Land disaster stress-based land maneuvering traffic degree evaluation model and method

Abstract

The invention provides a land maneuvering traffic degree evaluation method based on geological disaster stress. The land maneuvering traffic degree evaluation method comprises the following steps: (1) constructing a land maneuvering traffic degree evaluation data resource system based on geological disaster stress; (2) collecting and preprocessing multi-source data related to traffic degree evaluation in the land maneuvering traffic degree evaluation data resource system; and (3) constructing a passage attention mechanism neural network-based passage degree evaluation model in combination with different passage influence factor data provided by the onshore maneuvering passage degree evaluation data resource system, and calculating weights of the different passage influence factor data through the model to obtain a sensitive feature combination influencing onshore maneuvering passage. And the traffic degree evaluation model pays attention to the sensitive features in the training process, so that a traffic degree evaluation result is obtained. The method is reasonable in conception, comprehensively considers the influence of various influence factors such as geological disasters, weather, traffic and the like on land mobile traffic, can obtain a more reliable traffic degree evaluation result, and has a better application prospect.

Description

Technical Field

[0001] The present invention belongs to the field of remote sensing, computer science and technology and off-road motorized traffic technology, and in particular relates to a land motorized trafficability evaluation model and method based on geological disaster threat. Background Art

[0002] Chen Huiyan and others introduced an important concept in the article "A Review of the Development of Military Ground Unmanned Mobile Platform Technology": Accessibility usually represents the degree of difficulty in accessing an area, while land mobility accessibility is an indicator of the degree of difficulty in accessing land mobile equipment in off-road scenarios. Analyzing and evaluating land mobility accessibility is the basis for land mobile access. Accessibility represents the level of accessibility, which can provide a basis for route designation, risk control, and other aspects in actual access operations, significantly improving the success rate and safety of access operations.

[0003] Zhang Wenshi et al. conducted a study on terrain passability based on computer technology in the article "Research on Terrain Passability". They modeled mathematical models for terrain factors such as pavement materials and longitudinal slope, meteorological factors, and time factors, and analyzed the passability of tracked vehicles such as tanks traveling along the road. Although this is an analysis of road passability, it also has certain reference significance for the analysis of off-road passability in terms of terrain factors, meteorological factors, and time factors. In "Road Passability Analysis of Unmanned Tracked Platforms in Off-Road Environments", Zhou Mengru et al. proposed a method for analyzing the passability of unmanned tracked platforms in off-road environments, using camera and laser radar perception schemes to address the problems of fuzzy road features and complex terrain in off-road environments. Based on image semantic segmentation, the geometric features of the ground terrain were analyzed and proposed, and the longitudinal, lateral, and vertical wall passability of tracked vehicles were modeled and mathematically analyzed. Finally, the ground was divided into five levels according to geometric characteristics, namely easy to pass, difficult to pass, difficult to pass, impassable, and unknown passability. Among them, unknown passability refers to areas that are far away from the vehicle body or have poor image information collection results. Li Kunwei et al. proposed a soil accessibility assessment framework based on multivariate data in "Soil Off-Road Accessibility Assessment Based on Multi-Source Data", in which soil data, DEM data, vegetation data, and meteorological data are processed and evaluated, and finally compared with the vehicle cone index to determine whether it is passable. In the hierarchical off-road path planning method proposed by Yan Xingyu et al. in "Hierarchical Off-Road Path Planning Method Based on Accessibility Analysis", the surface cover is assigned corresponding values. For example, the surface cover type of bare land corresponds to a passability value of 0.1, and the wetland surface cover type corresponds to a passability value of 0.1. The lower the value, the more difficult it is to pass, and the higher the value, the easier it is to pass. This type of method is a quantitative analysis of passability in the accessibility analysis process.

[0004] There are two problems with the original technology: first, the factors that affect land motor traffic are not fully considered. The original technology mainly considers the impact of terrain and landforms on motor traffic, and often rarely considers other factors besides terrain and landforms, such as weather, transportation, and other factors, especially the impact of geological disasters. The original technology rarely considers this; second, the original technology often uses quantitative analysis methods to determine the impact of each factor on traffic. The method is relatively simple and has a low degree of intelligence. It is difficult to reflect the importance differences between different factors, it is difficult to distinguish the importance of different influencing factors, and the combination of sensitive features is unclear.

[0005] To sum up, it is necessary to make further innovations on existing technologies. Summary of the invention

[0006] In response to the technical problems existing in the above-mentioned background technology, the present invention proposes a reasonable concept. By constructing a land motor vehicle accessibility evaluation data resource system based on geological disaster coercion, the impact of multiple influencing factors such as geological disasters, meteorology, and transportation on land motor vehicle accessibility is comprehensively considered. At the same time, by introducing the channel attention mechanism, a accessibility evaluation model based on the channel attention mechanism neural network is constructed, so that the importance of features can be paid attention to during the model training process, the weights of different influencing factors can be calculated, and the sensitive feature combination affecting land motor vehicle accessibility is obtained, thereby obtaining a more reliable accessibility result, which has a good application prospect.

[0007] In order to solve the above technical problems, the present invention provides a land mobile traffic evaluation method based on geological disaster threat, which mainly includes the following steps:

[0008] (1) First, by integrating a series of data that can affect land motor traffic, analyzing the correlation between different data and land motor traffic, and constructing a land motor traffic evaluation data resource system based on geological disaster coercion;

[0009] (2) Collect and preprocess multi-source data related to the land mobile accessibility evaluation data resource system;

[0010] (3) Combined with the different traffic influencing factor data provided by the land mobile traffic evaluation data resource system, a traffic evaluation model based on the channel attention mechanism neural network is constructed. The weights of the different traffic influencing factor data are then calculated through the traffic evaluation model to obtain the sensitive feature combination that affects land mobile traffic. The traffic evaluation model pays attention to the sensitive features during the training process, thereby obtaining the traffic evaluation results.

[0011] The land motor accessibility evaluation method based on geological disaster threat, wherein the specific construction process of the land motor accessibility evaluation data resource system based on geological disaster threat in step (1) is as follows:

[0012] (1.1) Determine the evaluation objectives

[0013] Determine the spatial and temporal scope of the evaluation, as well as the access requirements in different combat maneuver environments;

[0014] (1.2) Identify evaluation factors

[0015] Comprehensively determine the specific element names related to the evaluation of maneuverability from three aspects: geometric passability elements, safety passability elements and support passability elements;

[0016] (1.3) Define evaluation indicators

[0017] Define specific quantitative indicators for each evaluation factor, which should accurately reflect the impact of changes in the factors on traffic flow;

[0018] (1.4) Establishing an indicator system structure

[0019] Based on the identified elements and defined indicators, an evaluation indicator system architecture is constructed.

[0020] The land mobile passability evaluation method based on geological disaster threat, wherein: the geometric passability factor in the step (1.2) is the ability of the vehicle to pass through uneven ground and various obstacles at a sufficiently high average speed; the geometric passability factor is determined by judging the constraint satisfaction of the vehicle geometric parameters and the terrain, obstacles, surface cover, and geological conditions, and the main indicators are gap failure and power failure.

[0021] The land mobile passability evaluation method based on geological disaster threat, wherein: the supporting passability factor in the step (1.2) is the ability of the vehicle to pass through soft ground at a sufficiently high average speed, and the main indicators are the hook traction, traction coefficient, driving efficiency and maximum speed.

[0022] The land mobility evaluation method based on geological disaster threat, wherein: the safety passability factor in the step (1.2) is the vehicle's safe passability under the influence of geological disaster hazards in the area and enemy anti-mobility threats.

[0023] The land mobile accessibility evaluation method based on geological disaster threat, wherein the specific process of step (2) is as follows:

[0024] (2.1) Collect multi-source data related to accessibility evaluation;

[0025] (2.2) Preprocess the collected multi-source data.

[0026] The method for evaluating land mobile accessibility based on geological disaster threat, wherein: the multi-source data in step (2.1) includes topography, geological disasters, meteorology and hydrology, vegetation coverage and obstacles; the multi-source data in step (2.1) can come from remote sensing images, geographic information system databases and field measurements.

[0027] The land mobile accessibility assessment method based on geological disaster threat, wherein: the preprocessing in the step (2.2) includes data cleaning, normalization, data enhancement and format conversion; the data cleaning mainly involves manually removing damaged or format-incompatible image files to prevent invalid data from affecting the training of the accessibility assessment model; the normalization involves scaling pixel values ​​to the range of [0, 1] or [-1, 1] using the minimum and maximum normalization method; the data enhancement involves changing the training data by rotation, flipping, noise addition and ratio adjustment to generate more samples, thereby enhancing the robustness and generalization ability of the accessibility assessment model; the format conversion involves adjusting the image resolution to match the input size of the accessibility assessment model through an interpolation method, and converting the image channel order to meet the input format requirements of the deep learning framework and the accessibility assessment model.

[0028] The land mobile accessibility evaluation method based on geological disaster threat, wherein the specific construction process of the accessibility evaluation model based on the channel attention mechanism neural network in step (3) is:

[0029] (3.1) Data preprocessing

[0030] Make sure the data is in a format suitable for neural network input and label each sample data;

[0031] (3.2) Design of traffic evaluation model architecture

[0032] Taking the fully connected neural network as the basic model, the channel attention mechanism based on the Squeeze-and-Excitation module (SE module) is introduced to enhance the network's attention to important channel features; the Squeeze function in the SE module is used to aggregate global information of each channel through global average pooling; the Excitation function in the SE module is used to calculate the importance weight of each channel through the fully connected layer; the Scale function in the SE module is used to multiply the weight by the original channel feature and readjust the weight of each channel; finally, the feature vector processed by the channel attention mechanism will be input into the fully connected layer for the final evaluation of maneuverability, and the evaluation result of the passability will be output;

[0033] (3.3) Training of traffic evaluation model

[0034] Select appropriate loss functions and optimizers, and adjust hyperparameters to achieve the best training results; perform multiple rounds of iterative training, input training data and labels, calculate outputs through forward propagation, calculate errors using loss functions, and update network weights through back propagation; introduce validation sets during training to monitor the performance of the passability evaluation model in real time to prevent overfitting;

[0035] (3.4) Testing and verification of accessibility evaluation model

[0036] The accessibility evaluation model is tested using a dataset that was not used in training and its performance on actual data is evaluated.

[0037] The land mobile accessibility evaluation method based on geological disaster threat, wherein the process of using the Squeeze function in the SE module in step (3.2) to aggregate global information of each channel through global average pooling is: applying global average pooling to the feature map of each channel to generate a channel descriptor; assuming that the input feature map is X∈R H×W×C , where H and W are the spatial dimensions of the feature map, and C is the number of channels; calculate the global average pooling result:

[0038]

[0039] In the above formula, z c is the global information aggregation result of the cth channel;

[0040] The process of using the Excitation function in the SE module in step (3.2) to calculate the importance weight of each channel through the fully connected layer is as follows: the global information aggregation result of each channel is processed through a fully connected layer to calculate the importance weight of each channel; assuming that the excitation network consists of two fully connected layers, the excitation function can use ReLU and Sigmoid;

[0041] s = σ(W2ReLU(W1z+b1)+b2);

[0042] In the above formula, W1 and W2 are the weight matrices of the fully connected layer, b1 and b2 are bias terms, σ is the Sigmoid function, and s∈R C is the channel weight vector;

[0043] In the step (3.2), the Scale function in the SE module is used to multiply the weight by the original channel feature. The process of re-adjusting the weight of each channel is as follows: applying the calculated importance weight of the channel to the original feature map, adjusting the weight of each channel, and the weighted feature map obtained for each channel cc is:

[0044] X′ c =s cX c ;

[0045] In the above formula, X c is the scaled c-th channel feature map, s c is the corresponding channel weight.

[0046] The land motor traffic evaluation method based on geological disaster threat, wherein the specific process of step (3.3) is:

[0047] The loss function is used to calculate the error between the prediction result of the passability evaluation model and the true label, and the cross entropy loss function suitable for classification tasks is used:

[0048]

[0049] In the above formula, y i,c is the c-th class true label of the ith sample, is the predicted probability of the accessibility evaluation model that the ith sample belongs to the cth category, where C is the number of categories;

[0050] The Adam optimizer is selected to update the weights of the traffic evaluation model to minimize the loss function. The calculation formula is:

[0051]

[0052]

[0053] In the above formula, m t and v t are the estimates of the first and second moments of the gradient, β1 and β2 are the decay rates of the first and second moments, and e is a small constant used to avoid division by zero errors; θ t is the parameter of the passability evaluation model of the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters of the passability evaluation model, θ t+1 is to use the optimizer to update the weights;

[0054] Calculate the output of the passability evaluation model through forward propagation:

[0055]

[0056] In the above formula, X is the input data, θ is the parameter of the passability evaluation model, and f is the forward propagation function of the passability evaluation model;

[0057] The gradient of the loss function with respect to the parameters of the passability evaluation model is calculated by backpropagation:

[0058]

[0059] During the training process, the validation set is used to monitor the performance of the passability evaluation model to prevent overfitting; the loss and accuracy of the validation set are calculated:

[0060] L val =Loss(f(X val ;θ),y val );

[0061]

[0062] In the above formula, X val and val is the input and label of the validation set, and Indicator is the indicator function used to calculate whether the prediction is correct.

[0063] The land mobile accessibility evaluation method based on geological disaster threat, wherein the evaluation process of the accessibility evaluation model in step (3.4) mainly includes the calculation of accuracy, precision, recall and F1-Score, and the formulas are as follows:

[0064]

[0065]

[0066] In the above formula, TP is the number of true positive examples, TN is the number of true negative examples, FP is the number of false positive examples, and FN is the number of false negative examples.

[0067] By adopting the above technical solution, the present invention has the following beneficial effects:

[0068] The land motor accessibility evaluation method based on geological disaster coercion of the present invention is reasonably conceived. By constructing a land motor accessibility evaluation data resource system based on geological disaster coercion, the impact of multiple influencing factors such as geological disasters, meteorology, and transportation on land motor accessibility is comprehensively considered. At the same time, the original method is relatively simple, it is difficult to reflect the importance difference of different land motor accessibility influencing factors, and the sensitive feature combination is unclear. By introducing the channel attention mechanism, a passability evaluation model based on the channel attention mechanism neural network is constructed, so that the passability evaluation model can pay attention to the feature importance during the training process, and the weights of different influencing factors can be calculated to obtain the sensitive feature combination that affects land motor accessibility, thereby obtaining a more reliable passability result. In recent years, the frequency of geological disasters has been increasing. The land motor accessibility under geological disaster coercion is an important factor that cannot be ignored. At the same time, deep learning technology including attention mechanism is the current mainstream machine learning technology with a wide range of uses. Therefore, the present invention has a good application prospect.

[0069] The present invention integrates a series of data that can affect land motor traffic, such as high-resolution satellite image data, geological disaster historical records, geological disaster susceptibility assessment, terrain data, geomorphological data, meteorological data, etc., to construct a land motor traffic assessment data resource system based on geological disaster threat. Compared with previous methods, the system takes into account the potential risks of geological disasters, and then uses a traffic assessment model based on a channel attention mechanism neural network to process and analyze the data: that is, firstly, the input influencing factor image data is normalized, and the pixel value is scaled to between 0 and 1; then the importance of each feature channel is weighted by using a channel attention mechanism based on a squeeze-and-excitation module, i.e., an SE module; wherein, each feature map is globally averaged pooled by a squeeze module in the squeeze-and-excitation module, and the spatial information of each feature map is compressed into a single global description vector, so that the model can focus on the overall information of each channel, and the information of the feature map is weighted by the excitation in the squeeze-and-excitation module. The on module passes the compressed vector through a small fully connected layer and generates the weight coefficient of each channel through the activation function. The generated weight coefficient is re-applied to the corresponding feature map through the Scale module in the Squeeze-and-Excitation module, and the strength of each feature channel is adjusted through channel-by-channel multiplication operations. The feature map processed by the channel attention mechanism is then flattened into a one-dimensional vector and input into the fully connected layer for further feature fusion and dimensionality reduction, further extracting high-order features, and finally directly predicting the off-road mobility evaluation result. The gradient is calculated by the back propagation algorithm, and the network weight is updated using the optimization algorithm to gradually reduce the loss function value and improve the prediction accuracy of the model. Finally, by comparing historical disaster events with actual traffic records, the error verification and evaluation of the accessibility evaluation model based on the channel attention mechanism neural network is carried out, thereby realizing high-precision and high-reliability land mobility evaluation under complex terrain conditions, providing a data basis for land mobility. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings in the following description are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0071] Figure 1 It is an overall flow chart of the land mobile trafficability evaluation method based on geological disaster threat of the present invention;

[0072] Figure 2 The present invention is a flowchart for constructing a neural network accessibility evaluation model based on a channel attention mechanism involved in the land mobile accessibility evaluation method based on geological disaster threat. DETAILED DESCRIPTION

[0073] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0074] The present invention is further explained below in conjunction with specific implementation modes.

[0075] like Figure 1 As shown, this embodiment provides a land mobile accessibility evaluation method based on geological disaster threat, which mainly includes:

[0076] S100. Constructing a data resource system for evaluating land mobility based on geological disasters

[0077] S101. Determine the evaluation objectives

[0078] Determine the spatial and temporal scope of the evaluation, as well as the access requirements in different combat maneuver environments;

[0079] S102. Identify evaluation factors

[0080] Comprehensively determine the specific element names related to the evaluation of maneuverability from three aspects: geometric passability elements, safety passability elements, and support passability elements;

[0081] S103. Define evaluation indicators

[0082] Define specific quantitative indicators for each evaluation factor, which should accurately reflect the impact of changes in the factors on traffic flow;

[0083] S104. Establish indicator system structure

[0084] Based on the identified elements and defined indicators, an evaluation indicator system architecture is constructed.

[0085] By analyzing the geological factors that affect land off-road mobility, the influencing factors are mainly divided into three categories: geometric passability factors, safe passability factors, and support passability factors. According to the classification of these three types of influencing factors, a data resource system affecting the evaluation of land mobility is constructed, among which geological disaster coercion belongs to the safe passability factor.

[0086] Geometric passability: The ability of a vehicle to pass through uneven ground and various obstacles at a sufficiently high average speed. Geometric passability is determined by judging the constraint satisfaction of the vehicle's geometric parameters (minimum ground clearance, longitudinal passing angle, approach angle, departure angle, minimum turning diameter, turning channel circle) and terrain (slope, slope direction), obstacles (natural, artificial), surface cover (vegetation, water system, residential areas, etc.), and geological conditions (hard rock, semi-hard rock). The main indicators are clearance failure and power failure.

[0087] Support passability: The ability of a vehicle to pass through soft ground (soft rock, hard soil, ordinary soil, loose soil, special soil) at a sufficiently high average speed. The main indicators are the hook traction, traction coefficient, driving efficiency and maximum speed.

[0088] Safe passability: The ability of vehicles to pass safely under the influence of potential geological disasters (collapse, landslide, mud-rock flow) in the area and the threat of enemy counter-mobility (firepower, manpower).

[0089] Table 1: Data resource system affecting off-road mobility planning

[0090]

[0091]

[0092] Some of the above data resource systems belong to basic data, and other data can be further calculated based on the basic data. For example, the sudden parameters and rainfall data in the supporting passability elements can be used to obtain the soil cone index through inversion. The soil cone index comes from the cone index method. The cone index method is a method for quickly determining the exercisability of soil proposed by the U.S. Army Engineering Corps Waterway Experiment Station. It is widely used in many countries such as the United States and the United Kingdom; including the cone index (CI), rated cone index (RCI), and vehicle cone index (VCI), which represent the following meanings respectively:

[0093] CI stands for soil strength or bearing capacity;

[0094] RC I The strength or bearing capacity of the soil after it has been reshaped by repeated vehicle passage;

[0095] VC I is the minimum soil cone index required for a vehicle to complete a certain number of passes in the same rut, which represents the soil bearing capacity required by a certain type of vehicle.

[0096] This paper mainly discusses the cone index (CI), which aggregates the mechanical properties that constitute the resistance and thrust characteristics of the soil into one value, and is generally measured by a cone index meter. Some studies have shown that RC I is related to the water content and clay content of the soil, as shown in the following formula:

[0097]

[0098] In the above formula, RC I is the rated cone index of the soil; C is the clay content in the soil; and M is the water content.

[0099] S200, data collection and preprocessing

[0100] S201. Data Collection

[0101] Collect multi-source data related to accessibility evaluation, including terrain, geological hazards, meteorology and hydrology, vegetation cover, obstacles, etc. The data can come from remote sensing images, geographic information system (GIS) databases, field measurements, etc.

[0102] S202, preprocess the collected data, including data cleaning, normalization, data enhancement, format conversion, etc. Among them, data cleaning mainly involves manually removing damaged or non-formatted image files to prevent invalid data from affecting the training of the accessibility evaluation model; normalization involves scaling pixel values ​​to the range of [0,1] or [-1,1] using the minimum-maximum normalization method; data enhancement involves changing the training data through rotation, flipping, noise addition, and ratio adjustment to generate more samples, thereby enhancing the robustness and generalization ability of the accessibility evaluation model; format conversion involves adjusting the image resolution to match the input size of the accessibility evaluation model through interpolation methods, and converting the image channel order to meet the input format requirements of the deep learning framework and the accessibility evaluation model.

[0103] S300, Passage Evaluation Model Based on Channel Attention Mechanism Neural Network

[0104] Attention Mechanism (AM) is a computer science concept that imitates the way humans allocate attention. It aims to enhance the processing ability of the accessibility evaluation model for input data, so that it can focus more on relevant parts when processing sequence or set data, thereby improving the performance of the accessibility evaluation model. The attention mechanism was originally introduced in the field of natural language processing, but was later widely used in image processing, speech recognition, machine translation and other fields. The core idea of ​​the attention mechanism is to dynamically assign weights to each input part according to the content of the input data. These weights represent the importance of each part to the task. The attention mechanism generally includes the following key components: Query: The query is a vector used to determine the importance of each input part. It is usually a hidden state or output in the network; Key: The key is a vector associated with each input part, which is used to represent the characteristics of the input data; Value: The value is the value corresponding to each key, which contains the specific information of the input data.

[0105] In the field of computer vision, there are four main attention mechanisms: channel attention, spatial attention, self-attention, and mixed attention. The channel attention mechanism is a form of attention mechanism in deep learning. It was proposed by Jie Hu, LiShen, and Gang Sun in 2017. This mechanism is a specific attention model that focuses on modeling the importance of different channels in convolutional neural networks. In this way, the model can learn which feature channels are more important, so as to pay more attention to these channels, improving the expressiveness and efficiency of the network.

[0106] The channel attention mechanism usually aggregates the spatial information of each channel through a global pooling operation, and then calculates the importance weight of each channel through a simple learning framework (such as a fully connected layer). These weights are then used to adjust the activation values ​​of the original channels to achieve feature recalibration.

[0107] In the channel attention mechanism, global pooling is a key step in compressing spatial information. Here you can choose global maximum pooling or global average pooling, and the choice of the two has a significant impact on the performance of the attention mechanism.

[0108] Global average pooling: This method takes the average of all elements in the feature map as the representative of the channel. This method can capture background information and more general features, but may ignore some important but infrequent features. Global maximum pooling: This method takes the maximum value in the feature map as the representative of the channel. Global maximum pooling can capture the most significant features, which is helpful for the passability evaluation model to identify the key and significant parts of the image, but may lose some useful background information.

[0109] By introducing the attention mechanism into the neural network, different factors may have different degrees of influence on the final evaluation results in the accessibility evaluation model. By introducing the attention mechanism, the accessibility evaluation model can automatically focus on more important factors, so as to more accurately judge the pros and cons of regional accessibility. Accessibility evaluation involves comprehensive consideration of multiple factors, such as geological disasters, transportation, topography, lithology, etc. The attention mechanism can help the accessibility evaluation model to associate between multiple factors, so as to better understand the mutual influence between different factors. In short, the introduction of the attention mechanism can make the accessibility evaluation model more intelligent, accurate and interpretable. This will help improve the decision-making support capabilities of the model and make it more valuable in practical applications. In general, the neural network model based on the attention mechanism can automatically pay attention to important information in the data, so that the model focuses on key features and directly outputs accessibility results.

[0110] like Figure 2As shown in the figure, the specific construction process of the above-mentioned neural network passability evaluation model based on the channel attention mechanism is as follows:

[0111] S301. Data preprocessing

[0112] Make sure the data is in a format suitable for the neural network input, such as processing image data into a fixed-size tensor and labeling each sample data;

[0113] S302, Traffic Evaluation Model Architecture Design

[0114] Taking the fully connected neural network as the basic model, a channel attention mechanism based on the Squeeze-and-Excitation (SE) module is introduced to enhance the network's attention to important channel features.

[0115] The Squeeze function in the SE module is used to aggregate global information for each channel through global average pooling. The specific process is: apply global average pooling to the feature map of each channel to generate a channel descriptor. Assume that the input feature map is X∈R H×W×C , where H and W are the spatial dimensions of the feature map, and C is the number of channels; calculate the global average pooling result:

[0116]

[0117] In the above formula, z c is the global information aggregation result of the cth channel;

[0118] The Excitation function in the SE module is used to calculate the importance weight of each channel through a fully connected layer. The specific process is: the global information aggregation result of each channel is processed through a fully connected layer to calculate the importance weight of each channel. Assuming that the excitation network consists of two fully connected layers, the excitation function can use ReLU and Sigmoid.

[0119] s = σ(W2ReLU(W1z+b1)+b2);

[0120] In the above formula, W1 and W2 are the weight matrices of the fully connected layer, b1 and b2 are bias terms, σ is the Sigmoid function, and s∈R C is the channel weight vector;

[0121] The Scale function in the SE module is used to multiply the weight by the original channel feature and readjust the weight of each channel. The specific process is: the importance weight of the calculated channel is applied to the original feature map, the weight of each channel is adjusted, and the weighted feature map obtained for each channel cc is:

[0122] X′ c =sc X c ;

[0123] In the above formula, X c is the scaled c-th channel feature map, s c is the corresponding channel weight;

[0124] Finally, the feature vector processed by the channel attention mechanism will be input into the fully connected layer for the final maneuverability evaluation, and the evaluation result of the passability will be output.

[0125] S303, traffic evaluation model training

[0126] Select appropriate loss functions and optimizers, and adjust hyperparameters such as learning rate and momentum to obtain the best training effect; perform multiple rounds of iterative training, input training data and labels, calculate output through forward propagation, use loss functions to calculate errors, and update network weights through back propagation; introduce validation sets during training to monitor the performance of the passability evaluation model in real time to prevent overfitting. The specific process is as follows: the loss function is used for the error between the prediction result of the passability evaluation model and the true label, and the cross entropy loss function suitable for classification tasks is used:

[0127]

[0128] In the above formula, y i,c is the c-th class true label of the ith sample, is the predicted probability of the accessibility evaluation model that the ith sample belongs to the cth category, where C is the number of categories;

[0129] The Adam optimizer is selected to update the weights of the traffic evaluation model to minimize the loss function. The calculation formula is:

[0130]

[0131]

[0132] In the above formula, m t and v t are the estimates of the first and second moments of the gradient, β1 and β2 are the decay rates of the first and second moments, and e is a small constant used to avoid division by zero errors; θ t is the parameter of the passability evaluation model of the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters of the passability evaluation model, θ t+1 is to use the optimizer to update the weights;

[0133] Calculate the output of the passability evaluation model through forward propagation:

[0134]

[0135] In the above formula, X is the input data, θ is the parameter of the passability evaluation model, and f is the forward propagation function of the passability evaluation model;

[0136] The gradient of the loss function with respect to the parameters of the passability evaluation model is calculated by backpropagation:

[0137]

[0138] During the training process, the validation set is used to monitor the performance of the passability evaluation model to prevent overfitting; the loss and accuracy of the validation set are calculated:

[0139] L val =Loss(f(X val ;θ),y val );

[0140]

[0141] In the above formula, X val and val is the input and label of the validation set, and Indicator is the indicator function used to calculate whether the prediction is correct.

[0142] S304, Traffic evaluation model testing and verification

[0143] The accessibility evaluation model is tested using a data set that has not participated in the training to evaluate its performance on actual data. The evaluation process of the accessibility evaluation model mainly includes the calculation of accuracy, precision, recall and F1-Score, and the formulas are as follows:

[0144]

[0145]

[0146] In the above formula, TP is the number of true positive examples, TN is the number of true negative examples, FP is the number of false positive examples, and FN is the number of false negative examples.

[0147] S400, calculating the weights of different traffic influencing factor data through the traffic evaluation model to obtain a sensitive feature combination that affects land motor traffic, so that the traffic evaluation model pays attention to the sensitive features during the training process, thereby obtaining a traffic evaluation result.

[0148] The present invention constructs a land motor vehicle accessibility evaluation data resource system based on geological disaster coercion, comprehensively considers the impact of multiple influencing factors such as geological disasters, meteorology, and transportation on land motor vehicle accessibility, and constructs a accessibility evaluation model based on a channel attention mechanism neural network by introducing a channel attention mechanism. The weights of influencing factors can be calculated to obtain land motor vehicle accessibility results with a combination of sensitive features. In recent years, the frequency of geological disasters has been on the rise. Geological disaster coercion is an important factor that cannot be ignored for land motor vehicle accessibility. At the same time, deep learning technology including attention mechanism is the current mainstream machine learning technology, which has a wide range of uses and good application prospects.

[0149] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A land mobile accessibility evaluation method based on geological disaster threat, characterized in that , mainly including the following steps: (1) First, by integrating a series of data that can affect land motor traffic, analyzing the correlation between different data and land motor traffic, and constructing a land motor traffic evaluation data resource system based on geological disaster coercion; (2) Collect and preprocess multi-source data related to the land mobile accessibility evaluation data resource system; (3) Combined with the different traffic influencing factor data provided by the land mobile traffic evaluation data resource system, a traffic evaluation model based on the channel attention mechanism neural network is constructed. The weights of the different traffic influencing factor data are then calculated through the traffic evaluation model to obtain the sensitive feature combination that affects land mobile traffic. The traffic evaluation model pays attention to the sensitive features during the training process, thereby obtaining the traffic evaluation results.

2. The land motor traffic evaluation method based on geological disaster threat as claimed in claim 1, characterized in that The specific construction process of the land mobile accessibility evaluation data resource system based on geological disaster coercion in step (1) is as follows: (1.1) Determine the evaluation objectives Determine the spatial and temporal scope of the evaluation, as well as the access requirements in different combat maneuver environments; (1.2) Identify evaluation factors Comprehensively determine the specific element names related to the evaluation of maneuverability from three aspects: geometric passability elements, safety passability elements and support passability elements; (1.3) Define evaluation indicators Define specific quantitative indicators for each evaluation factor, which should accurately reflect the impact of changes in the factors on traffic flow; (1.4) Establishing an indicator system structure Based on the identified elements and defined indicators, an evaluation indicator system architecture is constructed.

3. The land motor traffic assessment method based on geological disaster threat as claimed in claim 2, characterized in that: The geometric passability factor in step (1.2) is the ability of the vehicle to pass through uneven ground and various obstacles at a sufficiently high average speed; the geometric passability factor is determined by judging the constraint satisfaction of the vehicle's geometric parameters and terrain, obstacles, surface cover, and geological conditions, and the main indicators are gap failure and power failure.

4. The land motor traffic assessment method based on geological disaster threat as claimed in claim 2, characterized in that: The supporting passability factor in step (1.2) is the ability of the vehicle to pass through soft ground at a sufficiently high average speed, and the main indicators are the hook traction, traction coefficient, driving efficiency and maximum speed.

5. The land motor traffic assessment method based on geological disaster threat as claimed in claim 2, characterized in that: The safe passability factor in step (1.2) is the safe passage capability of vehicles under the influence of geological disaster hazards in the area and enemy anti-mobility threats.

6. The land motor traffic evaluation method based on geological disaster threat according to claim 1, characterized in that: The specific process of step (2) is as follows: (2.1) Collect multi-source data related to accessibility evaluation; (2.2) Preprocess the collected multi-source data.

7. The land motor traffic assessment method based on geological disaster threat according to claim 6, characterized in that: The multi-source data in step (2.1) include topography, geological hazards, meteorology and hydrology, vegetation cover and obstacles; the multi-source data in step (2.1) may come from remote sensing images, geographic information system databases and field measurements.

8. The land motor traffic assessment method based on geological disaster threat according to claim 6, characterized in that: The preprocessing in step (2.2) includes data cleaning, normalization, data enhancement and format conversion; The data cleaning mainly involves manually removing damaged or non-formatted image files to prevent invalid data from affecting the training of the passability evaluation model; The normalization process is to scale the pixel values ​​to the range of [0, 1] or [-1, 1] by using the minimum and maximum value normalization method; The data enhancement process generates more samples by changing the training data through rotation, flipping, noise addition, and ratio adjustment, thereby enhancing the robustness and generalization ability of the passability evaluation model; The format conversion process is to adjust the image resolution to match the accessibility assessment model input size through an interpolation method, and convert the image channel order to meet the input format requirements of the deep learning framework and the accessibility assessment model.

9. The land mobility evaluation method based on geological disaster threat according to claim 1, characterized in that: The specific construction process of the passability evaluation model based on the channel attention mechanism neural network in step (3) is as follows: (3.1) Data preprocessing Make sure the data is in a format suitable for neural network input and label each sample data; (3.2) Design of traffic evaluation model architecture Taking the fully connected neural network as the basic model, a channel attention mechanism based on the Squeeze-and-Excitation module (SE module) is introduced to enhance the network's attention to important channel features; the Squeeze function in the SE module is used to aggregate global information of each channel through global average pooling; The Excitation function in the SE module is used to calculate the importance weight of each channel through the fully connected layer; The Scale function in the SE module is used to multiply the weight by the original channel feature and readjust the weight of each channel. Finally, the feature vector processed by the channel attention mechanism will be input into the fully connected layer for the final maneuverability evaluation, and the evaluation result of the passability will be output. (3.3) Training of traffic evaluation model Select appropriate loss functions and optimizers, and adjust hyperparameters to achieve the best training results; perform multiple rounds of iterative training, input training data and labels, calculate outputs through forward propagation, calculate errors using loss functions, and update network weights through back propagation; introduce validation sets during training to monitor the performance of the passability evaluation model in real time to prevent overfitting; (3.4) Testing and verification of accessibility evaluation model The accessibility evaluation model is tested using a dataset that was not used in training and its performance on actual data is evaluated.

10. The land motor traffic assessment method based on geological disaster threat as claimed in claim 9, characterized in that ,The process of using the Squeeze function in the SE module in step (3.2) to aggregate global information for each channel through global average pooling is: apply global average pooling to the feature map of each channel to generate a channel descriptor; assuming that the input feature map is X∈R H×W×C , where H and W are the spatial dimensions of the feature map, and C is the number of channels; Calculate the global average pooling result: In the above formula, z c is the global information aggregation result of the cth channel; The process of using the Excitation function in the SE module in step (3.2) to calculate the importance weight of each channel through the fully connected layer is as follows: the global information aggregation result of each channel is processed through a fully connected layer to calculate the importance weight of each channel; assuming that the excitation network consists of two fully connected layers, the excitation function can use ReLU and Sigmoid; s = σ(W2ReLU(W1z+b1)+b2); In the above formula, W1 and W2 are the weight matrices of the fully connected layer, b1 and b2 are bias terms, σ is the Sigmoid function, and s∈R C is the channel weight vector; In the step (3.2), the Scale function in the SE module is used to multiply the weight by the original channel feature. The process of re-adjusting the weight of each channel is as follows: applying the calculated channel importance weight to the original feature map, adjusting the weight of each channel, and the weighted feature map obtained for each channel cc is: X′ c =s c X c ; In the above formula, X c is the scaled c-th channel feature map, s c is the corresponding channel weight.

11. The land motor traffic evaluation method based on geological disaster threat as claimed in claim 9, characterized in that , the specific process of step (3.3) is: The loss function is used to calculate the error between the prediction result of the passability evaluation model and the true label, and the cross entropy loss function suitable for classification tasks is used: In the above formula, y i,c is the c-th class true label of the ith sample, is the predicted probability of the accessibility evaluation model that the i-th sample belongs to the c-th category, and C is the number of categories; The Adam optimizer is selected to update the weights of the traffic evaluation model to minimize the loss function. The calculation formula is: In the above formula, m t and v t are the estimates of the first and second moments of the gradient, β1 and β2 are the decay rates of the first and second moments, and e is a small constant used to avoid division by zero errors; θ t is the parameter of the passability evaluation model of the tth iteration, η is the learning rate, is the gradient of the loss function with respect to the parameters of the passability evaluation model, θ t+1 is to use the optimizer to update the weights; Calculate the output of the passability evaluation model through forward propagation: In the above formula, X is the input data, θ is the parameter of the passability evaluation model, and f is the forward propagation function of the passability evaluation model; The gradient of the loss function with respect to the parameters of the passability evaluation model is calculated by backpropagation: During the training process, the validation set is used to monitor the performance of the passability evaluation model to prevent overfitting; the loss and accuracy of the validation set are calculated: L val =Loss(f(X val ;θ),y val ); In the above formula, X val and val is the input and label of the validation set, and Indicator is the indicator function used to calculate whether the prediction is correct.

12. The land motor traffic evaluation method based on geological disaster threat as claimed in claim 9, characterized in that ,The evaluation process of the accessibility evaluation model in step (3.4) mainly includes the calculation of accuracy, precision, recall and F1-Score, and the formulas are as follows: In the above formula, TP is the number of true positive examples, TN is the number of true negative examples, FP is the number of false positive examples, and FN is the number of false negative examples.

Images