A high-precision extraction system and method for remote sensing three-dimensional terrain features

Through the Hidden Markov model and deep learning technology, the model parameters are dynamically updated, combined with Viterbi algorithm and semantic segmentation, the problem of untimely update of terrain data in the existing technology is solved, real-time monitoring and high-precision prediction of terrain changes are realized, and real-time monitoring and high-precision prediction of terrain changes are improved. Data parsing capabilities and response speed are found.

CN118691986BActive Publication Date: 2024-11-08JIANGXI NUCLEAR IND SURVEYING & MAPPING INST GRP CO LTD +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411199747.4
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-29
Publication Date
2024-11-08
Estimated Expiration
2044-08-29

AI Technical Summary

Technical Problem

When processing rapidly changing terrain data, the parameter update and status prediction are not flexible enough, resulting in data processing and update in an emergency situation after the disaster, and the inability to monitor terrain changes in real time, affecting decision-making effects and safety management.

Method used

The hidden Markov model and deep learning technology are used to recognize the terrain state through satellite remote sensing images, update the model parameters dynamically, and process observation data using the Viterbi algorithm to predict the terrain change trend, and extract the three-dimensional terrain feature boundaries through semantic segmentation.

Benefits of technology

Real-time monitoring and high-precision prediction of terrain changes are realized, the availability and analysis capabilities of terrain data are improved, the response speed to complex terrain changes is improved, and more accurate and real-time data support is provided.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118691986B_ABST
    Figure CN118691986B_ABST
Patent Text Reader

Abstract

The present invention relates to the field of three-dimensional image processing technology, specifically to a high-precision extraction system and method for remote sensing and mapping three-dimensional terrain features, the system comprising a terrain state definition module, a terrain dynamic probability model module, a terrain change sequence decoding module, and a three-dimensional terrain feature boundary extraction module. In the present invention, by initializing a hidden Markov model and setting an initial state distribution, the dynamic monitoring capability of terrain feature extraction in remote sensing images is improved, in the terrain dynamic probability model, new remote sensing images are continuously captured and feature data are analyzed in real time, model parameters are updated in real time, monitoring and prediction accuracy are optimized, the Viterbi algorithm is applied to process observation data, the terrain state change sequence is accurately identified, and future change trends are effectively predicted, deep learning is used for semantic segmentation of terrain features, the availability and accuracy of data are improved, the response speed to complex terrain changes is enhanced, and accurate and real-time data support is provided for fields such as disaster management and environmental monitoring.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of three-dimensional image processing, and in particular to a high-precision extraction system and method for remote sensing and mapping of three-dimensional terrain features. Background Art

[0002] The field of three-dimensional image processing technology involves methods and systems for generating, processing and analyzing three-dimensional data from two-dimensional images or image sequences. The core technologies in this field include three-dimensional reconstruction, disparity map calculation, stereo matching and depth estimation. By using advanced algorithms and computing techniques, three-dimensional image processing can provide accurate three-dimensional visual representations for virtual reality, game development, film production, and industrial and medical applications. What is particularly important about this technology is that it can process and analyze data from a variety of sensors (such as stereo cameras, laser scanners, radars), thereby creating detailed three-dimensional models and terrain maps.

[0003] Among them, the high-precision extraction system for remote sensing mapping of three-dimensional terrain features is an advanced system specially designed to extract terrain features from remote sensing data. The core of the theme is to use remote sensing technology, such as satellite images and aerial photography, to accurately map the surface through three-dimensional image processing technology. The main uses of the system include environmental monitoring, urban planning, disaster management, and updating of geographic information systems (GIS). This system enables decision makers to better understand the surface conditions and terrain changes by providing high-precision and high-resolution terrain information, so as to make more informed decisions in various application fields.

[0004] Although existing technologies can support the analysis of multiple sensor data and the creation of three-dimensional models, their parameter updates and state predictions are usually not flexible enough when processing rapidly changing terrain data, resulting in insufficient timeliness in data processing and updating in emergency situations after disasters. When extracting terrain features from remote sensing data, existing technologies rely on static algorithms and models and lack the ability to dynamically monitor terrain changes. This static processing method is difficult to adapt to actual application scenarios with frequent environmental changes and high data update speed requirements. For example, in urban planning and disaster management, the real-time analysis and accurate prediction of terrain data are extremely critical. Existing technologies fail to provide sufficient support, resulting in delayed or inaccurate decision-making, which in turn affects the final application effect and safety management. Summary of the invention

[0005] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a high-precision extraction system and method for remote sensing three-dimensional terrain features.

[0006] In order to achieve the above-mentioned purpose, the present invention adopts the following technical scheme: A high-precision extraction system for remote sensing and mapping three-dimensional terrain features comprises:

[0007] The terrain state definition module identifies normal, warning, and post-disaster terrain states based on satellite remote sensing images, associates each state with a set of terrain features, obtains state definition data, initializes the hidden Markov model, sets the initial state distribution, and obtains the initialization probability matrix;

[0008] The terrain dynamic probability model module captures new remote sensing images based on the initialization probability matrix, calculates state transition and observation probabilities through terrain feature data, generates a dynamic probability model, and uses the dynamic probability model to update the parameters of the hidden Markov model to obtain updated model parameters;

[0009] The terrain change sequence decoding module applies the Viterbi algorithm to process the continuous observation data based on the updated model parameters, identifies the terrain state change sequence, generates the state change result, and predicts the terrain change trend in the future time period according to the state change result to obtain the terrain prediction analysis result;

[0010] The three-dimensional terrain feature boundary extraction module performs deep learning training of remote sensing images based on the terrain prediction analysis results, performs semantic segmentation on terrain features, generates a terrain feature distinction framework, applies the terrain feature distinction framework, extracts terrain feature boundaries, and obtains three-dimensional terrain feature boundary extraction data.

[0011] As a further solution of the present invention, the step of obtaining the initialization probability matrix is ​​specifically as follows:

[0012] Based on satellite remote sensing images, identify terrain states and associate a set of terrain features with each state using the formula:

[0013]

[0014] Get the state definition data, where It is The feature matrix of each state, It is The terrain feature weight of each state, Indicates state definition data, Indicates the total number of terrain states;

[0015] Use the state definition data and initialize the hidden Markov model, set the initial state distribution, and apply the formula:

[0016]

[0017] Generate the initialization state probability, where represents the initialization state probability, represents the normalization constant, represents the adjustment parameter, Represents state definition data;

[0018] According to the initialization state probability, a probability matrix is ​​constructed, and the probability matrix after averaging is calculated using the formula:

[0019]

[0020] Generate an initialization probability matrix, where represents the initialization probability matrix, From The initial state probability of the state, is the total number of states.

[0021] As a further solution of the present invention, the step of acquiring the dynamic probability model is specifically as follows:

[0022] Based on the initialization probability matrix, capture a new remote sensing image, pass the terrain feature data, and apply the formula:

[0023]

[0024] Generate the initial state transition probability, where represents the initial state transition probability, Indicates The terrain feature data of each state, represents the initialization probability matrix, Indicates adjustment parameters;

[0025] Based on the initial state transition probability, the regional feature information is updated using the new remote sensing image data, and the updated feature information is combined with the transition probability matrix using the formula:

[0026]

[0027] Get the initial observation probability, where represents the initial observation probability, represents the initial state transition probability, represents remote sensing image data, represents the normalization factor;

[0028] Combine the initial observation probability and the initial state transition probability, perform normalization, and strengthen the model's response to the state probability, using the formula:

[0029]

[0030] Generate a dynamic probability model where represents a dynamic probability model, represents the initial observation probability, represents the initial state transition probability.

[0031] As a further solution of the present invention, the step of obtaining the updated model parameters is specifically as follows:

[0032] According to the dynamic probability model, the state transition probability of the hidden Markov model is updated, referring to the complexity and adaptability of the model, using the formula:

[0033]

[0034] Generate updated state transition probabilities, where represents a dynamic probability model, represents the adjustment factor, is the normalization constant, is the updated state transition probability;

[0035] Combine the updated state transition probability and the current remote sensing data to update the initial observation probability, using the formula:

[0036]

[0037] Get the updated observation probability, where is the updated state transition probability, For new remote sensing data, and To adjust the parameters, is the updated observation probability;

[0038] Combining the updated observation probability and the updated state transition probability, the model parameters are re-estimated using the formula:

[0039]

[0040] Get the updated model parameters, where and are the updated state transition probability and observation probability, respectively. represents the weight of the state transition parameter, represents the weight of the observed parameter, is the normalization coefficient, Represents the updated model parameters.

[0041] As a further solution of the present invention, the step of obtaining the state change result is specifically:

[0042] According to the updated model parameters, the observed data is processed using the Viterbi algorithm, and the observed data and the updated parameters are combined using the formula:

[0043]

[0044] Generate initial path probabilities, where are the updated model parameters, is the observed data, is the initialization path probability;

[0045] Based on the initial path probability, the observation data is captured at each time point, and the path probabilities at multiple time points are gradually calculated and updated according to the current state transition probability and observation probability, applying the formula:

[0046]

[0047] Get the updated path probability, where is the path probability at the previous time point, For time point The model parameters, is the normalization constant, is the updated path probability;

[0048] Based on the updated path probabilities, the state change sequence is determined according to the optimal path probability calculated in the Viterbi algorithm, using the formula:

[0049]

[0050] By selecting the path with high probability, the state change result is determined, where is the state change result, is the updated path probability.

[0051] As a further solution of the present invention, the steps of obtaining the terrain prediction analysis results are specifically as follows:

[0052] According to the state change results, the monitoring data provides the historical height change information of each terrain unit, through the formula:

[0053]

[0054] Get the rate of change of each terrain unit, where represents the height change of terrain cells, represents the time interval, Indicates the rate of change of each terrain unit;

[0055] According to the change rate of each terrain unit and the environmental factors, the terrain adjustment coefficient is calculated by the formula:

[0056]

[0057] Get the adjusted rate of change, where Represents the rate of change of each terrain unit, Represents environmental factors, is a constant, represents the adjusted rate of change;

[0058] The adjusted rate of change is applied to the prediction model to analyze the terrain height in the future time period, through the formula:

[0059]

[0060] Generate terrain prediction analysis results, where Represents the current terrain height, represents the adjusted rate of change, represents the prediction time length, Represents the results of terrain prediction analysis.

[0061] As a further solution of the present invention, the step of acquiring the three-dimensional terrain feature boundary extraction data is specifically as follows:

[0062] According to the terrain prediction analysis results, key frame images are extracted from the remote sensing data set to perform preliminary terrain feature marking, using the formula:

[0063]

[0064] Generate a preliminary feature-labeled image set, where represents the preliminary feature labeled image set, represents a single remote sensing image as input, represents the critical area applied to the target terrain features, Indicates the total number of images;

[0065] Based on the preliminary feature-labeled image set, a deep neural network model is applied to learn and analyze the images using the formula:

[0066]

[0067] Generate a set of feature vectors, where represents the weights of the neural network layer, Indicates bias, represents the base of natural logarithm, represents the set of feature vectors, Indicates A preliminary feature labeling image set of images;

[0068] Based on the feature vector set, quantitative analysis is performed to distinguish differentiated terrain features using the formula:

[0069]

[0070] Get three-dimensional terrain feature boundary extraction data, where: represents a single feature vector extracted from the feature vector set, represents the total number of eigenvectors, represents the threshold used for terrain classification, Represents three-dimensional terrain feature boundary extraction data.

[0071] A high-precision extraction method for remote sensing and mapping three-dimensional terrain features, the high-precision extraction method for remote sensing and mapping three-dimensional terrain features is performed based on the high-precision extraction system for remote sensing and mapping three-dimensional terrain features, and comprises the following steps:

[0072] S1: Based on satellite remote sensing images, the identified images are analyzed to identify and classify the terrain status into three categories: normal, warning, and disaster. The initial state distribution of the hidden Markov model is set to obtain the initialization probability matrix.

[0073] S2: According to the initialization probability matrix, capture new remote sensing images, extract terrain feature data, update the state transition probability and observation probability of the hidden Markov model, and obtain updated model parameters;

[0074] S3: using the updated model parameters, processing the continuous remote sensing image data, identifying the terrain state change sequence through the Viterbi algorithm, parsing and recording the terrain state transition, predicting the terrain change trend in the future time period, and generating terrain prediction analysis results;

[0075] S4: Based on the terrain prediction and analysis results, execute a deep learning algorithm to process the remote sensing image, annotate terrain features through semantic segmentation, enhance the recognition accuracy of terrain features, and obtain a terrain feature distinction framework;

[0076] S5: Apply the terrain feature differentiation framework to analyze the newly captured remote sensing image, extract terrain feature boundaries, record and organize boundary data, and generate three-dimensional terrain feature boundary extraction data.

[0077] Compared with the prior art, the advantages and positive effects of the present invention are:

[0078] In the present invention, by initializing the hidden Markov model and setting the initial state distribution, the dynamic detection capability of extracting terrain features from remote sensing images is improved. In the terrain dynamic probability model, the continuous capture of new remote sensing images and the real-time analysis of feature data allow the system to update the model parameters in real time. This dynamic adjustment method optimizes the monitoring and prediction accuracy of terrain status. By using the Viterbi algorithm to process the observed data, the system can more accurately identify the sequence of terrain state changes, thereby effectively predicting future terrain change trends. Deep learning is used to perform semantic segmentation on terrain features, and three-dimensional terrain feature boundaries are extracted therefrom, which significantly improves the availability and accuracy of terrain data, enhances the parsing capability of terrain data and the response speed to complex terrain changes, and provides more accurate and real-time data support for disaster management, environmental monitoring and other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0079] Figure 1 is a system flow chart of the present invention;

[0080] Figure 2 This is a flow chart of initializing the probability matrix in the present invention;

[0081] Figure 3 is a flow chart of the dynamic probability model in the present invention;

[0082] Figure 4 is a flow chart of the updated model parameters in the present invention;

[0083] Figure 5 It is a flow chart of the state change result in the present invention;

[0084] Figure 6 A flow chart showing the terrain prediction analysis results in the present invention;

[0085] Figure 7 The flowchart of the three-dimensional terrain feature boundary extraction data in the present invention. DETAILED DESCRIPTION

[0086] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0087] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a target position, be constructed and operated in the target position, and therefore cannot be understood as a limitation on the present invention. In addition, in the description of the present invention, "multiple" means two or more, unless otherwise clearly and specifically defined.

[0088] See also Figure 1 The present invention provides a technical solution: a high-precision extraction system for remote sensing and mapping three-dimensional terrain features includes:

[0089] The terrain state definition module identifies normal, warning, and post-disaster terrain states based on satellite remote sensing images, associates each state with a set of terrain features, obtains state definition data, initializes the hidden Markov model, sets the initial state distribution, and obtains the initialization probability matrix;

[0090] The terrain dynamic probability model module captures new remote sensing images based on the initialization probability matrix, calculates the state transition and observation probability through terrain feature data, generates a dynamic probability model, uses the dynamic probability model to update the parameters of the hidden Markov model, and obtains the updated model parameters;

[0091] The terrain change sequence decoding module uses the Viterbi algorithm to process continuous observation data based on the updated model parameters, identifies the terrain state change sequence, generates state change results, and predicts the terrain change trend in the future time period based on the state change results to obtain terrain prediction analysis results;

[0092] The three-dimensional terrain feature boundary extraction module conducts deep learning training of remote sensing images based on the terrain prediction analysis results, performs semantic segmentation of terrain features, generates a terrain feature differentiation framework, applies the terrain feature differentiation framework, extracts terrain feature boundaries, and obtains three-dimensional terrain feature boundary extraction data.

[0093] The initialization probability matrix includes the initial state probability, state transition probability, and observation probability. The state definition data includes the terrain state label, terrain feature set, and state correlation. The dynamic probability model includes the state transition matrix, observation data probability, and model update frequency. The updated model parameters include the corrected state probability, updated observation probability, and adjusted state transition probability. The state change results include the state path, state change identification, and state change timeline. The terrain prediction analysis results include future terrain change trends, change intensity, and potential impact areas. The segmentation model includes the learning hierarchy, feature recognition accuracy, and model application scope. The three-dimensional terrain feature boundary extraction data includes boundary coordinate data, feature recognition details, and boundary accuracy evaluation.

[0094] See also Figure 2 , the specific steps to obtain the initialization probability matrix are:

[0095] Based on satellite remote sensing images, identify terrain states and associate a set of terrain features with each state using the formula:

[0096]

[0097] Get the state definition data, where It is The feature matrix of each state, It is The terrain feature weight of each state, Indicates state definition data, Indicates the total number of terrain states;

[0098] Use the state definition data and initialize the hidden Markov model, set the initial state distribution, and apply the formula:

[0099]

[0100] Generate the initialization state probability, where represents the initialization state probability, represents the normalization constant, represents the adjustment parameter, Represents state definition data;

[0101] According to the initialization state probability, the probability matrix is ​​constructed and the probability matrix after averaging is calculated using the formula:

[0102]

[0103] Generate an initialization probability matrix, where represents the initialization probability matrix, From The initial state probability of the state, is the total number of states.

[0104] State definition data formula:

[0105]

[0106] : It is a topographic feature The characteristic matrix represents the observation value or characteristic quantity of the target terrain;

[0107] : Terrain features The weights are derived through expert evaluation or historical data analysis;

[0108] : State definition data, which is a scalar that integrates all relevant terrain features and corresponding weights;

[0109] There are three types of terrain features , obtained through remote sensing image monitoring, such as (Rock coverage percentage), (terrain slope), (vegetation density), weight They were set to 0.5, 0.3, and 0.2, respectively, to quantitatively assess the importance of topography on ecological impact;

[0110] calculate :

[0111]

[0112] this The value represents the terrain state The comprehensive evaluation score is used for subsequent model parameterization.

[0113] Initialization state probability formula:

[0114]

[0115] : Normalization constant, ensuring that the probability sum is 1, by normalization All values ​​of are calculated;

[0116] : System adjustment parameters, affecting the sensitivity of the initial state probability distribution;

[0117] : Initial state distribution probability;

[0118] set up , The value passed in from the previous step is 70.5. To simplify the calculation, (a reasonable normalization constant);

[0119] calculate :

[0120]

[0121] The value represents the status as the probability of the initial state of the hidden Markov model.

[0122] Initialization probability matrix formula:

[0123]

[0124] : The probability component of each state calculated in the previous step;

[0125] : Number of states;

[0126] : Initialize the probability matrix;

[0127] There are three states, and the probability of each state is The result calculated from the previous step As a sample;

[0128] calculate :

[0129]

[0130]

[0131] The value of reflects the average information entropy of the probability matrix in the initial state, which plays an important role in the state transition prediction of the hidden Markov model.

[0132] See also Figure 3 , the specific steps for obtaining the dynamic probability model are:

[0133] Based on the initialization probability matrix, capture new remote sensing images, pass terrain feature data, and apply the formula:

[0134]

[0135] Generate the initial state transition probability, where represents the initial state transition probability, Indicates The terrain feature data of each state, represents the initialization probability matrix, Indicates adjustment parameters;

[0136] Based on the initial state transition probability, the regional feature information is updated using the new remote sensing image data, and the updated feature information is combined with the transition probability matrix using the formula:

[0137]

[0138] Get the initial observation probability, where represents the initial observation probability, represents the initial state transition probability, represents remote sensing image data, represents the normalization factor;

[0139] Combine the initial observation probability and the initial state transition probability, perform normalization, and strengthen the model's response to the state probability. Use the formula:

[0140]

[0141] Generate a dynamic probability model where represents a dynamic probability model, represents the initial observation probability, represents the initial state transition probability.

[0142] The formula for the initial state transition probability is:

[0143]

[0144] : Terrain feature data, which can be a terrain classification index obtained through remote sensing images, and the specific value is set to ;

[0145] : Initialize the probability matrix, for example, obtained by analyzing historical terrain data and set to 0.8;

[0146] : Adjust the parameters, and set the optimal parameters obtained through actual model debugging to 0.5;

[0147] : The number of states is set to the number of actual terrain classifications, which is set to 3;

[0148] calculate Values: ;

[0149] For each (from 1 to ),calculate , here only gives Calculation: ;

[0150] Calculate the denominator (Set another ) ;

[0151] calculate :

[0152]

[0153] Calculation results The transition probability of the first state is approximately , which represents the probability of transitioning from the current state to the next state.

[0154] The formula for the initial observation probability is:

[0155]

[0156] : Calculated from the previous step, take 0.333;

[0157] : New remote sensing image data, captured by remote sensing technology, set to 1.2;

[0158] : Normalization factor, ensuring the sum is 1, set to 0.7 through system testing;

[0159] calculate : ;

[0160] Calculate the denominator (set the three states similar): ;

[0161] calculate :

[0162]

[0163] Observation probability Indicates that under a given terrain state, the probability of observing a new image is approximately .

[0164] The formula of dynamic probability model is:

[0165]

[0166] calculate : ;

[0167] Take the maximum value (set to 0.110889, only one state): then it is 0.110889;

[0168] calculate :

[0169]

[0170] Dynamic Probabilistic Model It means that the model can fully reflect the dynamic relationship between the current state and the observed data, and represents the complete possibility of state transition.

[0171] See also Figure 4 , the specific steps for obtaining the updated model parameters are:

[0172] According to the dynamic probability model, the state transition probability of the hidden Markov model is updated, referring to the complexity and adaptability of the model, using the formula:

[0173]

[0174] Generate updated state transition probabilities, where represents a dynamic probability model, represents the adjustment factor, is the normalization constant, is the updated state transition probability;

[0175] Combine the updated state transition probability with the current remote sensing data to update the initial observation probability, using the formula:

[0176]

[0177] Get the updated observation probability, where is the updated state transition probability, For new remote sensing data, and To adjust the parameters, is the updated observation probability;

[0178] Combine the updated observation probability and the updated state transition probability to re-estimate the model parameters using the formula:

[0179]

[0180] Get the updated model parameters, where and are the updated state transition probability and observation probability, respectively. represents the weight of the state transition parameter, represents the weight of the observed parameter, is the normalization coefficient, Represents the updated model parameters.

[0181] Updated state transition probability formula:

[0182]

[0183] : The output of the dynamic probability model is taken from the actual calculation results, such as ;

[0184] : The adjustment coefficient determined based on historical data analysis is obtained through data regression analysis and is set to 0.3;

[0185] : The weight of relevant terrain feature data is based on expert evaluation and is taken as 0.2;

[0186] : The influence weight of environmental factors, obtained through environmental monitoring equipment, is set to 0.5;

[0187] : Normalization constant to ensure the sum is 1, calculated based on the sum of probabilities from the previous model, set to 1.0;

[0188] Calculation process:

[0189]

[0190]

[0191]

[0192] Calculated It indicates that the updated state transition probability is 31%, which reflects the model capability after adjustment according to the current dynamic data.

[0193] Updated observation probability formula:

[0194]

[0195] : 0.31 obtained from the previous step;

[0196] : The current remote sensing data, the value captured by remote sensing technology, is set to 0.8;

[0197] : The impact adjustment coefficient of observation data is set to 0.9;

[0198] : The constant used for equilibrium probability calculation, adjusted according to the sum of previous data, set to 0.1;

[0199] Calculation process:

[0200]

[0201]

[0202]

[0203] Calculated It indicates that the updated observation probability is 64.1%, reflecting the updated observation probability based on the current terrain data.

[0204] Updated model parameter formula:

[0205]

[0206] : The value obtained from the previous step is 0.31;

[0207] : The value obtained from the previous step is 0.641;

[0208] : The weight of the state transition parameter is set to 0.6;

[0209] : The weight of the observation parameter, set to 0.4;

[0210] : A constant used for normalization to ensure the consistency of the results, set to 1.0;

[0211] Calculation process:

[0212]

[0213]

[0214]

[0215] Calculated represents the synthesized model parameter value, which reflects the model capability after comprehensive update of state transition and observation probability.

[0216] See also Figure 5 , the specific steps for obtaining the status change results are:

[0217] According to the updated model parameters, the observed data is processed using the Viterbi algorithm, and the observed data and the updated parameters are combined using the formula:

[0218]

[0219] Generate initial path probabilities, where are the updated model parameters, is the observed data, is the initialization path probability;

[0220] Based on the initial path probability, the observation data is captured at each time point, and the path probability at multiple time points is gradually calculated and updated according to the current state transition probability and observation probability, applying the formula:

[0221]

[0222] Get the updated path probability, where is the path probability at the previous time point, For time point The model parameters, is the normalization constant, is the updated path probability;

[0223] Based on the updated path probabilities, the state change sequence is determined according to the optimal path probability calculated in the Viterbi algorithm, using the formula:

[0224]

[0225] By selecting the path with high probability, the state change result is determined, where is the state change result, is the updated path probability.

[0226] Initialization path probability formula:

[0227]

[0228] :The updated model parameters are obtained through machine learning algorithms based on historical data and current observation data. 0.9, this value is obtained through statistical analysis of historical data and indicates the possibility of state transition;

[0229] : Observation data, terrain data obtained through remote sensing technology, set as , which is the value extracted from the current remote sensing image;

[0230] Normalization constant Make sure the sum of probabilities is 1;

[0231] calculate ;

[0232] Normalize and set the three states similar, then ;

[0233] calculate ;

[0234] Initialize path probabilities represents the probability of each state being selected as the starting state given the initial observation.

[0235] Updated path probability formula:

[0236]

[0237] : The path probability at the previous time point is set to 0.333;

[0238] : Model parameters of target time and state, set to 0.9, obtained from data analysis;

[0239] Normalization constant : Set to a constant of 1 to simplify the calculation;

[0240] calculate ;

[0241] Update path probability ;

[0242] Calculated Indicates at time The probability of the optimal path.

[0243] State change result formula:

[0244]

[0245] : The optimal path probability array at all time points, obtained from the previous steps;

[0246] set up The array is [0.33, 0.27, 0.40], select the position of the maximum value, ;

[0247] It means that the state sequence ends at the third state, indicating that this is the terrain state change path.

[0248] See also Figure 6 , the specific steps for obtaining terrain prediction analysis results are:

[0249] According to the state change results, the monitoring data provides the historical height change information of each terrain unit, through the formula:

[0250]

[0251] Get the rate of change of each terrain unit, where represents the height change of terrain cells, represents the time interval, Indicates the rate of change of each terrain unit;

[0252] The terrain adjustment coefficient is calculated based on the change rate of each terrain unit and the environmental factors, using the formula:

[0253]

[0254] Get the adjusted rate of change, where Represents the rate of change of each terrain unit, Represents environmental factors, is a constant, represents the adjusted rate of change;

[0255] The adjusted rate of change is applied to the prediction model to analyze the terrain height in the future time period, through the formula:

[0256]

[0257] Generate terrain prediction analysis results, where Represents the current terrain height, represents the adjusted rate of change, represents the prediction time length, Represents the results of terrain prediction analysis.

[0258] The formula for the rate of change of each terrain unit is:

[0259]

[0260] : The height difference of terrain changes. The monitoring data shows that the change at the target location is 10 meters;

[0261] : Time interval. According to the monitoring data, this change occurs within one year, that is, 365 days;

[0262] The height data of location A at the beginning and end of the year are obtained through terrain monitoring equipment. The height difference is measured to be 10 meters, and the time interval is 365 days. The value is substituted into the formula to calculate the terrain change rate;

[0263] calculate :

[0264]

[0265] This rate represents the rate of change of altitude at location A on a daily basis.

[0266] Adjusted rate of change formula:

[0267]

[0268] : The rate of change calculated from the previous formula, 0.0274m / day;

[0269] : Environmental factors, including geological and climate impacts, are set as environmental index 2;

[0270] : Normalization constant, used to adjust the results to a reasonable range, set to 1000;

[0271] The rate obtained from the previous formula is is 0.0274m / day;

[0272] Geological and climatic influences Set to 2;

[0273] Normalization constant Set to 1000 to ensure within a reasonable range;

[0274] Substitute the formula to calculate the terrain adjustment coefficient:

[0275]

[0276] This coefficient is used to adjust the rate value of the subsequent prediction model.

[0277] Terrain prediction analysis result formula:

[0278]

[0279] : Current altitude, set to 500m;

[0280] : The adjusted rate of change calculated from the previous formula, 0.00000105 units;

[0281] : Forecast duration, considering the next 5 years, that is, 1825 days;

[0282] Current altitude 500 meters;

[0283] Using the adjustment factor obtained in the previous formula 0.00000105 units;

[0284] Prediction time 5 years, or 1825 days;

[0285] Substitute the formula to calculate the future terrain height:

[0286]

[0287] This height forecast value represents the expected change in height at location A over the next 5 years.

[0288] See also Figure 7 ,The specific steps for obtaining the 3D terrain feature boundary extraction data are as follows:

[0289] According to the terrain prediction analysis results, key frame images are extracted from the remote sensing data set to perform preliminary terrain feature marking, using the formula:

[0290]

[0291] Generate a preliminary feature-labeled image set, where represents the preliminary feature labeled image set, represents a single remote sensing image as input, represents the critical area applied to the target terrain features, Indicates the total number of images;

[0292] Based on the preliminary feature-labeled image set, a deep neural network model is applied to learn and analyze the images using the formula:

[0293]

[0294] Generate a set of feature vectors, where represents the weights of the neural network layer, Indicates bias, represents the base of natural logarithm, represents the set of feature vectors, Indicates A preliminary feature labeling image set of images;

[0295] Based on the feature vector set, quantitative analysis is performed to distinguish differentiated terrain features using the formula:

[0296]

[0297] Get three-dimensional terrain feature boundary extraction data, where: represents a single feature vector extracted from the feature vector set, represents the total number of eigenvectors, represents the threshold used for terrain classification, Represents three-dimensional terrain feature boundary extraction data.

[0298] Preliminary feature labeled image set formula:

[0299]

[0300] : No. An image, for example, an image in a set of remote sensing images, whose average pixel value can be obtained by image processing software and is set to 120;

[0301] : The corresponding key area for feature extraction, which is a two-dimensional image pre-defined according to the terrain features. The value of the terrain feature area in the key area is set to 1, and the rest are 0;

[0302] : The total number of images processed, set to 5;

[0303] set up , for each (Set the average value of each image to 120), (40% of the area in the key area is 1, is 0);

[0304] calculate :

[0305]

[0306] This means that the sum of the image data set after weighting by the feature extraction key area is 240, which can be used to evaluate the strength of terrain features in the image.

[0307] Preliminary feature labeled image set formula:

[0308]

[0309] : The weight of the neural network layer is set to 0.05;

[0310] : Bias, set to 0.1;

[0311] : The image data obtained in the previous step is 240;

[0312] : The number of images is 5;

[0313] Using the value calculated in the previous step ;

[0314] calculate :

[0315]

[0316] This means that the sum of the feature vectors after processing by the neural network is 5, reflecting the feature strength of the input image.

[0317] Three-dimensional terrain feature boundary extraction data formula:

[0318]

[0319] : A single feature vector extracted from the feature vector set, set Divide into 5 equal parts, each part is 1;

[0320] : The total number of components is 5;

[0321] : Classification threshold, set to 0.2;

[0322] Use the previous step ;

[0323] calculate :

[0324]

[0325] This means that the terrain feature boundary extraction data obtained after feature vector weighting is 1, which can be regarded as the final recognition result of the terrain boundary.

[0326] A high-precision extraction method for remote sensing and mapping three-dimensional terrain features, the high-precision extraction method for remote sensing and mapping three-dimensional terrain features is performed based on the high-precision extraction system for remote sensing and mapping three-dimensional terrain features, and comprises the following steps:

[0327] S1: Based on satellite remote sensing images, the identified images are analyzed to identify the terrain in the image, and the terrain status is classified into three types: normal, warning, and post-disaster. Parameters are set according to the relationship between the status and terrain characteristics, and the initial state distribution of the hidden Markov model is set to obtain the initialization probability matrix;

[0328] S2: According to the initialization probability matrix, capture new remote sensing images, extract terrain feature data in the image, analyze the feature data to update the state transition probability and observation probability, adjust the parameters of the hidden Markov model through probability, optimize the model, and obtain updated model parameters;

[0329] S3: Use the updated model parameters to process the continuous remote sensing image data, identify the terrain state change sequence through the Viterbi algorithm, analyze and record the terrain state transition, predict the terrain change trend in the future time period, and generate terrain prediction analysis results;

[0330] S4: Based on the terrain prediction analysis results, execute deep learning algorithms to process remote sensing images, annotate terrain features through semantic segmentation, enhance the recognition accuracy of terrain features, and obtain a terrain feature distinction framework;

[0331] S5: Apply the terrain feature differentiation framework to analyze the newly captured remote sensing images, extract the terrain feature boundaries, record and organize the boundary data, and generate three-dimensional terrain feature boundary extraction data.

[0332] The above are only preferred embodiments of the present invention and are not intended to limit the present invention in other forms. Any technician familiar with the profession may use the technical contents disclosed above to change or modify them into equivalent embodiments with equivalent changes and apply them to other fields. However, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention without departing from the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A high-precision extraction system for remote sensing three-dimensional terrain features, characterized in that: The system comprises: The terrain state definition module identifies normal, warning, and post-disaster terrain states based on satellite remote sensing images, associates each state with a set of terrain features, obtains state definition data, initializes the hidden Markov model, sets the initial state distribution, and obtains the initialization probability matrix; The terrain dynamic probability model module captures new remote sensing images based on the initialization probability matrix, calculates state transition and observation probabilities through terrain feature data, generates a dynamic probability model, and uses the dynamic probability model to update the parameters of the hidden Markov model to obtain updated model parameters; The steps for obtaining the dynamic probability model are specifically as follows: Based on the initialization probability matrix, capture a new remote sensing image, pass the terrain feature data, and apply the formula: Generate the initial state transition probability, where τ b represents the initial state transition probability, F b,i Represents the terrain feature data of the i-th state, P a represents the initialization probability matrix, α b Indicates adjustment parameters; Based on the initial state transition probability, the regional feature information is updated using the new remote sensing image data, and the updated feature information is combined with the transition probability matrix using the formula: Get the initial observation probability, where ω b represents the initial observation probability, τ b represents the initial state transition probability, I b represents remote sensing image data, β b represents the normalization factor; Combine the initial observation probability and the initial state transition probability, perform normalization, and strengthen the model's response to the state probability, using the formula: Generate a dynamic probability model, where M b represents the dynamic probability model, ω b represents the initial observation probability, τ b represents the initial state transition probability; The terrain change sequence decoding module applies the Viterbi algorithm to process the continuous observation data based on the updated model parameters, identifies the terrain state change sequence, generates the state change result, and predicts the terrain change trend in the future time period according to the state change result to obtain the terrain prediction analysis result; The three-dimensional terrain feature boundary extraction module performs deep learning training of remote sensing images based on the terrain prediction analysis results, performs semantic segmentation on terrain features, generates a terrain feature distinction framework, applies the terrain feature distinction framework, extracts terrain feature boundaries, and obtains three-dimensional terrain feature boundary extraction data.

2. The high-precision extraction system for remote sensing and mapping three-dimensional terrain features according to claim 1 is characterized in that: The steps of obtaining the initialization probability matrix are specifically as follows: Based on satellite remote sensing images, identify terrain states and associate a set of terrain features with each state using the formula: Get the state definition data, where X a,i is the feature matrix of the i-th state, F a,i is the terrain feature weight of the i-th state, S a represents the state definition data, n represents the total number of terrain states; Use the state definition data and initialize the hidden Markov model, set the initial state distribution, and apply the formula: Generate the initial state probability, where π a represents the initialization state probability, Z a represents the normalization constant, β a represents the adjustment parameter, S a Represents state definition data; According to the initialization state probability, a probability matrix is ​​constructed, and the probability matrix after averaging is calculated using the formula: Generate an initialization probability matrix, where P a represents the initialization probability matrix, π a,i Initialize the state probability from the i-th state, n is the total number of states.

3. The high-precision extraction system for remote sensing and mapping three-dimensional terrain features according to claim 1 is characterized in that: The steps for obtaining the updated model parameters are specifically as follows: According to the dynamic probability model, the state transition probability of the hidden Markov model is updated, referring to the complexity and adaptability of the model, using the formula: Generate updated state transition probability, where M b represents the dynamic probability model, α c , β c , γ c represents the adjustment coefficient, λ c is the normalization constant, T c is the updated state transition probability; Combine the updated state transition probability and the current remote sensing data to update the initial observation probability, using the formula: Get the updated observation probability, where T c is the updated state transition probability, D c is the new remote sensing data, δ c and c To adjust the parameters, O c is the updated observation probability; Combining the updated observation probability and the updated state transition probability, the model parameters are re-estimated using the formula: Get the updated model parameters, where T c and O c are the updated state transition probability and observation probability, η c represents the weight of the state transition parameter, θ c represents the weight of the observation parameter, ζ c is the normalization coefficient, H c Represents the updated model parameters.

4. The high-precision extraction system for remote sensing three-dimensional terrain features according to claim 3 is characterized in that: The steps for obtaining the state change result are specifically as follows: According to the updated model parameters, the observed data is processed using the Viterbi algorithm, and the observed data and the updated parameters are combined using the formula: Generate initial path probability, where H c is the updated model parameter, D d is the observed data, Π d is the initialization path probability; Based on the initial path probability, the observation data is captured at each time point, and the path probabilities at multiple time points are gradually calculated and updated according to the current state transition probability and observation probability, applying the formula: Get the updated path probability, where P d,t-1 is the path probability at the previous time point, H d,tk is the model parameter at time point t, λ d is the normalization constant, P d is the updated path probability; Based on the updated path probabilities, the state change sequence is determined according to the optimal path probability calculated in the Viterbi algorithm, using the formula: S d =arg max(P d ) By selecting the path with high probability, the state change result is determined, where S d is the state change result, P d is the updated path probability.

5. The high-precision extraction system for remote sensing and mapping three-dimensional terrain features according to claim 4 is characterized in that: The steps for obtaining the terrain prediction analysis results are specifically as follows: According to the state change results, the monitoring data provides the historical height change information of each terrain unit, through the formula: Get the rate of change of each terrain unit, where Δs h Represents the height change of terrain unit, Δt h represents the time interval, v h Indicates the rate of change of each terrain unit; According to the change rate of each terrain unit and the environmental factors, the terrain adjustment coefficient is calculated by the formula: Get the adjusted rate of change, where v h Represents the rate of change of each terrain unit, E h represents the environmental factor, Z is a constant, a h represents the adjusted rate of change; The adjusted rate of change is applied to the prediction model to analyze the terrain height in the future time period, through the formula: Generate terrain prediction analysis results, where s 0h Represents the current terrain height, a h represents the adjusted rate of change, t future represents the prediction time length, P h Represents the results of terrain prediction analysis.

6. The high-precision extraction system for remote sensing and mapping three-dimensional terrain features according to claim 5 is characterized in that: The steps for obtaining the three-dimensional terrain feature boundary extraction data are specifically as follows: According to the terrain prediction analysis results, key frame images are extracted from the remote sensing data set to perform preliminary terrain feature marking, using the formula: Generate a preliminary feature-labeled image set, where D g represents the preliminary feature labeled image set, I gk represents a single remote sensing image as input, represents the key area applied to the target terrain features, and K represents the total number of images; Based on the preliminary feature-labeled image set, a deep neural network model is applied to learn and analyze the images using the formula: Generate a set of feature vectors, where w k represents the weight of the neural network layer, b k represents the bias, e represents the natural logarithm base, F g represents the feature vector set, D gk represents the preliminary feature labeled image set of the k-th image; Based on the feature vector set, quantitative analysis is performed to distinguish differentiated terrain features using the formula: Get the 3D terrain feature boundary extraction data, where F gj represents a single feature vector extracted from the feature vector set, J represents the total number of feature vectors, and C j represents the threshold for terrain classification, B g Represents three-dimensional terrain feature boundary extraction data.

7. A high-precision extraction method for remote sensing three-dimensional terrain features, characterized in that: The method according to any one of claims 1 to 6 is implemented by the high-precision extraction system for remote sensing and mapping of three-dimensional terrain features, comprising the following steps: Based on satellite remote sensing images, the identified images are analyzed to identify and classify the terrain status into three categories: normal, warning, and disaster. The initial state distribution of the hidden Markov model is set to obtain the initialization probability matrix. According to the initialization probability matrix, new remote sensing images are captured, terrain feature data are extracted, and state transition probabilities and observation probabilities of the hidden Markov model are updated to obtain updated model parameters; Using the updated model parameters, continuous remote sensing image data is processed, a terrain state change sequence is identified through a Viterbi algorithm, the terrain state transition is analyzed and recorded, terrain change trends in a future time period are predicted, and terrain prediction analysis results are generated; Based on the terrain prediction and analysis results, a deep learning algorithm is executed to process the remote sensing image, terrain features are annotated through semantic segmentation, the recognition accuracy of terrain features is enhanced, and a terrain feature distinction framework is obtained; The terrain feature differentiation framework is applied to analyze newly captured remote sensing images, extract terrain feature boundaries, record and organize boundary data, and generate three-dimensional terrain feature boundary extraction data.