to data collection errors and market emergencies.

[0126] During the training process, feature optimization and discriminability enhancement can also be performed. Specifically, t-distributed Stochastic Neighbor Embedding (t-SNE) can be used to project the features before and after optimization into low-dimensional space, visually verifying the class separability of feature clusters, and guiding parameter tuning.

[0127] After training the model, in order to improve the evaluation accuracy of the model, in one of the embodiments, after the step of training the preset model by using the multi-modal feature database to obtain a land evaluation model, the method can further include the following sub-steps:

[0128] S21, updating parameters of the land evaluation model based on an online learning mechanism, and calculating a standard deviation of a prediction result of the land evaluation model based on a Monte Carlo technique to generate a land value prediction confidence interval using the standard deviation.

[0129] Specifically, the online learning mechanism can be used for parameter updating, and a sliding window (for example, Window Size: 12 months) can be used to real-time input the latest land transaction data and remote sensing images, and the model parameters can be dynamically updated through incremental learning to adapt to changes in land market policies and the impact of unexpected events.

[0130] Meanwhile, the uncertainty of model evaluation can be quantified, the standard deviation σ of the prediction result can be calculated based on the Monte Carlo Dropout technique, and the land value prediction confidence interval can be generated, and the formula is:

[0131]

[0132] Wherein, S is the number of Monte Carlo sampling, and z is the standard normal distribution quantile.

[0133] After training the model, the model can be called for land value evaluation, and a spatio-temporal convolutional neural network can be constructed to realize high-precision modeling and prediction of the dynamic evolution of land value, and the step 3 includes:

[0134] During prediction, the input data can be distilled and optimized to obtain a feature matrix (including three-dimensional attributes of the land block, social and economic indicators, and time series remote sensing features), and the above feature matrix is used as the model input, and the dimension is N x T x C (N is the number of land blocks, T is the time step, and C is the number of feature channels) input to the model for evaluation processing.

[0135] The definition of the dynamic prediction task of the model can be as follows: output the land price change rate ΔP and the value grade probability distribution P of the future Δt period (such as quarterly / annual) class , and the loss function adopts a weighted mean square error (WMSE) and a cross-entropy joint optimization:

[0136]

[0137] Wherein, λ is a task weight hyperparameter, which is determined by grid search optimization.

[0138] In addition, an adversarial sample generation network (GAN) is introduced to add a slight disturbance to the feature matrix of the input data, thereby improving the robustness of the model to noisy data and outliers.

[0139] S13, constructing a land value heat map using the land assessment data and visually displaying the land value heat map.

[0140] In an embodiment, after obtaining the land assessment data, the land value heat map can be constructed using the land assessment data, and finally the land value heat map can be visually displayed on the screen of the device for the user to view. Through the land value heat map, the user can more intuitively view the assessment results, thereby facilitating land resource management, development and utilization according to the assessment results.

[0141] In one embodiment, step S13 can include the following sub-steps:

[0142] S131, converting the non-spatial attribute data of the land assessment data into a geographic coordinate system by inverse geocoding, and normalizing the land assessment data to obtain normalized data.

[0143] S132, generating a continuous spatial grid surface using the normalized data, and adjusting the color gradient of the continuous spatial grid surface using a color band tool to obtain a land value heat map.

[0144] S133, after format conversion of the land value heat map, the land value heat map is visually displayed.

[0145] In an embodiment, the land assessment data can contain latitude and longitude coordinates and value grades. After obtaining the land assessment data, the non-spatial attribute data can be converted into a geographic coordinate system by inverse geocoding, and the land assessment data and the value indicators in the database can be normalized. A continuous spatial grid surface can be generated using Kriging interpolation or inverse distance weighting algorithm, and the color gradient can be adjusted using a color band tool to define a non-linear color gradient (blue to red spectrum corresponding to low value to high value), and the transparency can be dynamically adjusted in combination with the confidence interval quantization result.

[0146] Finally, the interpolated grid data can be converted into Quantized-Mesh format by Cesium engine, overlaid to the base map in WMS service or real-time streaming form, and integrated with the parcel boundary vector layer to realize multi-layer spatial registration. Finally, the land value heat map can be visually displayed. Using LOD dynamic detail level control and WebGL GPU accelerated rendering strategy, the resolution can be adaptively switched according to the view zoom level, while embedding the time axis control to support the time series comparison of historical and predicted heat maps, realizing the local dynamic update of incremental data, and ensuring the efficient visualization and interactive analysis of large-scale spatial data.

[0147] Referring to Figure 2 , a flow chart of an operation of a land assessment data visualization display method provided by an embodiment of the present application is shown.

[0148] Specifically, the operation of the land assessment data visualization display method can include the following operations:

[0149] First, data acquisition and preprocessing, multi-source heterogeneous data is collected in parallel, and the data is standardized, spatially aligned, and time series corrected.

[0150] Second, data distillation and optimization, which can include redundant screening (compressing redundant features based on feature importance evaluation) and dimensionality reduction enhancement (automatic encoder dimensionality reduction, attention mechanism dynamic weighting).

[0151] Third, spatio-temporal convolutional neural network modeling, which can include dual-flow networks (a spatial flow: 3D convolution extracts spatial structure features; b time flow: GRU network modeling time evolution) and feature fusion and prediction (cross-attention fusion dual-flow features, output land price change rate and value grade probability).

[0152] Fourth, dynamic monitoring and optimization, including online learning (sliding window updating model parameters, real-time adaptation to market changes) and uncertainty quantification (Monte Carlo Dropout technology generates land value prediction confidence interval, supports risk early warning).

[0153] Fifth, visualization, including hierarchical architecture (data access layer, feature processing layer, model service layer, and application interaction layer) and decision support (GIS heat map display, early warning push, and automatic report generation).

[0154] In this embodiment, the land assessment data visualization display method provided by the embodiment of the present application has the beneficial effects that the present application can construct a multi-modal feature database of land resources; the multi-modal feature database is used to train a preset model to obtain a land assessment model, and the land assessment model is called to perform assessment processing to obtain land assessment data; the land assessment data is used to construct a land value heat map, and the land value heat map is visualized; through database training model and assessment, the assessment accuracy can be improved, and through heat map visualization, the display effect can be improved.

[0155] The embodiment of the present application also provides a land assessment data visualization display device, referring to Figure 3 , a structural schematic diagram of a land assessment data visualization display device provided by an embodiment of the present application is shown.

[0156] The visualization display device of the land assessment data may include, for example:

[0157] The construction module 201 is configured to construct a multi-modal feature database of land resources, the multi-modal feature database being a database constructed by using land value monitoring indicators and land heterogeneous data;

[0158] The evaluation module 202 is configured to train a preset model by using the multi-modal feature database to obtain a land assessment model, and to call the land assessment model to perform evaluation processing to obtain land assessment data;

[0159] The display module 203 is configured to construct a land value heat map by using the land assessment data, and to visually display the land value heat map.

[0160] Optionally, the multi-modal feature database of land resources includes:

[0161] The land value monitoring indicators and the land heterogeneous data are determined, and an initial database is constructed by using the land value monitoring indicators and the land heterogeneous data;

[0162] The initial database is subjected to data distillation to obtain a multi-modal feature database.

[0163] Optionally, the determination of the land value monitoring indicators and the land heterogeneous data, and the construction of the initial database by using the land value monitoring indicators and the land heterogeneous data include:

[0164] The land value monitoring indicators predefined by a user are determined, and land heterogeneous data is obtained;

[0165] The land heterogeneous data is preprocessed to obtain heterogeneous processing data;

[0166] The heterogeneous processing data is sequentially subjected to coordinate conversion, multi-scale decomposition and merging to obtain static feature data and dynamic feature data;

[0167] The initial database is constructed by using the land value monitoring indicators, the static feature data and the dynamic feature data.

[0168] Optionally, the data distillation of the initial database to obtain a multi-modal feature database includes:

[0169] The importance score value of each data of the initial database is calculated based on a random forest;

[0170] The initial database is subjected to data distillation according to the importance score value according to a dynamic distillation strategy to obtain a multi-modal feature database.

[0171] Optionally, the preset model is a spatio-temporal convolutional neural network architecture.

[0172] The spatio-temporal convolutional neural network architecture comprises a spatial stream network and a temporal stream network, and dynamically fuses output features of the spatial stream network and the temporal stream network through a cross-attention mechanism.

[0173] The training of the preset model adopts multi-task learning and regularization constraint.

[0174] Optionally, the device further comprises:

[0175] The updating module is configured to, after the step of training the preset model by using the multi-modal feature database to obtain a land assessment model, update parameters of the land assessment model based on an online learning mechanism, and calculate a standard deviation of a prediction result of the land assessment model based on a Monte Carlo technique to generate a land value prediction confidence interval by using the standard deviation.

[0176] Optionally, the construction of the land value heat map by using the land assessment data and the visual display of the land value heat map comprise:

[0177] The non-spatial attribute data of the land assessment data is converted into a geographic coordinate system through reverse geocoding, and the land assessment data is normalized to obtain normalized data.

[0178] The normalized data is used to generate a continuous spatial grid surface, and a color gradient of the continuous spatial grid surface is adjusted by using a color band tool to obtain a land value heat map.

[0179] After the format conversion of the land value heat map, the land value heat map is visually displayed.

[0180] The embodiment of the present application also provides a visual display system of land assessment data, referring to Figure 4 , a structural schematic diagram of a visual display system of land assessment data provided by an embodiment of the present application is shown.

[0181] Among them, as an example, the visual display system of land assessment data can comprise:

[0182] The data access layer supports real-time / offline access of multi-source heterogeneous data, including remote sensing image stream (based on GDAL library analysis), land ownership database (SQL interface), social and economic API (RESTful interface), sensor data (MQTT protocol), etc., and provides data format standardization and outlier cleaning functions.

[0183] Feature processing layer: integrates multi-modal feature fusion engine (based on Spark distributed computing framework) and data distillation module, realizes feature alignment, dimensionality reduction optimization and spatio-temporal attention weighting, and outputs standardized feature matrix to model service layer.

[0184] Model service layer: deploy spatio-temporal convolutional neural network (ST-CNN) prediction model, provide high-concurrency inference service based on TensorFlow Serving, support online learning and model version management (MLflow).

[0185] Application interaction layer: provides Web visualization interface (Vue.js + ECharts), GIS map engine (Cesium) and API interface.

[0186] Data access layer realizes dynamic collection and standardized processing of multi-source heterogeneous data, including: accessing multi-temporal high-resolution remote sensing images, performing radiation correction and atmospheric correction based on GDAL library; synchronously crawling land ownership data, enterprise registration information and project approval data from public platforms, using entity recognition technology to build structured knowledge graph; integrating Apache Kafka message queue to realize second-level transmission and buffering of real-time data such as population heat and land transaction flow.

[0187] Feature processing layer realizes dynamic prediction of land value through automatic feature engineering and deep learning model, including: calling RSBuilding model API to extract three-dimensional geometric parameters (height, contour, volume) of building at plot level, combining Prophet algorithm to decompose trend and periodic terms of land price time series data; based on Spark distributed framework, spatio-temporal attention weighted fusion of multi-source features is realized, and low-dimensional high-discriminability feature matrix is generated.

[0188] Application interaction layer provides multi-dimensional interactive decision support tools, including: building GIS visualization dashboard based on Cesium engine, superimposing land value heat map, development state warning layer and historical comparison view, supporting spatio-temporal sliding analysis; integrating NLP large language model, using domain-adapted Prompt template and LoRA light-weight fine-tuning technology to automatically convert structured prediction data into compliance analysis report, covering potential plot list, policy recommendations and visual chart references; after report generation, regular expression verification logic integrity, output PDF / Excel format file, and trigger overdue undeveloped plot warning through rule engine (Drools).

[0189] Those skilled in the art can clearly understand that, for the convenience of description and brevity, the specific working process of the above-described device can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0190] Further, the embodiment of the present application further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the program to realize the method for visualizing and displaying land assessment data according to the above embodiment.

[0191] Further, the embodiment of the present application further provides a computer readable storage medium, which stores a computer executable program, and the computer executable program is used to make a computer execute the method for visualizing and displaying land assessment data according to the above embodiment.

[0192] In the description of the embodiment of the present application, it should be noted that the terms "upper", "lower", and the like indicate the orientation or positional relationship shown in the drawings, and are only used to facilitate the description of the embodiment of the present application and simplify the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation of the present application. When an element such as a layer, a region, or a substrate is referred to as "on" or "above" another element, it can be directly on the other element, or there can be an intermediate element. In contrast, when an element is referred to as "directly on" or "directly above" another element, there is no intermediate element. It should also be understood that when an element is referred to as "below" or "under" another element, it can be directly below or under the other element, or there can be an intermediate element. In contrast, when an element is referred to as "directly below" or "directly under" another element, there is no intermediate element. Unless otherwise specified and limited, the terms "mount", "connect", "connect" should be understood in a broad sense, for example, it can be fixedly connected, or it can be detachably connected, or integrally connected; it can be mechanically connected, or it can be electrically connected; it can be directly connected, or it can be indirectly connected through an intermediate medium, or it can be connected inside two elements. For those skilled in the art, the specific meaning of the above terms in the present application can be understood according to the specific circumstances.

[0193] Those skilled in the art will appreciate that the embodiments of the present application can also provide a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer usable program code.

[0194] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0195] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the function specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0196] The computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks. Figure 1 one or more flow or blocks. Figure 1 one or more flow or blocks.

[0197] The above only is the preferred embodiment of the present application, it should be pointed out that, for those skilled in the technical field, without departing from the technical principles of the present application, can also make a number of improvements and variations, these improvements and variations should also be considered as the protection scope of the present application.

Claims

1. A method of visualizing land assessment data, characterized by, The method comprises: constructing a multi-modal feature database of land resources, the multi-modal feature database being a database constructed using land value monitoring indicators and land heterogeneous data; training a preset model using the multi-modal feature database to obtain a land assessment model, and calling the land assessment model for assessment processing to obtain land assessment data; constructing a land value heat map using the land assessment data, and visually displaying the land value heat map.

2. The method of claim 1, wherein, The method of constructing a multi-modal feature database of land resources comprises: determining land value monitoring indicators and land heterogeneous data, and constructing an initial database using the land value monitoring indicators and the land heterogeneous data; data distillation is performed on the initial database to obtain a multi-modal feature database.

3. The method of claim 2, wherein, The method of determining land value monitoring indicators and land heterogeneous data, and constructing an initial database using the land value monitoring indicators and the land heterogeneous data comprises: determining land value monitoring indicators predefined by a user, and obtaining land heterogeneous data; preprocessing the land heterogeneous data to obtain heterogeneous processed data; performing coordinate conversion, multi-scale decomposition and merging on the heterogeneous processed data in sequence to obtain static feature data and dynamic feature data; constructing an initial database using the land value monitoring indicators, the static feature data and the dynamic feature data.

4. The method of claim 2, wherein, The method of performing data distillation on the initial database to obtain a multi-modal feature database comprises: calculating an importance score value of each data of the initial database based on a random forest; performing data distillation on the initial database according to the importance score value based on a dynamic distillation strategy to obtain a multi-modal feature database.

5. The method of claim 1, wherein, The preset model is a spatio-temporal convolutional neural network architecture; The spatio-temporal convolutional neural network architecture comprises a spatial stream network and a temporal stream network, and dynamically fuses output features of the spatial stream network and the temporal stream network through a cross-attention mechanism; The training of the preset model adopts multi-task learning and regularization constraint.

6. The method of claim 1-5, wherein, After the step of training a preset model using the multi-modal feature database to obtain a land assessment model, the method further comprises: updating parameters of the land assessment model based on an online learning mechanism, and calculating a standard deviation of a prediction result of the land assessment model based on a Monte Carlo technique to generate a land value prediction confidence interval using the standard deviation.

7. The method of claim 1-5, wherein, The method of constructing a land value heat map using the land assessment data, and visually displaying the land value heat map comprises: converting non-spatial attribute data of the land assessment data into a geographic coordinate system through reverse geocoding, and performing normalization processing on the land assessment data to obtain normalized data; generating a continuous spatial grid surface using the normalized data, and adjusting a color gradient of the continuous spatial grid surface using a color band tool to obtain a land value heat map; after format conversion of the land value heat map, the land value heat map is visually displayed.

8. An apparatus for visualizing land assessment data, characterized by The device comprises: The construction module is configured to construct a multi-modal feature database of land resources, which is a database constructed by using land value monitoring indexes and land heterogeneous data; The evaluation module is configured to train a preset model by using the multi-modal feature database to obtain a land evaluation model, and to call the land evaluation model to perform evaluation processing to obtain land evaluation data; The display module is configured to construct a land value heat map by using the land evaluation data, and to visually display the land value heat map.

9. An electronic device comprising: The memory, the processor, and the computer program stored in the memory and executable on the processor are characterized in that the processor executes the program to implement the method for visually displaying land evaluation data according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions for causing a computer to execute the method for visually displaying land evaluation data according to any one of claims 1-7.

Images