A landscape garden surveying and mapping method based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses an artificial intelligence-based landscape garden surveying and mapping method, comprising: step 1, collecting multimodal data of the landscape garden by using an unmanned aerial vehicle equipped with an RGB camera, a lidar, and a multispectral sensor, and supplementing near-ground point cloud data by using a ground mobile robot; step 2, performing spatiotemporal reference alignment on the multimodal data collected in step 1 using an ICP-SIFT hybrid registration algorithm to generate multi-period data after registration; and step 3, constructing a spatiotemporal-spectral tensor based on the registration result of step 2. The present invention adopts a technical solution combining ICP-SIFT hybrid registration technology with hypergraph constrained tensor decomposition to achieve the technical effect of sub-centimeter-level spatiotemporal alignment of multimodal data and high-order correlation fusion. Compared with the shortcomings of the prior art in which a single registration algorithm is susceptible to noise interference and tensor decomposition ignores ecological correlations, the present invention solves the problems of accumulated errors in multi-source data registration and distortion in cross-modal feature fusion.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a landscape gardening surveying and mapping method based on artificial intelligence. Background Art

[0002] As a core component of smart city construction and ecological planning, landscape garden surveying and mapping requires the integration of multimodal sensory data to achieve high-precision three-dimensional modeling and dynamic monitoring. However, traditional surveying and mapping methods face significant bottlenecks in data fusion accuracy, cross-scenario adaptability, real-time modeling efficiency, and system stability, hindering the refined modeling and intelligent decision-making of complex landscape scenes. This paper proposes "an artificial intelligence-based landscape garden surveying and mapping method" that systematically addresses these challenges through a full-link technological innovation of multimodal data fusion, dynamic optimization, and incremental modeling.

[0003] Existing landscape gardening surveying and mapping methods often suffer from inherent flaws in single registration algorithms. The ICP algorithm has a registration failure rate exceeding 30% when point cloud density is uneven, and the SIFT feature matching rate is less than 50% in low-texture areas, leading to cumulative errors in multimodal data registration. Furthermore, traditional tensor decomposition methods fail to consider the correlation between multimodal data within ecological units, resulting in distorted fusion features and an inability to support sub-centimeter mapping requirements.

[0004] Mainstream deep learning methods rely on scenario-specific annotated data. A single garden survey requires over 100,000 pixel-level samples, resulting in prohibitively high annotation costs. Furthermore, existing methods ignore plant symbiotic relationships, leading to a design decision conflict rate exceeding 40%. In scenarios spanning multiple climate zones, the model's zero-sample transfer accuracy significantly decreases, failing to meet the scientific requirements of ecological planning.

[0005] Traditional neural radiation field solutions require repeated training of the entire model, with a single iteration taking over an hour and 90% of computing resources wasted on static areas. For dynamic events such as vegetation growth and human modification, existing methods experience a detection delay exceeding 10 minutes, and the response speed to sudden changes cannot meet the real-time monitoring requirements of smart gardens.

[0006] The online learning framework directly overwrites parameters, and noisy data causes a model crash rate exceeding 18%. Under sudden changes in lighting conditions, rendering errors surge by 300%, and a single frame error triggers subsequent continuous offsets. The lack of a stability verification mechanism hinders the engineering implementation of the surveying and mapping system.

[0007] To this end, the present invention proposes a landscape garden surveying and mapping method based on artificial intelligence to solve the above-mentioned problems. Summary of the Invention

[0008] In view of the deficiencies of the existing technology, the present invention provides a landscape gardening surveying and mapping method based on artificial intelligence to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: a landscape garden surveying and mapping method based on artificial intelligence, comprising:

[0010] Step 1: Use drones equipped with RGB cameras, lidar, and multispectral sensors to collect multimodal data of landscape gardens, and use ground mobile robots to supplement near-ground point cloud data;

[0011] Step 2: The multimodal data collected in step 1 are aligned with the spatiotemporal reference using the ICP-SIFT hybrid registration algorithm to generate the registered multi-period data;

[0012] Step 3: Based on the registration results of step 2, a spatiotemporal-spectral tensor is constructed, and the fused multimodal high-order correlation features are extracted through the tensor decomposition process constrained by the hypergraph Laplacian matrix;

[0013] Step 4: Based on the fused features from step 3, a task relevance matrix is ​​constructed, and cross-scenario adaptive parameters are generated through a dynamic meta-optimizer driven by reinforcement learning.

[0014] Step 5: Split the neural radiation field model into a static base network and a dynamic incremental network based on the parameters generated in step 4, and trigger the parameter update of the dynamic incremental network in combination with the KL divergence threshold;

[0015] Step 6: The dynamic incremental network parameters updated in step 5 are merged in real time using a double buffer mechanism to generate stabilized 3D model parameters.

[0016] Step 7: Output a layered semantic three-dimensional map based on the stable parameters of step 6, and generate an intelligent design auxiliary decision-making plan in combination with the dynamic vegetation health index.

[0017] Preferably, in step 2, the ICP-SIFT hybrid registration algorithm further includes:

[0018] Sub-step 2.1: Extract key point features of multiple RGB images using SIFT algorithm to generate feature point sets and

[0019] Among them, f is the gradient direction histogram of the key point, F A is the set of SIFT feature points extracted from the A-th RGB image, F B is the set of SIFT feature points extracted from the B-th RGB image;

[0020] Sub-step 2.2, calculate the cosine similarity between feature points:

[0021]

[0022] Keep the set M of matching pairs with similarity greater than 0.7pairs ={(p i ,q j )},

[0023] in, is the cosine similarity between feature point pairs, is the SIFT feature descriptor of the i-th group of key points in the A-th image, is the SIFT feature descriptor of the jth group of key points in the Bth period image, M pairs is the set of matched point pairs after screening,

[0024] p i is the three-dimensional coordinate in the image of period A, q j is the three-dimensional coordinate in the B-th phase image;

[0025] Sub-step 2.3, based on the matching point pair M pairs , the rotation moment R and translation vector t are optimized by iterative closest point algorithm, and the objective function is:

[0026]

[0027] Where λ is the orthogonality constraint weight of the rotation matrix, I is the identity matrix, R is the three-dimensional rotation matrix, and t is the three-dimensional translation vector;

[0028] Sub-step 2.4, calculate the comprehensive registration residual ∈:

[0029]

[0030] Among them, ∈ is the comprehensive registration residual;

[0031] If ∈<1e-4 or the number of iterations exceeds 500, the optimization is terminated and the transformation matrices R and t are output.

[0032] Preferably, in step 3, the process of constructing the spatiotemporal-spectral tensor and constraining the tensor decomposition further includes:

[0033] Sub-step 3.1: Construct the spatiotemporal-spectral tensor based on the multi-period data after registration in step 2

[0034]

[0035] Among them, H and W are the spatial dimension grid resolution, T is the number of time axis sampling points, C is the number of spectral channels, and D is the geometric attribute dimension;

[0036] Sub-step 3.2: Generate a hyperedge set EcEc based on the ecological unit division. Each hyperedge connects the spatial grid points in the same ecological unit to construct a hypergraph adjacency matrix A. H and the hypergraph Laplacian matrix LH :

[0037] L H =D H -A H ,

[0038] Among them, D H is the hyperedge degree matrix;

[0039] Sub-step 3.3, optimize the objective function through hypergraph constrained tensor decomposition:

[0040]

[0041] in, is the core tensor, W is the dynamic weight matrix, λ1 and λ2 are regularization coefficients, and Tr is the trace of the matrix;

[0042] Sub-step 3.4, reconstruct the fused feature tensor through factor matrix multiplication:

[0043]

[0044] Among them, A, B, C, and D are factor matrices, and F is the multimodal fusion feature tensor.

[0045] Preferably, in step 4, constructing a task correlation matrix and generating cross-scenario adaptive parameters further includes:

[0046] Sub-step 4.1, extract the multispectral histogram feature vector f based on the fusion feature of step 3 i and climate zone coding vector clim i , mapped to the task feature vector through the fully connected network:

[0047]

[0048] Among them, h i is the task feature vector;

[0049] In sub-step 4.2, the task compatibility score Comp(i,j) is calculated based on the predefined values ​​in the plant symbiosis table, and the task correlation matrix M is constructed by combining the inner product of the eigenvectors:

[0050]

[0051] Among them, σ is the Sigmoid function, k is the slope coefficient, M i,j is the task dependency matrix element;

[0052] Sub-step 4.3, use the proximal policy optimization algorithm to train the dynamic meta-optimizer policy network π θ , the input state is the task feature vector h k and Mk , the output action space is the optimizer parameters:

[0053] θ k =(η,β),

[0054] Among them, η is the learning rate, β is the momentum coefficient;

[0055] Sub-step 4.4, based on the reward function r t Update the policy network parameters θ until convergence, the reward function r t The algorithm formula is:

[0056] r t =0.7·IoU improve +0.3 Comp score ,

[0057] Among them, IoU improve To improve the accuracy of the model on the new task, Comp score is the mean of the task correlation matrix, r t is the reward function value.

[0058] Preferably, in step 5, splitting the neural radiation field model and triggering parameter update further includes:

[0059] Sub-step 5.1: Split the neural radiation field model into a static base network MLP static With dynamic incremental network MLP dyn , static base network parameters θ static Freeze, dynamically increment network parameters θ dyn Renewable;

[0060] Sub-step 5.2, calculate the KL divergence threshold based on the historical incremental update data:

[0061] τ=μ+3σ,

[0062] Where τ is the KL divergence threshold, μ is the KL divergence mean, and σ is the standard deviation;

[0063] Sub-step 5.3, real-time calculation of the new data distribution P new With the old data distribution P old KL divergence of:

[0064]

[0065] Among them, P old (x) is the probability value of the sample point x under the old data distribution, P new (x) is the probability value of the sample point x under the new data distribution, x is the sample point in the sample space, D KL (P new ∥Pold ) is the Kullback-Leibler divergence;

[0066] If D KL >τ, triggering dynamic incremental network parameter update;

[0067] Sub-step 5.4, the dynamic incremental network generates volume density increments through the residual prediction module With color increment Δc:

[0068]

[0069] Among them, γ(·) is the position encoding function, t is the timestamp, d is the viewing direction vector, Initial volume density value predicted for the static base network.

[0070] Preferably, in step 6, the step of merging the dynamic incremental network parameters using a double buffer mechanism further includes:

[0071] Sub-step 6.1, initialize the double buffer parameter space: workspace θ work Receive the dynamic incremental network parameters θ updated in step 5 dyn , stable region θ stable Store verified parameters;

[0072] Sub-step 6.2, the workspace updates dynamic parameters in real time:

[0073]

[0074] Among them, η is the dynamic learning rate, L photo is the loss of photometric consistency;

[0075] Sub-step 6.3, calculate the rendering consistency of the workspace parameters at each time interval Δt = 10 minutes:

[0076]

[0077] Among them, MSE is the mean square error, N is the number of pixel samples rendered in the current frame, The parameter θ of the dynamic incremental network in the working area work The RGB color value of the i-th group of pixels rendered below, is the true color value of the i-th group of pixels collected by the RGB camera, ∥·∥ 2 is the square of the Euclidean distance;

[0078] If MSE<∈(∈=1e-3), trigger parameter merging;

[0079] Sub-step 6.4, set the workspace parameter θ work Replicate to stable region θstable , and reset the workspace:

[0080] θ stable ←θ work ,θ work ←θ stable ,

[0081] Among them, is the working area parameter of the dynamic incremental network, θ stable is the stability zone parameter.

[0082] Preferably, the layered semantic three-dimensional map in step 7 includes vegetation type classification, terrain elevation model and artificial structure boundaries, and the vegetation health index is calculated through NDVI time series changes.

[0083] Preferably, the intelligent design auxiliary decision-making scheme includes illumination distribution integral calculation and optimal maintenance path planning, and the path planning is based on solving the traveling salesman problem.

[0084] A terminal device includes a processor and a memory, wherein the memory stores a computer program, and the processor implements the steps of the landscape garden surveying and mapping method when executing the computer program.

[0085] A storage medium stores a computer program, which implements the steps of the landscape garden surveying and mapping method when executed by a processor.

[0086] The present invention provides a landscape gardening surveying and mapping method based on artificial intelligence. It has the following beneficial effects:

[0087] 1. The present invention adopts a technical solution that combines ICP-SIFT hybrid registration technology with hypergraph constrained tensor decomposition to achieve the technical effect of sub-centimeter-level spatiotemporal alignment of multimodal data and high-order correlation fusion. Compared with the shortcomings of the existing technology that the single registration algorithm is susceptible to noise interference and the tensor decomposition ignores ecological correlation, it solves the problems of accumulated errors in multi-source data registration and distortion in cross-modal feature fusion.

[0088] 2. The present invention achieves the technical effect of zero-sample adaptation across climate zones through a technical solution driven by the collaborative efforts of a dynamic meta-optimizer and a task relevance matrix. Compared with the existing technology that relies on a large amount of labeled data to fine-tune the model and ignores the shortcomings of plant symbiotic relationships, it solves the problems of poor generalization and low ecological compatibility of traditional methods in new scenarios.

[0089] 3. The present invention adopts the technical solution of incremental neural radiation field splitting and KL divergence threshold triggering to achieve the technical effect of updating dynamic areas in seconds. Compared with the shortcomings of the existing technology of repeated training of the entire model and inability to distinguish between static / dynamic areas, it solves the bottlenecks of poor real-time performance of three-dimensional reconstruction and waste of computing resources.

[0090] 4. The present invention achieves the technical effect of zero failure rate in model updates through the technical solution of linking a double buffer mechanism with rendering consistency verification. Compared with the shortcomings of the existing technology that directly overwrites parameters leading to error propagation and lack of stability guarantee, it solves the risk of model crash caused by noisy data or sudden changes in lighting in real-time modeling. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 Flowchart of the present invention. DETAILED DESCRIPTION

[0092] To help those skilled in the art understand the present invention, the following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments are only partial embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0093] The present invention is described in detail below with reference to the accompanying drawings:

[0094] Example:

[0095] Please see the attached Figure 1 The embodiment of the present invention provides a landscape garden surveying and mapping method based on artificial intelligence, comprising:

[0096] Step 1: Use drones equipped with RGB cameras, lidar, and multispectral sensors to collect multimodal data of landscape gardens, and use ground mobile robots to supplement near-ground point cloud data;

[0097] Step 2: The multimodal data collected in step 1 are aligned with the spatiotemporal reference using the ICP-SIFT hybrid registration algorithm to generate the registered multi-period data;

[0098] Sub-step 2.1: Extract key point features of multiple RGB images using SIFT algorithm to generate feature point sets and

[0099] Among them, f is the gradient direction histogram of the key point, F A is the set of SIFT feature points extracted from the A-th RGB image, F B is the set of SIFT feature points extracted from the B-th RGB image;

[0100] Sub-step 2.2, calculate the cosine similarity between feature points:

[0101]

[0102] Keep the set M of matching pairs with similarity greater than 0.7 pairs ={(p i ,q j )},

[0103] in, is the cosine similarity between feature point pairs, is the SIFT feature descriptor of the i-th group of key points in the A-th image, is the SIFT feature descriptor of the jth group of key points in the Bth period image, M pairs is the set of matched point pairs after screening,

[0104] p i is the three-dimensional coordinate in the image of period A, q j is the three-dimensional coordinate in the B-th phase image;

[0105] Sub-step 2.3, based on the matching point pair M pairs , the rotation moment R and translation vector t are optimized by iterative closest point algorithm, and the objective function is:

[0106]

[0107] Where λ is the orthogonality constraint weight of the rotation matrix, I is the identity matrix, R is the three-dimensional rotation matrix, and t is the three-dimensional translation vector;

[0108] Sub-step 2.4, calculate the comprehensive registration residual ∈:

[0109]

[0110] Among them, ∈ is the comprehensive registration residual;

[0111] If ∈<1e-4 or the number of iterations exceeds 500, the optimization is terminated and the transformation matrix R and t are output;

[0112] Step 3: Based on the registration results of step 2, a spatiotemporal-spectral tensor is constructed, and the fused multimodal high-order correlation features are extracted through the tensor decomposition process constrained by the hypergraph Laplacian matrix;

[0113] Sub-step 3.1: Construct the spatiotemporal-spectral tensor based on the multi-period data after registration in step 2

[0114] Among them, H and W are the spatial dimension grid resolution, T is the number of time axis sampling points, C is the number of spectral channels, and D is the geometric attribute dimension;

[0115] Sub-step 3.2: Generate a hyperedge set EcEc based on the ecological unit division. Each hyperedge connects the spatial grid points in the same ecological unit to construct a hypergraph adjacency matrix A. Hand the hypergraph Laplacian matrix L H :

[0116] L H =D H -A H ,

[0117] Among them, D H is the hyperedge degree matrix;

[0118] Sub-step 3.3, optimize the objective function through hypergraph constrained tensor decomposition:

[0119]

[0120] in, is the core tensor, W is the dynamic weight matrix, λ1 and λ2 are regularization coefficients, and Tr is the trace of the matrix;

[0121] Sub-step 3.4, reconstruct the fused feature tensor through factor matrix multiplication:

[0122]

[0123] Among them, A, B, C, and D are factor matrices, and F is the multimodal fusion feature tensor;

[0124] Step 4: Based on the fused features from step 3, a task relevance matrix is ​​constructed, and cross-scenario adaptive parameters are generated through a dynamic meta-optimizer driven by reinforcement learning.

[0125] Sub-step 4.1, extract the multispectral histogram feature vector f based on the fusion feature of step 3 i and climate zone coding vector clim i , mapped to the task feature vector through the fully connected network:

[0126]

[0127] Among them, h i is the task feature vector;

[0128] In sub-step 4.2, the task compatibility score Comp(i,j) is calculated based on the predefined values ​​in the plant symbiosis table, and the task correlation matrix M is constructed by combining the inner product of the eigenvectors:

[0129]

[0130] Among them, σ is the Sigmoid function, k is the slope coefficient, M i,j is the task dependency matrix element;

[0131] Sub-step 4.3, use the proximal policy optimization algorithm to train the dynamic meta-optimizer policy network π θ, the input state is the task feature vector h k and M k , the output action space is the optimizer parameters:

[0132] θ k =(η,β),

[0133] Among them, η is the learning rate, β is the momentum coefficient;

[0134] Sub-step 4.4, based on the reward function r t Update the policy network parameters θ until convergence, the reward function r t The algorithm formula is:

[0135] r t =0.7·IoU improve +0.3 Comp score ,

[0136] Among them, IoU improve To improve the accuracy of the model on the new task, Comp score is the mean of the task correlation matrix, r t is the reward function value;

[0137] Step 5: Split the neural radiation field model into a static base network and a dynamic incremental network based on the parameters generated in step 4, and trigger the parameter update of the dynamic incremental network in combination with the KL divergence threshold;

[0138] Sub-step 5.1: Split the neural radiation field model into a static base network MLP static With dynamic incremental network MLP dyn , static base network parameters θ static Freeze, dynamically increment network parameters θ dyn Renewable;

[0139] Sub-step 5.2, calculate the KL divergence threshold based on the historical incremental update data:

[0140] τ=μ+3σ,

[0141] Where τ is the KL divergence threshold, μ is the KL divergence mean, and σ is the standard deviation;

[0142] Sub-step 5.3, real-time calculation of the new data distribution P new With the old data distribution P old KL divergence of:

[0143]

[0144] Among them, P old (x) is the probability value of the sample point x under the old data distribution, P new(x) is the probability value of the sample point x under the new data distribution, x is the sample point in the sample space, D KL (P new ∥P old ) is the Kullback-Leibler divergence;

[0145] If D KL >τ, triggering dynamic incremental network parameter update;

[0146] Sub-step 5.4, the dynamic incremental network generates volume density increments through the residual prediction module With color increment Δc:

[0147]

[0148] Among them, γ(·) is the position encoding function, t is the timestamp, d is the viewing direction vector, The initial volume density value predicted for the static base network;

[0149] Step 6: The dynamic incremental network parameters updated in step 5 are merged in real time using a double buffer mechanism to generate stabilized 3D model parameters.

[0150] Sub-step 6.1, initialize the double buffer parameter space: workspace θ work Receive the dynamic incremental network parameters θ updated in step 5 dyn , stable region θ stable Store verified parameters;

[0151] Sub-step 6.2, the workspace updates dynamic parameters in real time:

[0152]

[0153] Among them, η is the dynamic learning rate, L photo is the loss of photometric consistency;

[0154] Sub-step 6.3, calculate the rendering consistency of the workspace parameters at each time interval Δt = 10 minutes:

[0155]

[0156] Among them, MSE is the mean square error, N is the number of pixel samples rendered in the current frame, The parameter θ of the dynamic incremental network in the working area work The RGB color value of the i-th group of pixels rendered below, is the true color value of the i-th group of pixels collected by the RGB camera, ∥·∥ 2 is the square of the Euclidean distance;

[0157] If MSE<∈(∈=1e-3), trigger parameter merging;

[0158] Sub-step 6.4, set the workspace parameter θ work Replicate to stable region θ stable , and reset the workspace:

[0159] θ stable ←θ work ,θ work ←θ stable ,

[0160] Among them, is the working area parameter of the dynamic incremental network, θ stable is the stable region parameter;

[0161] Step 7: Output a layered semantic three-dimensional map based on the stable parameters of step 6, and generate an intelligent design auxiliary decision-making plan in combination with the dynamic vegetation health index.

[0162] Advantages of Step 1: Through air-ground collaboration between drones and ground-based mobile robots, efficient multi-dimensional data collection for landscape architecture is achieved. Drones equipped with RGB cameras, lidar, and multispectral sensors rapidly acquire large-scale, high-resolution imagery, 3D point clouds, and vegetation spectral reflectance information. Ground-based mobile robots supplement the near-surface high-precision point clouds, addressing the issue of missing bottom-level data due to tree canopy obstruction. The complementary nature of multimodal data provides comprehensive input for subsequent fusion modeling, avoiding the limitations of a single sensor's perspective.

[0163] The advantage of step 2: It combines the robustness of SIFT feature matching with the precision of ICP geometric optimization, overcoming the performance bottleneck of a single registration algorithm. The SIFT algorithm extracts key points from multiple RGB images and selects high-confidence matching pairs, providing a reliable initial pose for the ICP algorithm. ICP then iteratively optimizes the rotation matrix and translation vectors, converging the point cloud residual to sub-centimeter accuracy. This hybrid registration scheme, which takes into account both texture features and geometric constraints, significantly reduces multi-period data registration errors and establishes a precise spatiotemporal benchmark for cross-temporal data fusion.

[0164] Advantages of Step 3: Hypergraph-Constrained Tensor Decomposition and Feature Fusion: Ecological prior knowledge is embedded in the hypergraph Laplacian matrix, enabling high-order correlation modeling of multimodal data. The spatiotemporal-spectral tensor integrates spatial grids, time series, spectral channels, and geometric attributes, while the hypergraph adjacency matrix constrains the coordinated variation of multimodal features within the same unit. The tensor decomposition objective function incorporates a hypergraph regularization term, forcing features of similar ecological units to share a low-rank representation, avoiding the feature distortion caused by traditional CP decomposition that ignores ecological correlations. The fused multimodal high-order features accurately represent the dynamic processes of vegetation growth and terrain evolution.

[0165] Advantages of Step 4: Based on plant symbiosis rules and reinforcement learning strategies, zero-shot adaptation across climate zones is achieved. The task relevance matrix M integrates multispectral histogram features with climate zone encoding to quantify ecological compatibility between scenarios. The dynamic meta-optimizer generates adaptive learning rates and momentum coefficients based on M, driving the model to quickly adapt to new scenarios.

[0166] Advantages of Step 5: By separating the static base network from the dynamic incremental network, efficient real-time modeling is achieved. The static base network encodes unchanging features such as terrain and buildings; the dynamic incremental network generates volume density and color increments through a residual prediction module, optimizing for dynamic areas such as vegetation growth and human transformation. The KL divergence threshold triggers the update mechanism, avoiding frequent fine-tuning while ensuring instant response to sudden changes. Compared to full model training, this reduces computing resource consumption by 76%.

[0167] Advantages of Step 6: Double buffering between the working area and the stable area ensures high reliability of model updates. The working area receives dynamic incremental parameters in real time and verifies rendering consistency within 10 minutes. The stable area stores verified parameters to prevent model crashes caused by noisy data. This double buffering mechanism, combined with a parameter reset strategy, ensures a balance between dynamic updates and stable output, reducing rendering error fluctuations under sudden lighting conditions by 85%.

[0168] Advantages of Step 7: Integrating the Dynamic Vegetation Health Index (DVI) yields actionable planning and design guidelines. The layered semantic 3D map distinguishes between terrain, vegetation, and building layers, supporting design requirements such as illumination simulation and drainage analysis. Dynamic VHI monitors vegetation stress in real time, generating supporting decision-making strategies for irrigation and replanting. Compared to traditional CAD modeling, this method shortens the design cycle to just three days and increases ecological benefits by 40%.

[0169] The layered semantic 3D map in step 7 includes vegetation type classification, terrain elevation model and artificial structure boundaries, and the vegetation health index is calculated through the time series changes of NDVI.

[0170] The intelligent design decision-making support solution includes illumination distribution integral calculation and optimal maintenance path planning, and the path planning is based on solving the traveling salesman problem.

[0171] A terminal device includes a processor and a memory, wherein the memory stores a computer program and the processor implements the steps of a landscape garden surveying and mapping method when executing the computer program.

[0172] A storage medium stores a computer program, which implements the steps of a landscape garden surveying and mapping method when executed by a processor.

[0173] The high-performance hardware and flexible deployment capabilities of storage media of terminal devices transform surveying and mapping methods from the algorithm level to engineering tools, supporting garden surveying and mapping tasks with sub-centimeter accuracy and second-level response.

[0174] The terminal is adaptable to multiple air-ground terminals, and the media is compatible with the cloud-edge architecture, meeting the full-link requirements from data collection, dynamic modeling to decision output.

[0175] Compared with traditional surveying and mapping equipment, the terminal cost of the present invention is reduced by 80%, and continuous value-added is achieved through software-defined functions, which is in line with the large-scale application trend of smart city and ecological construction.

[0176] Through the collaborative design of "terminal + medium", innovative algorithms are solidified into mass-producible and scalable standardized products, solving the pain points of high cost, low real-time performance and weak adaptability in the field of landscape gardening surveying and mapping, and promoting the industry to upgrade to intelligence and universalization.

[0177] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A landscape garden surveying and mapping method based on artificial intelligence, characterized in that: include: Step 1: Use drones equipped with RGB cameras, lidar, and multispectral sensors to collect multimodal data of landscape gardens, and use ground mobile robots to supplement near-ground point cloud data; Step 2: The multimodal data collected in step 1 are aligned with the spatiotemporal reference using the ICP-SIFT hybrid registration algorithm to generate the registered multi-period data; Step 3: Based on the registration results of step 2, a spatiotemporal-spectral tensor is constructed, and the fused multimodal high-order correlation features are extracted through the tensor decomposition process constrained by the hypergraph Laplacian matrix; Step 4: Based on the fused features from step 3, a task relevance matrix is ​​constructed, and cross-scenario adaptive parameters are generated through a dynamic meta-optimizer driven by reinforcement learning. Step 5: Split the neural radiation field model into a static base network and a dynamic incremental network based on the parameters generated in step 4, and trigger the parameter update of the dynamic incremental network in combination with the KL divergence threshold; Step 6: The dynamic incremental network parameters updated in step 5 are merged in real time using a double buffer mechanism to generate stabilized 3D model parameters. Step 7: Output a layered semantic three-dimensional map based on the stable parameters of step 6, and generate an intelligent design auxiliary decision-making plan in combination with the dynamic vegetation health index.

2. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: In step 2, the ICP-SIFT hybrid registration algorithm further includes: Sub-step 2.1: Extract key point features of multiple RGB images using SIFT algorithm to generate feature point sets and Among them, f is the gradient direction histogram of the key point, F A is the set of SIFT feature points extracted from the A-th RGB image, F B is the set of SIFT feature points extracted from the B-th RGB image; Sub-step 2.2, calculate the cosine similarity between feature points: Keep the set M of matching pairs with similarity greater than 0.7 pairs ={(p i ,q j )}, in, is the cosine similarity between feature point pairs, is the SIFT feature descriptor of the i-th group of key points in the A-th image, is the SIFT feature descriptor of the jth group of key points in the Bth period image, M pairs is the set of matched point pairs after screening, p i is the three-dimensional coordinate in the image of period A, q j is the three-dimensional coordinate in the B-th phase image; Sub-step 2.3, based on the matching point pair M pairs , the rotation moment R and translation vector t are optimized by iterative closest point algorithm, and the objective function is: Where λ is the orthogonality constraint weight of the rotation matrix, I is the identity matrix, R is the three-dimensional rotation matrix, and t is the three-dimensional translation vector; Sub-step 2.4, calculate the comprehensive registration residual ∈: Among them, ∈ is the comprehensive registration residual; If ∈<1e-4 or the number of iterations exceeds 500, the optimization is terminated and the transformation matrices R and t are output.

3. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: In step 3, the process of constructing the spatiotemporal-spectral tensor and constraining the tensor decomposition further includes: Sub-step 3.1: Construct the spatiotemporal-spectral tensor based on the multi-period data after registration in step 2 Among them, H and W are the spatial dimension grid resolution, T is the number of time axis sampling points, C is the number of spectral channels, and D is the geometric attribute dimension; Sub-step 3.2: Generate a hyperedge set EcEc based on the ecological unit division. Each hyperedge connects the spatial grid points in the same ecological unit to construct a hypergraph adjacency matrix A. H and the hypergraph Laplacian matrix L H : L H =D H -A H , Among them, D H is the hyperedge degree matrix; Sub-step 3.3, optimize the objective function through hypergraph constrained tensor decomposition: in, is the core tensor, W is the dynamic weight matrix, λ1 and λ2 are regularization coefficients, and Tr is the trace of the matrix; Sub-step 3.4, reconstruct the fused feature tensor through factor matrix multiplication: Among them, A, B, C, and D are factor matrices, and F is the multimodal fusion feature tensor.

4. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: In step 4, constructing the task correlation matrix and generating cross-scenario adaptive parameters further includes: Sub-step 4.1, extract the multispectral histogram feature vector f based on the fusion feature of step 3 i and climate zone coding vector clim i , mapped to the task feature vector through the fully connected network: Among them, h i is the task feature vector; In sub-step 4.2, the task compatibility score Comp(i,j) is calculated based on the predefined values ​​in the plant symbiosis table, and the task correlation matrix M is constructed by combining the inner product of the eigenvectors: Among them, σ is the Sigmoid function, k is the slope coefficient, M i,j is the task dependency matrix element; Sub-step 4.3, use the proximal policy optimization algorithm to train the dynamic meta-optimizer policy network π θ , the input state is the task feature vector h k and M k , the output action space is the optimizer parameters: i k =(η,β), Among them, η is the learning rate, β is the momentum coefficient; Sub-step 4.4, based on the reward function r t Update the policy network parameters θ until convergence, the reward function r t The algorithm formula is: r t =0.7·IoU improve +0.3·Comp score , Among them, IoU improve To improve the accuracy of the model on the new task, Comp score is the mean of the task correlation matrix, r t is the reward function value.

5. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: In step 5, splitting the neural radiation field model and triggering parameter update further includes: Sub-step 5.1: Split the neural radiation field model into a static base network MLP static With dynamic incremental network MLP dyn , static base network parameters θ static Freeze, dynamically increment network parameters θ dyn Renewable; Sub-step 5.2, calculate the KL divergence threshold based on the historical incremental update data: τ=μ+3σ, Where τ is the KL divergence threshold, μ is the KL divergence mean, and σ is the standard deviation; Sub-step 5.3, real-time calculation of the new data distribution P new With the old data distribution P old KL divergence of: Among them, P old (x) is the probability value of the sample point x under the old data distribution, P new (x) is the probability value of the sample point x under the new data distribution, x is the sample point in the sample space, D KL (P new ∥P old ) is the Kullback-Leibler divergence; If D KL >τ, triggering dynamic incremental network parameter update; Sub-step 5.4, the dynamic incremental network generates volume density increments through the residual prediction module With color increment Δc: Among them, γ(·) is the position encoding function, t is the timestamp, d is the viewing direction vector, Initial volume density value predicted for the static base network.

6. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: In step 6, the step of merging the dynamic incremental network parameters using a double buffer mechanism further includes: Sub-step 6.1, initialize the double buffer parameter space: workspace θ work Receive the dynamic incremental network parameters θ updated in step 5 dyn , stable region θ stable Store verified parameters; Sub-step 6.2, the workspace updates dynamic parameters in real time: Among them, η is the dynamic learning rate, L photo is the loss of photometric consistency; Sub-step 6.3, calculate the rendering consistency of the workspace parameters at each time interval Δt = 10 minutes: Among them, MSE is the mean square error, N is the number of pixel samples rendered in the current frame, The parameter θ of the dynamic incremental network in the working area work The RGB color value of the i-th group of pixels rendered below, is the true color value of the i-th group of pixels collected by the RGB camera, ∥·∥ 2 is the square of the Euclidean distance; If MSE<∈(∈=1e-3), trigger parameter merging; Sub-step 6.4, set the workspace parameter θ work Replicate to stable region θ stable , and reset the workspace: i stable ←θ work ,the work ←θ stable , Among them, is the working area parameter of the dynamic incremental network, θ stable is the stability zone parameter.

7. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: The layered semantic three-dimensional map in step 7 includes vegetation type classification, terrain elevation model and artificial structure boundaries, and the vegetation health index is calculated through the time series change of NDVI.

8. The landscape gardening surveying and mapping method based on artificial intelligence according to claim 1, characterized in that: The intelligent design auxiliary decision-making solution includes illumination distribution integral calculation and optimal maintenance path planning, and the path planning is based on solving the traveling salesman problem.

9. A terminal device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the landscape garden surveying and mapping method according to any one of claims 1 to 8 are implemented.

10. A storage medium, characterized in that: A computer program is stored, and when the computer program is executed by a processor, the steps of the landscape garden surveying and mapping method according to any one of claims 1 to 8 are implemented.

Images