A land surveying method based on aerial image analysis

Abstract

The application discloses a land surveying method based on aerial image analysis and relates to the technical field of land surveying. The application solves phase jump error caused by sudden deformation of frozen soil through a quantum optimization framework, significantly improves the continuity of the deformation field, simultaneously fuses a classifier designed according to the temperature-elastic modulus attenuation law, enhances the recognition ability of the thermal driving deformation, combines an edge computing architecture with a parameter self-adaptive mechanism, realizes minute-level deformation early warning response, and further introduces a cross-modal fusion strategy to expand the monitoring dimension, effectively compensates for the scene limitation of a single sensor, and simultaneously supports a derivative application such as illegal building detection through a capsule network and a differentiable morphological operator.

Description

Technical Field

[0001] This invention relates to the field of land surveying technology, and in particular to a land surveying method based on aerial image analysis. Background Technology

[0002] In the permafrost regions of the plateau, road engineering faces the problem of periodic deformation caused by freeze-thaw cycles; the surface freezes and thaws repeatedly due to diurnal temperature differences and seasonal climate changes, resulting in non-uniform and abrupt settlement characteristics of the roadbed.

[0003] In recent years, temporal interferometric synthetic aperture radar (TS-InSAR) and three-dimensional laser scanning technology have become the mainstream monitoring methods in this region. TS-InSAR extracts millimeter-level deformation through multi-phase satellite imagery, while three-dimensional laser scanning can achieve centimeter-level surface modeling. The combination of the two provides dynamic deformation monitoring capabilities for permafrost engineering.

[0004] In practical applications, TS-InSAR relies on continuous phase unwrapping algorithms. Sudden deformation in permafrost regions causes the phase gradient to frequently exceed the 2π jump threshold, triggering a chain reaction of unwrapping path errors, ultimately resulting in fracture artifacts in the deformation field. Although 3D laser scanning can capture instantaneous deformation, it is limited by the harsh weather conditions at high altitudes, resulting in insufficient coverage in a single scan. In recent years, multi-source data fusion schemes have attempted to combine satellite, UAV, and ground sensor data, but the spatiotemporal alignment error of heterogeneous data still exceeds the magnitude of the deformation signal, and there is a lack of effective coupling mechanisms at the algorithm level. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a land surveying method based on aerial image analysis to solve the problems of deformation inversion distortion and response lag caused by the limitations of existing plateau permafrost monitoring technology, which is constrained by the accumulation of phase unwrapping errors, insufficient permafrost mechanical modeling, and sensor coverage defects.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a land surveying method based on aerial image analysis, comprising:

[0009] Step S1: Obtain a multi-temporal aerial image sequence of the target area, wherein the aerial images include visible light band and near-infrared band;

[0010] Step S2: Perform interferometric processing on the aerial image sequence to generate a differential interferometric phase map with millimeter-level precision;

[0011] Step S3: The quantum annealing algorithm is used to optimize the path of the abrupt phase gradient field in the differential interference phase diagram to obtain the continuously deformed phase field.

[0012] Step S4: Construct a deformation pattern classifier based on historical permafrost monitoring data and generate a mapping relationship library between freeze-thaw cycle parameters and deformation patterns;

[0013] Step S5: Perform spatiotemporal correlation analysis between the continuous deformation phase field and the mapping relationship library to output a road settlement risk level distribution map.

[0014] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, step S3 includes:

[0015] The phase gradient difference between adjacent pixels is mapped to the spin variable of the Ising model;

[0016] The optimal phase untangling path is searched using quantum tunneling effect simulation.

[0017] Terrain elevation data is introduced as a regularization constraint for path search.

[0018] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, in step S3, during the process of mapping adjacent phase gradient differences to Ising spin variables, in the pixel grid... Above, select any connecting pixel i With pixels j ordered edges ;

[0019] Normalizing the package phase difference using a 2π periodicity yields the normalized phase gradient difference:

[0020] ,

[0021] in, For the edge k The normalized phase gradient difference, P pi For pixels i The wrapping phase, P pj For example j The wrapping phase is π, and round(⋅) is the rounding function;

[0022] The sign of the normalized gradient difference is quantized into the Ising spin value:

[0023] ,

[0024] in, S sk For the edge k The spin variable takes the value +1 or -1. It is a unit step function;

[0025] Introducing a coupling coefficient to characterize the energy penalty for phase abrupt changes in the Hamiltonian, the coupling coefficient is expressed as:

[0026] ,

[0027] in, Let be the coupling coefficient of edge k. The inversion scale factor, This is the absolute value of the gradient difference. The standard deviation of the globally normalized gradient difference;

[0028] The inversion factor is calculated based on the quantum tunneling frequency and the annealing temperature.

[0029] ,

[0030] Where ℏ is the reduced Planck constant, F Ω The tunneling angular frequency, T is the Boltzmann constant. a This is the initial annealing temperature.

[0031] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, step S3 further includes:

[0032] An adaptive learning mechanism is established to dynamically adjust the temperature decay parameters of the quantum annealing algorithm based on real-time deformation monitoring data.

[0033] The execution process of the adaptive learning mechanism includes:

[0034] a) Construct a phase unwrapping quality evaluation function, whose input is the degree of agreement between the continuity index of the unwrapped phase field and the historical deformation data;

[0035] b) Optimize the temperature parameter control strategy using the gradient descent method;

[0036] c) When a freeze-thaw cycle mode switch is detected, the parameter reset module of the quantum annealing algorithm is automatically triggered.

[0037] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, the construction of the deformation pattern classifier in step S4 includes:

[0038] Extract the temperature-elastic modulus decay curve from the experimental data of frozen soil core samples;

[0039] A graph neural network model is established, whose node features include temperature time series, deformation rate and frozen soil type encoding;

[0040] The classification results are correlated with the real-time deformation phase field through the spatiotemporal alignment module.

[0041] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, step S4, the step of constructing a deformation pattern classifier includes:

[0042] In the common feature space of permafrost core sample experiments and aerial deformation data over the years, a nodal feature vector is constructed for a single sample n:

[0043] ,

[0044] in, This represents the time-series temperature mean of sample n, in °C. This represents the deformation rate of the corresponding pixel, in units of... , Indicates the unique thermal code for permafrost type, F n Let T be a column vector, and T be the transpose sign.

[0045] Connect all sample nodes to form a spacetime graph. In the l Updating hidden features in layered graph convolution:

[0046] ,

[0047] in, H l Let l be the hidden feature matrix of the l-th layer. For the normalized adjacency matrix, W gl Let l be the trainable weight matrix of the l-th layer. To modify the activation function of the linear unit;

[0048] When the convolution depth reaches L layers, a fully connected classification head is used to output the deformation pattern probability:

[0049] ,

[0050] Where P is the probability matrix, W O To output the weight matrix;

[0051] Time-aligned weighted cross-entropy is used as the loss function:

[0052] ,

[0053] in, For the total loss, For time alignment weights, This is the truth label for sample n, taking the value 0 or 1. To correspond to the predicted probability, m represents the category number;

[0054] In the formula:

[0055] ,

[0056] in, For temperature sensitivity coefficient, For real-time temperature monitoring, the unit is °C. The mean squared value of historical temperatures is expressed in °C.

[0057] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, step S5 includes:

[0058] Establish a dynamic correlation model between deformation rate, frozen soil elastic modulus and vehicle load pressure;

[0059] Integrate meteorological forecast data to predict deformation trends;

[0060] The edge computing device outputs emergency repair priority suggestions.

[0061] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, it further includes:

[0062] A capsule network architecture was used to extract building roof features from aerial images;

[0063] Optimize building outline boundaries using differentiable morphological operators;

[0064] Generate an illegal construction inspection report with three-dimensional coordinates.

[0065] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, the capsule network architecture includes:

[0066] Primary capsule layer, extracting local texture features;

[0067] The dynamic routing layer aggregates spatial correlation features;

[0068] The pose matrix transformation layer adjusts the feature mapping relationship according to the imaging height.

[0069] As a preferred embodiment of the land surveying method based on aerial image analysis described in this invention, it further includes:

[0070] Cross-modal fusion of LiDAR point cloud data and aerial imagery;

[0071] Aligning spatiotemporal benchmarks of multi-source data through attention mechanisms;

[0072] Generate a monitoring map that integrates surface deformation and underground temperature field.

[0073] The beneficial effects of this invention are as follows: This invention solves the phase jump error caused by abrupt deformation of permafrost through a quantum optimization framework, significantly improving the continuity of the deformation field; at the same time, it integrates a classifier design based on the temperature-elastic modulus decay law, enhancing the ability to identify thermally driven deformation; the combination of edge computing architecture and parameter adaptive mechanism achieves minute-level deformation early warning response; in addition, this invention introduces a cross-modal fusion strategy to expand the monitoring dimensions, effectively compensating for the limitations of single sensors in various scenarios; and the introduced capsule network and differentiable morphological operator simultaneously support derivative applications such as illegal construction detection. Attached Figure Description

[0074] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0075] Figure 1 This is a flowchart illustrating the land surveying method based on aerial image analysis in Example 1. Detailed Implementation

[0076] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0077] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0078] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0079] This invention proposes an intelligent land surveying method for monitoring road settlement in plateau permafrost regions. Through the integration of multidisciplinary technologies and innovative algorithm architecture, it systematically solves core problems faced by traditional aerial surveying in complex permafrost environments, such as phase unwrapping distortion, deformation mode misjudgment, and fragmented multi-source data. This invention uses quantum computing, graph neural networks, and cross-modal sensing as its core technological pillars, constructing a complete technology system encompassing data acquisition, feature optimization, and decision output. This achieves a breakthrough improvement in monitoring accuracy and efficiency while maintaining existing hardware conditions.

[0080] At the structural level, the solution employs a three-tiered collaborative processing framework: The front end deploys an array of drones equipped with multispectral cameras to collect centimeter-level surface deformation data; the mid-end utilizes quantum annealing algorithms and adaptive learning mechanisms to construct a dedicated unwrapping engine for permafrost abrupt deformation; and the back end fuses multi-source heterogeneous data through a spatiotemporal graph convolutional network to form a multi-dimensional monitoring map covering both the surface and subsurface. This architecture breaks through the traditional serial processing mode by optimizing parallel feature extraction and distributed computing.

[0081] Example 1, referring to Figure 1 This embodiment provides a land surveying method based on aerial image analysis, including the following steps:

[0082] Step S1: Obtain a multi-temporal aerial image sequence of the target area. The aerial images include the visible light band and the near-infrared band.

[0083] Step S2: Perform interferometric processing on the aerial image sequence to generate a differential interferometric phase map with millimeter-level precision;

[0084] Step S3: The quantum annealing algorithm is used to optimize the path of the abrupt phase gradient field in the differential interferometric phase diagram to obtain the continuously deformed phase field.

[0085] Step S3 includes:

[0086] The phase gradient difference between adjacent pixels is mapped to the spin variable of the Ising model;

[0087] The optimal phase untangling path is searched using quantum tunneling effect simulation.

[0088] Terrain elevation data is introduced as a regularization constraint for path search;

[0089] In step S3, during the process of mapping adjacent phase gradient differences to Ising spin variables, in the pixel grid... Choose any ordered edge connecting pixel i and pixel j. ;

[0090] Normalizing the package phase difference using a 2π periodicity yields the normalized phase gradient difference:

[0091] ,

[0092] in, Let k be the normalized phase gradient difference. P pi For pixels i The wrapping phase, P pj For pixels j The phase of the envelope, where π is the mathematical constant pi. round(⋅) It is a rounding function;

[0093] The sign of the normalized gradient difference is quantized into the Ising spin value:

[0094] ,

[0095] in, Let be the spin variable of edge k, taking the value +1 or -1. It is a unit step function;

[0096] Introducing a coupling coefficient to characterize the energy penalty for phase abrupt changes in the Hamiltonian, the coupling coefficient is expressed as:

[0097] ,

[0098] in, Let Q be the coupling coefficient of edge k. b The inversion scale factor, This is the absolute value of the gradient difference. The standard deviation of the globally normalized gradient difference;

[0099] The inversion factor is calculated based on the quantum tunneling frequency and the annealing temperature.

[0100] ,

[0101] Where ℏ is the reduced Planck constant, F Ω K is the tunneling angular frequency. B T is the Boltzmann constant. a This is the initial annealing temperature;

[0102] Specifically, here, periodic normalization eliminates the interference of 2π phase jumps on gradient calculation, ensuring that the gradient field is consistent with the actual deformation. Symbolic quantization transforms the phase untangling problem into a spin-flipping task of the Ising model, greatly reducing the search dimension. The coupling coefficient measures the intensity of local mutations by the ratio of gradient magnitude to global standard deviation, with high-intensity edges receiving greater energy penalties. Severe jumps are corrected first during quantum annealing. The inversion factor is expanded twice by tunneling frequency and annealing temperature. Hardware-adjustable parameters are introduced, which can adaptively adjust the system energy scale according to environmental noise, thereby achieving a balance between coherent deformation recovery and noise suppression, and enhancing the reliability of the continuous deformation phase field.

[0103] Step S3 also includes:

[0104] An adaptive learning mechanism is established to dynamically adjust the temperature decay parameters of the quantum annealing algorithm based on real-time deformation monitoring data.

[0105] The execution process of the adaptive learning mechanism includes:

[0106] a) Construct a phase unwrapping quality evaluation function, whose input is the degree of agreement between the continuity index of the unwrapped phase field and the historical deformation data;

[0107] b) Optimize the temperature parameter control strategy using the gradient descent method;

[0108] c) When a freeze-thaw cycle mode switch is detected, the parameter reset module of the quantum annealing algorithm is automatically triggered;

[0109] Step S4: Construct a deformation pattern classifier based on historical permafrost monitoring data and generate a mapping relationship library between freeze-thaw cycle parameters and deformation patterns;

[0110] The construction of the deformation pattern classifier in step S4 includes:

[0111] Extract the temperature-elastic modulus decay curve from the experimental data of frozen soil core samples;

[0112] A graph neural network model is established, whose node features include temperature time series, deformation rate and frozen soil type encoding;

[0113] The classification results are correlated with the real-time deformation phase field through the spatiotemporal alignment module;

[0114] In step S4, the steps for constructing the deformation pattern classifier include:

[0115] In the common feature space of permafrost core sample experiments and aerial deformation data over the years, a nodal feature vector is constructed for a single sample n:

[0116] ,

[0117] in, This represents the time-series temperature mean of sample n, in °C. This represents the deformation rate of the corresponding pixel, in units of... , Indicates the unique thermal code for permafrost type, F n Let T be a column vector, and T be the transpose sign.

[0118] Connect all sample nodes to form a spacetime graph. In the l Updating hidden features in layered graph convolution:

[0119] ,

[0120] in, This is the first layer hidden feature matrix. For the normalized adjacency matrix, W gl For the first l Layer trainable weight matrix, To modify the activation function of the linear unit;

[0121] When the convolution depth reaches L layers, a fully connected classification head is used to output the deformation pattern probability:

[0122] ,

[0123] in, P c Let W be a probability matrix. O To output the weight matrix;

[0124] Time-aligned weighted cross-entropy is used as the loss function:

[0125] ,

[0126] in, For the total loss, For time alignment weights, This is the truth label for sample n, taking the value 0 or 1. To correspond to the predicted probability, m represents the category number;

[0127] In the formula:

[0128] ,

[0129] in, a t For temperature sensitivity coefficient, For real-time temperature monitoring, the unit is °C. The historical temperature mean square deviation is expressed in °C.

[0130] Specifically, this step integrates temperature, deformation rate, and permafrost category into node features, captures landform continuity and seasonal cycle effects with a spatiotemporal adjacency matrix, and uses graph convolution to propagate information within the local topology through a multi-layer weight sharing mechanism to improve the ability to distinguish heterogeneous permafrost zones. The Softmax head provides multi-class probability outputs that can be directly mapped to deformation pattern indices. Temperature alignment weights are embedded in the weighted cross-entropy, amplifying the loss when the real-time temperature deviates from the historical mean. This encourages the model to focus on rare deformation patterns under extreme temperatures during training, taking into account both thermal driving and deformation dynamics, and enhancing the generalization performance of the classifier under different seasons and soil conditions.

[0131] Step S5: Perform spatiotemporal correlation analysis between the continuous deformation phase field and the mapping relationship library to output a road settlement risk level distribution map;

[0132] Step S5 includes:

[0133] Establish a dynamic correlation model between deformation rate, frozen soil elastic modulus and vehicle load pressure;

[0134] Integrate meteorological forecast data to predict deformation trends;

[0135] The edge computing device outputs emergency repair priority suggestions;

[0136] Land surveying methods also include:

[0137] A capsule network architecture was used to extract building roof features from aerial images;

[0138] Optimize building outline boundaries using differentiable morphological operators;

[0139] Generate an illegal construction detection report with three-dimensional coordinates;

[0140] The capsule network architecture includes:

[0141] Primary capsule layer, extracting local texture features;

[0142] The dynamic routing layer aggregates spatial correlation features;

[0143] The pose matrix transformation layer adjusts the feature mapping relationship according to the imaging height;

[0144] Cross-modal fusion of LiDAR point cloud data and aerial imagery;

[0145] Aligning spatiotemporal benchmarks of multi-source data through attention mechanisms;

[0146] Generate a monitoring map that integrates surface deformation and underground temperature field.

[0147] In summary, this invention achieves three major technological breakthroughs:

[0148] 1. Quantum-optimized phase untangling mechanism: By creatively introducing the Ising model and quantum tunneling effect, the phase untangling problem is transformed into a task of minimizing the energy of the spin system. Through normalized gradient processing and dynamic adjustment of coupling coefficient, the phase breakage caused by ±8cm deformation jumps commonly seen in permafrost regions is effectively suppressed.

[0149] 2. Embedded Modeling of Frozen Soil Mechanics: A spatiotemporal graph neural network classifier was developed to fuse frozen soil core sample experimental data with aerial deformation fields at the feature level. By modeling the temperature-elastic modulus decay curve and designing a spatiotemporally aligned weighted loss function, accurate distinction between thermodynamically driven deformation and load-driven deformation was achieved, addressing the shortcomings of existing methods in modeling frozen soil properties.

[0150] 3. Cross-modal dynamic perception system: A collaborative architecture of capsule network and differentiable morphological operator is designed to simultaneously realize road settlement monitoring and illegal construction detection in a single aerial data stream. By using an attitude matrix transformation layer and a multi-view photon transmission model, the limitations of vegetation occlusion and perspective distortion on feature extraction are overcome, expanding the applicability of the solution in complex urban scenarios.

[0151] This solution, through original algorithmic innovation, breaks through the inherent bottlenecks of traditional measurement technologies limited by the performance of physical sensors, providing a new generation of technological paradigms for engineering safety monitoring in high-altitude and high-risk areas. Its value lies not only in the order-of-magnitude improvement in monitoring accuracy, but also in building an intelligent analysis bridge connecting the microscopic mechanical properties of permafrost with macroscopic surface deformation, providing reliable decision-making support for infrastructure construction and ecological protection in permafrost regions.

[0152] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A land surveying method based on aerial image analysis, characterized in that, include, Step S1: Obtain a multi-temporal aerial image sequence of the target area, wherein the aerial images include visible light band and near-infrared band; Step S2: Perform interferometric processing on the aerial image sequence to generate a differential interferometric phase map with millimeter-level precision; Step S3: The quantum annealing algorithm is used to optimize the path of the abrupt phase gradient field in the differential interference phase diagram to obtain the continuously deformed phase field. Step S4: Construct a deformation pattern classifier based on historical permafrost monitoring data and generate a mapping relationship library between freeze-thaw cycle parameters and deformation patterns; Step S5: Perform spatiotemporal correlation analysis between the continuous deformation phase field and the mapping relationship library to output a road settlement risk level distribution map; Step S3 includes: The phase gradient difference between adjacent pixels is mapped to the spin variable of the Ising model; The optimal phase untangling path is searched using quantum tunneling effect simulation. Terrain elevation data is introduced as a regularization constraint for path search.

2. The land surveying method based on aerial image analysis as described in claim 1, characterized in that, In step S3, during the process of mapping adjacent phase gradient differences to Ising spin variables, in the pixel grid... Above, select any connecting pixel With pixels ordered edges ; Perform phase difference analysis on package Periodic normalization yields the normalized phase gradient difference: ， in, For the edge The normalized phase gradient difference, For pixels The wrapping phase, For pixels The wrapping phase, Pi It is a rounding function; The sign of the normalized gradient difference is quantized into the Ising spin value: ， in, Let k be the spin variable of edge k, taking the value +1 or , It is a unit step function; Introducing a coupling coefficient to characterize the energy penalty for phase abrupt changes in the Hamiltonian, the coupling coefficient is expressed as: ， in, For the edge The coupling coefficient, The inversion scale factor, This is the absolute value of the gradient difference. The standard deviation of the globally normalized gradient difference; The inversion factor is calculated based on the quantum tunneling frequency and the annealing temperature. ， in, To reduce Planck's constant, The tunneling angular frequency, Boltzmann's constant, This is the initial annealing temperature.

3. The land surveying method based on aerial image analysis as described in claim 2, characterized in that, Step S3 also includes: An adaptive learning mechanism is established to dynamically adjust the temperature decay parameters of the quantum annealing algorithm based on real-time deformation monitoring data. The execution process of the adaptive learning mechanism includes: a) Construct a phase unwrapping quality evaluation function, whose input is the degree of agreement between the continuity index of the unwrapped phase field and the historical deformation data; b) Optimize the temperature parameter control strategy using the gradient descent method; c) When a freeze-thaw cycle mode switch is detected, the parameter reset module of the quantum annealing algorithm is automatically triggered.

4. The land surveying method based on aerial image analysis as described in claim 1, characterized in that, The construction of the deformation pattern classifier in step S4 includes: Extract the temperature-elastic modulus decay curve from the experimental data of frozen soil core samples; A graph neural network model is established, whose node features include temperature time series, deformation rate and frozen soil type encoding; The classification results are correlated with the real-time deformation phase field through the spatiotemporal alignment module.

5. The land surveying method based on aerial image analysis as described in claim 4, characterized in that, In step S4, the steps for constructing the deformation pattern classifier include: Within the common feature space of permafrost core sample experiments and aerial deformation data over the years, for a single sample Construct node eigenvectors: ， in, Indicates sample The time-series average temperature, in °C. This represents the deformation rate of the corresponding pixel, in units of... , Indicates the unique thermal code for permafrost type. It is a column vector. It is the transpose symbol; Connect all sample nodes to form a spacetime graph. In the Updating hidden features in layered graph convolution: ， in, For the first Hidden feature matrix of layer, For the normalized adjacency matrix, For the first Layer trainable weight matrix, To modify the activation function of the linear unit; When the convolution depth reaches L layers, a fully connected classification head is used to output the deformation pattern probability: ， in, It is a probability matrix. To output the weight matrix; Time-aligned weighted cross-entropy is used as the loss function: ， in, For the total loss, For time alignment weights, For the sample The truth label takes the value 0 or 1. For the corresponding predicted probability, Indicates the category number; In the formula: ， in, For temperature sensitivity coefficient, For real-time temperature monitoring, the unit is °C. The mean squared value of historical temperatures is expressed in °C.

6. The land surveying method based on aerial image analysis as described in claim 1, characterized in that, Step S5 includes: Establish a dynamic correlation model between deformation rate, frozen soil elastic modulus and vehicle load pressure; Integrate meteorological forecast data to predict deformation trends; The edge computing device outputs emergency repair priority suggestions.

7. The land surveying method based on aerial image analysis as described in claim 6, characterized in that, Also includes: A capsule network architecture was used to extract building roof features from aerial images; Optimize building outline boundaries using differentiable morphological operators; Generate an illegal construction inspection report with three-dimensional coordinates.

8. The land surveying method based on aerial image analysis as described in claim 7, characterized in that, The capsule network architecture includes: Primary capsule layer, extracting local texture features; The dynamic routing layer aggregates spatial correlation features; The pose matrix transformation layer adjusts the feature mapping relationship according to the imaging height.

9. A land surveying method based on aerial image analysis as described in claim 1, characterized in that, Also includes: Cross-modal fusion of LiDAR point cloud data and aerial imagery; Aligning spatiotemporal benchmarks of multi-source data through attention mechanisms; Generate a monitoring map that integrates surface deformation and underground temperature field.

Images