Geological ecological restoration evaluation method and system for geological disaster prevention and control, electronic equipment and storage medium

Abstract

The invention belongs to the technical field of ecological restoration evaluation, and provides a geological ecological restoration evaluation method and system for geological disaster prevention and control, electronic equipment and a storage medium. The method comprises the steps of research data acquisition, geological disaster type setting, initial evaluation, data preprocessing, model construction and training, risk determination before restoration, information acquisition after restoration, geological disaster risk prediction and geological ecological restoration evaluation. According to the method, the model is trained by using the remote sensing image, the sound wave data and the initial evaluation data, the evaluation result is predicted through the remote sensing image and the sound wave data, the risk evaluation difficulty and time are reduced, in addition, remote sensing and sound waves are combined, meanwhile, surface features and structural features are extracted, and the prediction effect of the model is optimized; through a multi-mode proximity perception model, the perception capability of the surrounding environment is improved, and the prediction precision of a single block is improved; and through the repair effect evaluation rule, a repair effect evaluation mechanism is optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of ecological restoration evaluation, in particular to a geological ecological restoration evaluation method and system for geological disaster prevention, an electronic device and a storage medium. BACKGROUND

[0002] Due to special lithology, hydrogeological conditions and human activities, karst landforms become high-risk areas of collapse, landslide, debris flow and other geological disasters. With the large-scale promotion of ecological protection and restoration projects in karst regions, geological ecological restoration effect evaluation has become a core link to verify the effectiveness of the project and guide the optimization of subsequent restoration.

[0003] Early evaluation relies on single geological parameters and empirical formulas, focusing only on the static risk of a single disaster, and has not formed a multi-disaster collaborative evaluation logic. Moreover, traditional geological evaluation requires manual field data collection, which consumes a lot of time and manpower, and the reliability of the results is poor due to incomplete data collection. Some methods rely only on remote sensing images or ground geological survey data, resulting in evaluation results limited by data dimensions. Traditional models such as linear regression and analytic hierarchy process are used in academic research, which can achieve risk classification but cannot effectively extract deep features from data, and the model generalization ability is weak. SUMMARY

[0004] In order to overcome the shortcomings of the prior art, the present application provides a geological ecological restoration evaluation method and system for geological disaster prevention, an electronic device and a storage medium, which solves the problems of traditional methods relying on empirical formulas, high cost, single data dimension and poor fitting effect.

[0005] To achieve the above object, the present application provides the following scheme:

[0006] A geological ecological restoration evaluation method for geological disaster prevention, comprising:

[0007] Collecting geological information, multi-dimensional remote sensing images and ultrasonic detection data of a target karst landform research area;

[0008] Setting a geological disaster type; the geological disaster type includes collapse disaster, landslide disaster and debris flow disaster;

[0009] According to the geological information, a pre-constructed risk evaluation formula is used to calculate the occurrence risk of each geological disaster type, and pre-statistical geological disaster risk data is obtained;

[0010] Pretreating the pre-statistical geological disaster risk data, the multi-dimensional remote sensing images and the ultrasonic detection data to obtain original training data;

[0011] training the multi-modal neighboring perception model by using the original training data to obtain a geological disaster risk prediction model;

[0012] determining a target karst landform monitoring area and original geological disaster risk data of the target karst landform monitoring area before geological ecological restoration;

[0013] after the target karst landform monitoring area is subjected to geological ecological restoration, repeatedly collecting the multi-dimensional remote sensing image and the ultrasonic detection data of the target karst landform monitoring area at a fixed time length as a period to obtain post-restoration environmental information; the fixed time length includes any one of 10 days, 15 days, and 20 days;

[0014] inputting the post-restoration environmental information into the geological disaster risk prediction model in sequence to obtain post-restoration geological disaster risk data;

[0015] according to the original geological disaster risk data, using a pre-set restoration effect evaluation rule to classify the post-restoration geological disaster risk data to obtain a geological ecological restoration evaluation result.

[0016] Preferably, the geological information, multi-dimensional remote sensing image, and ultrasonic detection data of the target karst landform research area are collected, including:

[0017] a vertical camera, an inclined camera, and an ultrasonic acquisition device are fixedly arranged on a drone; the vertical camera is perpendicular to the main body platform of the drone; the angle between the inclined camera and the vertical camera is 45°;

[0018] the target karst landform research area is divided into a plurality of square blocks of the same size by using GIS to obtain an original grid;

[0019] a collection track and a flight height are set; the requirements of the collection track include: respectively crossing each square block along the southeast, southwest, and northwest directions;

[0020] the flight height and the collection track are used to control the flight of the drone; after the flight distance reaches half of the distance corresponding to the square block each time, the vertical camera, the inclined camera, and the ultrasonic acquisition device are used to perform data acquisition once to obtain the multi-dimensional remote sensing image and the ultrasonic detection data; the multi-dimensional remote sensing image includes a vertical multi-dimensional remote sensing image and an azimuth multi-dimensional remote sensing image; the overlap degree of adjacent vertical multi-dimensional remote sensing images or azimuth multi-dimensional remote sensing images is 45%.

[0021] Preferably, the occurrence risk of each type of geological disaster is calculated by using a pre-constructed risk evaluation formula according to the geological information to obtain pre-statistical geological disaster risk data, including:

[0022] The collapse risk probability formula is constructed; the collapse risk probability formula is: Wherein,

[0023]

[0024] H a = ω1 × H a1 + ω2 × H a2 + ω3 × H a3 ; f ak = 1.2 × V s - 200; P1 represents the collapse risk probability; a0, a1, a2, a3 are respectively the basic regression coefficient, the first regression coefficient, the second regression coefficient, the third regression coefficient; R1 is the collapse risk value; Δh is the monthly amplitude of groundwater level; L t is the distance from the fault; ω f , ω h , ω Ha , ω n are respectively the bearing capacity weight, the water depth weight, the activity intensity weight, the void weight; f ak is the natural foundation bearing capacity of rock-soil mass; h is the groundwater level depth; H a is the human activity intensity comprehensive index; n is the rock-soil mass porosity; ω1, ω2, ω3 are respectively the mining activity weight, the underground engineering weight, the farmland proportion weight; H a1 , H a2 , H a3 are respectively the mining activity influence coefficient, the underground engineering influence coefficient, the farmland proportion influence coefficient; V s is the rock-soil mass shear wave velocity;

[0025] The landslide risk probability formula is constructed; the landslide risk probability formula is: P2 = 1-exp (-λ × R2); wherein,

[0026]

[0027] P2 represents the landslide risk probability; λ is the Poisson correction coefficient; R2 is the landslide risk value; ω α , ω Hsl , ω Fs , ω P5 , ω NDVI are respectively the slope weight, the height weight, the stability weight, the cumulative rainfall weight, the vegetation index weight; α is the slope gradient; H sl is the slope height; F s is the rock-soil mass stability coefficient; P5 is the continuous 5-day cumulative rainfall; NDVI is the normalized vegetation index; γ i is the natural specific gravity of rock-soil mass; h iφ represents the thickness of the soil and rock mass. i γ is the internal friction angle of the rock and soil; sat The saturated unit weight of the soil and rock mass; i The length of the sliding surface;

[0028] Construct a debris flow risk probability formula; the debris flow risk probability formula is as follows:

[0029] Where R3 represents the probability of debris flow risk; ω Ws ω I1 ω J These are the weights for sediment source, maximum rainfall, and longitudinal gradient, respectively; W s I1 represents the debris flow source volume; I1 represents the maximum hourly rainfall; J represents the longitudinal gradient of the valley.

[0030] The geological information is calculated using the collapse risk probability formula, the landslide risk probability formula, and the debris flow risk probability formula to obtain the pre-statistical geological hazard risk data; the pre-statistical geological hazard risk data includes: collapse probability data, landslide probability data, and debris flow probability data.

[0031] Preferably, the multimodal proximity perception model includes: an input layer, a feature extraction layer, a proximity perception embedding layer, a multimodal cross-attention fusion layer, and a multi-target output layer connected in sequence.

[0032] Preferably, the restored geological hazard risk data is classified according to the original geological hazard risk data using preset restoration effect evaluation rules to obtain a geological ecological restoration evaluation result, including:

[0033] The original geological hazard risk data and all the restored geological hazard risk data are classified according to the type of hazard to obtain single-indicator statistical data; the data in the single-indicator statistical data are arranged in chronological order.

[0034] Extract the difference between the first and last data points in the single-indicator statistical data to obtain the risk reduction before and after the repair.

[0035] Linear fitting is performed on the single-indicator statistical data to obtain a fitted curve, and the slope of the fitted curve is extracted to obtain the slope of the risk change trend.

[0036] A pre-constructed dynamic weighting formula is used to weight the risk reduction before and after the repair and the slope of the risk change trend to obtain a comprehensive evaluation score; the expression of the dynamic weighting formula is: Wherein, S is the comprehensive evaluation score; ΔR is the risk reduction before and after the repair; K is the adjustment constant; R0 is the first data in the single indicator statistics; and k is the slope of the risk change trend.

[0037] The comprehensive evaluation score is classified according to the preset scoring interval division standard to obtain the geological and ecological restoration evaluation result.

[0038] Preferably, the pre-statistical geological disaster risk data, the multi-dimensional remote sensing image, and the ultrasonic detection data are preprocessed to obtain raw training data, including:

[0039] The multidimensional remote sensing image is cropped according to the original grid division boundary to obtain a standard-sized image;

[0040] Use the coordinates corresponding to the center of the standard-sized image as the alignment label of the standard-sized image;

[0041] The ultrasonic detection data is denoised using wavelet transform and corrected using the propagation velocity parameters of karst rock mass to obtain calibrated detection data.

[0042] The coordinates corresponding to the calibration probe data center are used as the alignment labels of the calibration probe data. The standard-sized images and the calibration probe data with the same alignment labels are fused to obtain the alignment data.

[0043] The original training data is obtained by fusing the aligned data and the pre-statistical geological disaster risk data.

[0044] Preferably, the multimodal proximity sensing model is trained using the original training data to obtain a geological disaster risk prediction model, including:

[0045] The input layer is used to receive the original training data, and a block index is assigned to each group of aligned data.

[0046] In the feature extraction layer, ResNet-50 with embedded dilated convolution and residual attention is used to extract features from the standard-size image to obtain image features, and MLP with embedded BN and Dropout is used to extract features from the calibration detection data to obtain numerical features.

[0047] In the proximity-aware embedding layer, the block indices of the eight neighboring blocks adjacent to the current block are matched, and the image features of the neighboring blocks are fused with the image features of the current block to obtain image-aware enhanced features. The numerical features of the neighboring blocks are fused with the numerical features of the current block to obtain numerical-aware enhanced features.

[0048] The image perception enhancement feature is set as the query vector, and the numerical perception enhancement feature is set as the key vector and the value vector. Based on the cross-attention mechanism, the multimodal cross-attention fusion layer is used to calculate the weights of the query vector, the key vector, and the value vector and perform weighted fusion to obtain cross-modal fusion features.

[0049] The multi-objective output layer is used to perform multi-objective probability mapping on the cross-modal fusion features to obtain the model prediction results.

[0050] Preferably, a geological ecological restoration assessment system for geological disaster prevention and control includes:

[0051] The research data acquisition module is used to collect geological information, multi-dimensional remote sensing images, and ultrasonic detection data of the target karst landform research area.

[0052] The type setting module is used to set the geological hazard type; the geological hazard types include: collapse hazard, landslide hazard, and debris flow hazard;

[0053] The initial assessment module is used to calculate the occurrence risk of each type of geological hazard based on the geological information using a pre-built risk assessment formula, and obtain pre-statistical geological hazard risk data;

[0054] The preprocessing module is used to preprocess the pre-statistical geological disaster risk data, the multi-dimensional remote sensing image, and the ultrasonic detection data to obtain the original training data.

[0055] The model building module is used to train the multimodal proximity perception model using the original training data to obtain a geological disaster risk prediction model;

[0056] The pre-repair risk determination module is used to determine the target karst landform monitoring area and the original geological hazard risk data of the target karst landform monitoring area before geological and ecological restoration;

[0057] The post-repair information acquisition module is used to repeatedly acquire the multi-dimensional remote sensing images and ultrasonic detection data of the target karst landform monitoring area at fixed intervals after geological and ecological restoration, so as to obtain post-repair environmental information; the fixed intervals include any one of 10 days, 15 days, and 20 days.

[0058] The risk prediction module is used to sequentially input the remediated environmental information into the geological hazard risk prediction model to obtain remediated geological hazard risk data.

[0059] The effect evaluation module is used to classify the effect of the restored geological disaster risk data according to the original geological disaster risk data using preset restoration effect evaluation rules, and obtain the geological ecological restoration evaluation result.

[0060] Preferably, an electronic device includes: at least one processor and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned geological ecological restoration assessment method for geological disaster prevention and control.

[0061] Preferably, a non-transitory computer-readable storage medium stores computer instructions for causing a computer to execute the aforementioned geological ecological restoration assessment method for geological disaster prevention and control.

[0062] The present invention discloses the following technical effects:

[0063] This invention provides a geological ecological restoration assessment method, system, electronic device, and storage medium for geological disaster prevention and control. By training the model using remote sensing images, acoustic data, and initial assessment data, it overcomes the shortcomings of traditional methods, such as high assessment difficulty and cost, and limited data analysis dimensions, achieving prediction of assessment results using remote sensing images and acoustic data. Through a multimodal proximity sensing model, it addresses the problem of large fluctuations in prediction results caused by conventional models only considering the current block, thus improving the ability to perceive the surrounding environment. Finally, through restoration effect assessment rules, it addresses the problem of lacking before-and-after risk comparison due to assessment based solely on the final evaluation value, thus optimizing the restoration effect assessment mechanism. Attached Figure Description

[0064] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0065] Figure 1 This is a schematic diagram of a geological ecological restoration assessment process for geological disaster prevention and control provided in an embodiment of the present invention;

[0066] Figure 2 A flowchart for geological and ecological restoration assessment provided in this embodiment of the invention;

[0067] Figure 3 This is a schematic diagram of the acquisition direction provided in an embodiment of the present invention;

[0068] Figure 4This is a schematic diagram of the data collection points provided in an embodiment of the present invention;

[0069] Figure 5 This is a schematic diagram of the model architecture provided in an embodiment of the present invention. Detailed Implementation

[0070] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0071] The purpose of this invention is to provide a geological ecological restoration assessment method, system, electronic device and storage medium for geological disaster prevention and control, which solves the problems of traditional methods relying on empirical formulas, high cost, single data dimension and poor fitting effect.

[0072] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0073] Figure 1 This is a schematic diagram of a geological ecological restoration assessment process for geological disaster prevention and control provided in an embodiment of the present invention. Figure 2 A flowchart for geological and ecological restoration assessment provided in the embodiments of the present invention, such as Figure 1 and Figure 2 As shown, this invention provides a geological ecological restoration assessment method for geological disaster prevention and control, comprising:

[0074] Collect geological information, multidimensional remote sensing images, and ultrasonic detection data of the target karst landform study area;

[0075] Define the types of geological hazards; the types of geological hazards include: collapse hazards, landslide hazards, and debris flow hazards;

[0076] Based on the geological information, the occurrence risk of each type of geological hazard is calculated using a pre-constructed risk assessment formula to obtain pre-statistical geological hazard risk data;

[0077] The pre-statistical geological disaster risk data, the multi-dimensional remote sensing images, and the ultrasonic detection data are preprocessed to obtain the original training data;

[0078] The multimodal proximity perception model is trained using the original training data to obtain a geological disaster risk prediction model;

[0079] Determine the target karst landform monitoring area and the original geological hazard risk data of the target karst landform monitoring area before geological and ecological restoration;

[0080] After geological and ecological restoration of the target karst landform monitoring area, the multidimensional remote sensing images and ultrasonic detection data of the target karst landform monitoring area are repeatedly collected at fixed intervals to obtain post-restoration environmental information; the fixed intervals include any one of 10 days, 15 days, and 20 days.

[0081] The restored environmental information is sequentially input into the geological hazard risk prediction model to obtain restored geological hazard risk data;

[0082] Based on the original geological hazard risk data, the effects of the restored geological hazard risk data are classified using preset restoration effect evaluation rules to obtain geological ecological restoration evaluation results.

[0083] refer to Figure 3 and Figure 4 The study collected geological information, multi-dimensional remote sensing images, and ultrasonic detection data of the target karst landform research area, including:

[0084] A vertical camera, a tilting camera, and an ultrasonic acquisition device are fixedly installed on the drone; the vertical camera is perpendicular to the main platform of the drone; the angle between the tilting camera and the vertical camera is 45°.

[0085] The target karst landform study area was divided into several square blocks of the same size using GIS to obtain the original grid;

[0086] Set the data collection trajectory and flight altitude; the requirements for the data collection trajectory include: traversing each of the four square blocks in the four directions of east, south, west, and north respectively;

[0087] The UAV is controlled to fly according to the flight altitude and the acquisition trajectory. Each time the flight distance reaches half the distance corresponding to the square block, data is acquired using the vertical camera, the tilt camera, and the ultrasonic acquisition device to obtain the multidimensional remote sensing image and the ultrasonic detection data. The multidimensional remote sensing image includes: a vertical multidimensional remote sensing image and an azimuth multidimensional remote sensing image. The overlap between adjacent vertical multidimensional remote sensing images or azimuth multidimensional remote sensing images is 45%.

[0088] Specifically, based on the geological information, the occurrence risk of each type of geological hazard is calculated using a pre-constructed risk assessment formula to obtain pre-statistical geological hazard risk data, including:

[0089] Construct a collapse risk probability formula; the collapse risk probability formula is as follows: in,

[0090]

[0091] H a =ω1×H a1 +ω2×H a2 +ω3×H a3 ;f ak =1.2×V s -200; P1 represents the probability of collapse risk; a0, a1, a2, and a3 are the basic regression coefficient, the first regression coefficient, the second regression coefficient, and the third regression coefficient, respectively; R1 is the collapse risk value; Δh is the monthly variation of the groundwater level; L t ω is the distance from the fault; f ω h ω Ha ω n These are, respectively, the weights for bearing capacity, water depth, activity intensity, and porosity; f ak H represents the natural bearing capacity of the soil and rock foundation; h represents the depth of the groundwater level; H a The comprehensive index of human activity intensity; n represents the porosity of the soil and rock mass; ω1, ω2, and ω3 are the weights for mining activities, underground engineering, and the proportion of arable land, respectively; H a1 H a2 H a3 These are the impact coefficients for mining activities, underground engineering, and arable land proportion, respectively; V s Shear wave velocity of the rock and soil mass;

[0092] A landslide risk probability formula is constructed; the landslide risk probability formula is: P2=1-exp(-λ×R2); where,

[0093]

[0094] P2 represents the probability of landslide risk; λ is the Poisson correction coefficient; R2 is the landslide risk value; ω α ω Hsl ω Fs ω P5 ω NDVI These are the weights for slope, height, stability, cumulative rainfall, and vegetation index, respectively; α is the slope; H sl F represents the slope height. s P5 represents the stability coefficient of the soil and rock mass; P5 represents the cumulative rainfall over 5 consecutive days; NDVI represents the normalized difference in vegetation index; γ i The natural unit weight of the rock and soil mass; h i φ represents the thickness of the soil and rock mass. i γ is the internal friction angle of the rock and soil;sat The saturated unit weight of the soil and rock mass; i The length of the sliding surface;

[0095] Construct a debris flow risk probability formula; the debris flow risk probability formula is as follows:

[0096] Where R3 represents the probability of debris flow risk; ω Ws ω I1 ω J These are the weights for sediment source, maximum rainfall, and longitudinal gradient, respectively; W s I1 represents the debris flow source volume; I1 represents the maximum hourly rainfall; J represents the longitudinal gradient of the valley.

[0097] The geological information is calculated using the collapse risk probability formula, the landslide risk probability formula, and the debris flow risk probability formula to obtain the pre-statistical geological hazard risk data; the pre-statistical geological hazard risk data includes: collapse probability data, landslide probability data, and debris flow probability data.

[0098] refer to Figure 5 The multimodal proximity perception model includes: an input layer, a feature extraction layer, a proximity perception embedding layer, a multimodal cross-attention fusion layer, and a multi-target output layer connected in sequence.

[0099] Specifically, based on the original geological hazard risk data, the post-repair geological hazard risk data is classified according to preset remediation effect evaluation rules to obtain geological ecological restoration evaluation results, including:

[0100] The original geological hazard risk data and all the restored geological hazard risk data are classified according to the type of hazard to obtain single-indicator statistical data; the data in the single-indicator statistical data are arranged in chronological order.

[0101] Extract the difference between the first and last data points in the single-indicator statistical data to obtain the risk reduction before and after the repair.

[0102] Linear fitting is performed on the single-indicator statistical data to obtain a fitted curve, and the slope of the fitted curve is extracted to obtain the slope of the risk change trend.

[0103] A pre-constructed dynamic weighting formula is used to weight the risk reduction before and after the repair and the slope of the risk change trend to obtain a comprehensive evaluation score; the expression of the dynamic weighting formula is: Wherein, S is the comprehensive evaluation score; ΔR is the risk reduction before and after the repair; K is the adjustment constant; R0 is the first data in the single indicator statistics; and k is the slope of the risk change trend.

[0104] The comprehensive evaluation score is classified according to the preset scoring interval division standard to obtain the geological and ecological restoration evaluation result.

[0105] Furthermore, the pre-statistical geological disaster risk data, the multi-dimensional remote sensing images, and the ultrasonic detection data are preprocessed to obtain raw training data, including:

[0106] The multidimensional remote sensing image is cropped according to the original grid division boundary to obtain a standard-sized image;

[0107] Use the coordinates corresponding to the center of the standard-sized image as the alignment label of the standard-sized image;

[0108] The ultrasonic detection data is denoised using wavelet transform and corrected using the propagation velocity parameters of karst rock mass to obtain calibrated detection data.

[0109] The coordinates corresponding to the calibration probe data center are used as the alignment labels of the calibration probe data. The standard-sized images and the calibration probe data with the same alignment labels are fused to obtain the alignment data.

[0110] The original training data is obtained by fusing the aligned data and the pre-statistical geological disaster risk data.

[0111] refer to Figure 5 The multimodal proximity sensing model is trained using the original training data to obtain a geological disaster risk prediction model, including:

[0112] The input layer is used to receive the original training data, and a block index is assigned to each group of aligned data.

[0113] In the feature extraction layer, ResNet-50 with embedded dilated convolution and residual attention is used to extract features from the standard-size image to obtain image features, and MLP with embedded BN and Dropout is used to extract features from the calibration detection data to obtain numerical features.

[0114] In the proximity-aware embedding layer, the block indices of the eight neighboring blocks adjacent to the current block are matched, and the image features of the neighboring blocks are fused with the image features of the current block to obtain image-aware enhanced features. The numerical features of the neighboring blocks are fused with the numerical features of the current block to obtain numerical-aware enhanced features.

[0115] The image perception enhancement feature is set as the query vector, and the numerical perception enhancement feature is set as the key vector and the value vector. Based on the cross-attention mechanism, the multimodal cross-attention fusion layer is used to calculate the weights of the query vector, the key vector, and the value vector and perform weighted fusion to obtain cross-modal fusion features.

[0116] The multi-objective output layer is used to perform multi-objective probability mapping on the cross-modal fusion features to obtain the model prediction results.

[0117] Specifically, ecological restoration assessments for karst landforms mainly include:

[0118] 1) Collecting multi-source data: Using drones equipped with vertical cameras, oblique cameras, and ultrasonic equipment, and combined with GIS technology, the study area was divided into equal-sized square areas. Within each area, the drones flew in the four cardinal directions (north, south, east, and west), collecting data every half-area to ensure comprehensive coverage of multi-angle remote sensing images (vertical and azimuth, with a 45% overlap) and ultrasonic detection data.

[0119] 2) Disaster risk quantification: Establish corresponding risk probability formulas for different disasters, input geological information into the formulas, and calculate preliminary risk data for collapse, landslide, and debris flow disasters.

[0120] 3) Data Preprocessing and Model Training: Remote sensing images are cropped and aligned according to a grid. Ultrasonic data, after wavelet denoising and rock mass velocity correction, is fused with the aforementioned preliminary risk data to obtain training data. This data is then used to train a multimodal proximity sensing model. This model extracts image features using ResNet-50, extracts numerical features using MLP, fuses features from eight neighboring regions, and performs cross-attention weighting to ultimately map and output risk prediction results for various disasters.

[0121] 4) Evaluate the remediation effect: Before remediation, determine the initial risk level. After remediation, collect data every 10, 15, or 20 days and input the data into the model to obtain the post-remediation risk level. By calculating the magnitude of risk reduction and the slope of the trend, use a dynamic weighting formula to calculate the comprehensive score and score interval division to obtain the final evaluation result.

[0122] Further, 1) the detailed process of the corresponding steps includes: a vertical camera, a tilting camera, and an ultrasonic acquisition device are fixedly mounted on the UAV, with the vertical and tilting cameras using the same model and resolution; before operation, the ultrasonic acquisition device is set to a detection frequency suitable for karst landforms, referring to existing literature. ArcGIS software is used to divide the target karst landform study area into a grid. After grid division, the UAV's flight path is set, and the UAV needs to traverse each block from four directions, referring to... Figure 3 The distance between two consecutive data collections by the drone is half the side length of the block. (Reference) Figure 4 This results in an image overlap of 45% between adjacent blocks. For flight altitude, local literature needs to be consulted; the stable flight altitude of the drone should not be lower than the highest altitude of the target area, and the drone needs to maintain the same altitude during data acquisition.

[0123] Optionally, existing literature on ecological restoration of karst landforms mainly includes vegetation reconstruction, soil improvement, and hydrological regulation, some of which are as follows: 1) Vegetation reconstruction technology: Prioritize native tree species and herbaceous plants that are tolerant of poor soil and have strong soil-fixing ability, and restore vegetation structure through a mixed tree-shrub-vine-grass model. For example, in a certain area, mulberry trees are mixed with leguminous forage grasses to form biological crusts that fix carbon and nitrogen, while providing feed for herbivorous livestock. Studies have shown that natural restoration measures (such as closing mountains for afforestation) are more effective than artificial afforestation in improving biodiversity, so they are given priority in areas where human activities are controllable; inoculate with arbuscular mycorrhizal fungi to improve plant drought resistance and nutrient absorption capacity; and address the problem of the single ecological function of artificial pure forests by replanting native tree species and shrubs to construct multi-layered communities. 2) Applying calcium alginate-rice straw biochar composite material can fix soil phosphorus and nitrate nitrogen, reducing leaching; at the same time, it can increase the activity of alkaline phosphatase and urease, promoting soil nutrient cycling; in view of the characteristics of high calcium and magnesium content and phosphorus deficiency in karst soils, the use of low-phosphorus compound fertilizer combined with winter short-shoot pruning technology can stabilize the soil pH at 6.5 to 7.0; integrated crop-livestock models, such as grass-fed cattle-coated earthworm-coated earthworm castings-returned-to-field systems. 3) In view of the serious problem of surface water seepage in karst areas, technologies such as bare rock surface runoff collection and gully stormwater interception can be used to improve the efficiency of slope runoff interception; an integrated air-ground-space monitoring network can be constructed, integrating Beidou positioning, vegetation sensors and UAV inspections to monitor groundwater level changes in real time; the water self-adaptation characteristics of plants such as mulberry trees can be utilized, combined with mycorrhizal symbiosis to enhance drought resistance and carry out seasonal drought adaptation management.

[0124] Preferably, referring to existing literature, this embodiment constructs a single-hazard risk assessment system by combining soil conservation and water retention requirements:

[0125] 1) Probability of collapse risk:

[0126]

[0127] H a =ω1×H a1 +ω2×H a2 +ω3×H a3

[0128] f ak =1.2×V s -200

[0129] In the formula, P1 represents the probability of collapse risk; a0, a1, a2, and a3 are the basic regression coefficient, the first regression coefficient, the second regression coefficient, and the third regression coefficient, respectively, which are set to -1.2, 3.5, 2.8, and -1.5 in this embodiment; R1 is the collapse risk value; Δh is the monthly variation of the groundwater level; L t ω is the distance from the fault; f ω h ω Ha ω n These are the weights for bearing capacity, water depth, activity intensity, and porosity, respectively, set to 0.35, 0.3, 0.2, and 0.15 in this embodiment; f ak H represents the natural bearing capacity of the soil and rock foundation; h represents the depth of the groundwater level; H a ω1 represents the comprehensive index of human activity intensity; ω2 represents the porosity of the soil and rock mass; ω1, ω2, and ω3 represent the weights for mining activities, underground engineering, and the proportion of arable land, respectively, which are set to 0.4, 0.3, and 0.3 in this embodiment; H a1 H a2 H a3 These are the impact coefficients for mining activities, underground engineering, and arable land proportion, respectively; V s This represents the shear wave velocity of the soil and rock mass.

[0130] 2) Probability of landslide risk:

[0131] P2 = 1 - exp(-λ × R2)

[0132]

[0133] γ sat =γ i +0.8

[0134] l i =h i / cosα

[0135] In the formula, P2 represents the probability of landslide risk; λ is the Poisson correction coefficient, with a value of 2.2; R2 is the landslide risk value; ω α ω Hsl ω Fs ω P5 ω NDVIThese are the weights for slope, height, stability, cumulative rainfall, and vegetation index, respectively, set to 0.25, 0.2, 0.15, and 0.1 in this embodiment; α is the slope; H... sl F represents the slope height. s P5 represents the stability coefficient of the soil and rock mass; P5 represents the cumulative rainfall over 5 consecutive days; NDVI represents the normalized difference in vegetation index; γ i The natural unit weight of the rock and soil mass; h i φ represents the thickness of the soil and rock mass. i γ is the internal friction angle of the rock and soil; sat The saturated unit weight of the soil and rock mass; i The length of the sliding surface.

[0136] 3) Probability of debris flow risk:

[0137]

[0138] In the formula, R3 represents the probability of debris flow risk; ω Ws ω I1 ω J These are the weights for sediment source, maximum rainfall, and longitudinal gradient, respectively; W s I1 represents the debris flow source volume; I1 represents the maximum hourly rainfall; and J represents the longitudinal gradient of the valley.

[0139] Optionally, the acquired remote sensing images are cropped according to the grid boundaries of the blocks defined by the GIS, ensuring that all images maintain a consistent size. The cropped remote sensing images are then used as input for the subsequent ResNet-50. Ultrasonic detection data from karst landforms are susceptible to interference from rock fissures and air gaps, generating high-frequency noise. This embodiment uses wavelet transform to filter out high-frequency interference and retain effective detection signals. Furthermore, since different rock masses have different propagation speeds, the acquired acoustic data needs to be corrected based on the propagation speed parameters of the karst landform rock masses. The processed data is then used as input for the subsequent MLP. The processed remote sensing images and acoustic data are aligned and bound to the pre-statistical risk data using the center coordinates of their respective blocks, providing a location index for subsequent perception enhancement.

[0140] refer to Figure 5This embodiment provides a multimodal proximity perception model, which includes: an input layer, a feature extraction layer, a proximity perception embedding layer, a multimodal cross-attention fusion layer, and a multi-target output layer. The model's workflow is as follows: the input layer receives preprocessed raw training data and assigns a unique block index to each group of aligned data, marking the geographic block to which the data belongs. For image feature extraction, this embodiment selects ResNet-50 with embedded dilated convolution and residual attention. Dilated convolution can expand the image perception range, and residual attention can further focus on key disaster-related areas in the image. For acoustic data feature extraction, this embodiment uses an MLP with embedded Batch Normalization (BN) and Dropout. Conventional MLPs suffer from low efficiency and overfitting issues when extracting data features; by adding BN and Dropout... The module can effectively stabilize the training process. Considering that karst landform disasters are affected not only by the current block but also by the surrounding environment, this embodiment matches the current block with 8 neighboring blocks based on the block index, and fuses the extracted features of the neighboring blocks with the features of the current block to achieve the effect of perception enhancement. The image perception enhancement features are set as query vectors, and the numerical perception enhancement features are used as key vectors and value vectors, allowing the model to automatically judge the contribution of image features and numerical features to disaster prediction, and perform dynamic weight allocation and feature fusion. Through the multi-objective output layer, the cross-modal fused features are subjected to multi-objective probability mapping, and the risk probabilities of collapse, landslide, and debris flow are output simultaneously.

[0141] Furthermore, the karst landform areas to be monitored are identified, and the original geological hazard risk data for these areas before geological and ecological restoration is obtained. After the geological and ecological restoration is completed in these areas, multi-dimensional remote sensing images and ultrasonic detection data are repeatedly collected at 10, 15, or 20-day intervals to obtain post-restoration environmental information. Finally, this post-restoration environmental information is sequentially input into the geological hazard risk prediction model to obtain post-restoration geological hazard risk data. Based on disaster type (collapse, landslide, debris flow), the original geological disaster risk data before restoration and the geological disaster risk data for all periods after restoration are categorized into four independent datasets, namely single-indicator statistical data, with each dataset arranged in chronological order of collection time. The risk reduction before and after restoration for each disaster type is extracted by calculating the difference between the first and last data points in the single-indicator statistical data. Linear fitting is performed on the single-indicator statistical data, and the slope of the curve is extracted to obtain the risk change trend slope. A negative slope indicates a continuous decrease in risk over time, meaning the ecological restoration is effective; a positive slope indicates an increase in risk over time, suggesting that the ecological restoration has failed to achieve its intended effect. This embodiment establishes a dynamic weighting formula that combines the risk reduction before and after restoration with the risk change trend slope.

[0142]

[0143] In the formula, S is the comprehensive evaluation score; ΔR is the risk reduction before and after the repair; K is the adjustment constant; R0 is the first data in the single indicator statistics; and k is the slope of the risk change trend.

[0144] After obtaining the comprehensive evaluation score for each indicator, the corresponding repair effect level is matched according to the pre-set standards for each indicator.

[0145] As an optional implementation, this embodiment also provides a geological ecological restoration assessment system for geological disaster prevention and control, comprising:

[0146] The research data acquisition module is used to collect geological information, multi-dimensional remote sensing images, and ultrasonic detection data of the target karst landform research area.

[0147] The type setting module is used to set the geological hazard type; the geological hazard types include: collapse hazard, landslide hazard, and debris flow hazard;

[0148] The initial assessment module is used to calculate the occurrence risk of each type of geological hazard based on the geological information using a pre-built risk assessment formula, and obtain pre-statistical geological hazard risk data;

[0149] The preprocessing module is used to preprocess the pre-statistical geological disaster risk data, the multi-dimensional remote sensing image, and the ultrasonic detection data to obtain the original training data.

[0150] The model building module is used to train the multimodal proximity perception model using the original training data to obtain a geological disaster risk prediction model;

[0151] The pre-repair risk determination module is used to determine the target karst landform monitoring area and the original geological hazard risk data of the target karst landform monitoring area before geological and ecological restoration;

[0152] The post-repair information acquisition module is used to repeatedly acquire the multi-dimensional remote sensing images and ultrasonic detection data of the target karst landform monitoring area at fixed intervals after geological and ecological restoration, so as to obtain post-repair environmental information; the fixed intervals include any one of 10 days, 15 days, and 20 days.

[0153] The risk prediction module is used to sequentially input the remediated environmental information into the geological hazard risk prediction model to obtain remediated geological hazard risk data.

[0154] The effect evaluation module is used to classify the effect of the restored geological disaster risk data according to the original geological disaster risk data using preset restoration effect evaluation rules, and obtain the geological ecological restoration evaluation result.

[0155] As an optional implementation, this embodiment also provides an electronic device, including: at least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform the aforementioned geological ecological restoration assessment method for geological disaster prevention and control.

[0156] As an optional implementation, this embodiment also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the aforementioned geological ecological restoration assessment method for geological disaster prevention and control.

[0157] The beneficial effects of this invention are as follows:

[0158] This invention trains a model using remote sensing images, acoustic data, and initial assessment data, enabling the prediction of assessment results based on remote sensing images and acoustic data. This reduces the difficulty and time required for risk assessment. Furthermore, the combination of remote sensing and acoustic data simultaneously extracts surface and structural features, optimizing the model's predictive performance. The multimodal proximity perception model enhances the perception capability of the surrounding environment and improves the prediction accuracy of single blocks. Finally, the repair effect assessment mechanism is optimized through repair effect assessment rules.

[0159] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0160] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. Furthermore, those skilled in the art will recognize that, based on the ideas of the present invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A geological ecological restoration assessment method for geological disaster prevention and control, characterized in that, include: Collect geological information, multidimensional remote sensing images, and ultrasonic detection data of the target karst landform study area; Define the types of geological hazards; The types of geological hazards include: collapse hazards, landslide hazards, and debris flow hazards; Based on the geological information, the occurrence risk of each type of geological hazard is calculated using a pre-constructed risk assessment formula to obtain pre-statistical geological hazard risk data; The pre-statistical geological disaster risk data, the multi-dimensional remote sensing images, and the ultrasonic detection data are preprocessed to obtain the original training data; The multimodal proximity perception model is trained using the original training data to obtain a geological disaster risk prediction model; Determine the target karst landform monitoring area and the original geological hazard risk data of the target karst landform monitoring area before geological and ecological restoration; After geological and ecological restoration of the target karst landform monitoring area, the multidimensional remote sensing images and ultrasonic detection data of the target karst landform monitoring area are repeatedly collected at fixed intervals to obtain post-restoration environmental information; the fixed intervals include any one of 10 days, 15 days, and 20 days. The restored environmental information is sequentially input into the geological hazard risk prediction model to obtain restored geological hazard risk data; Based on the original geological hazard risk data, the effects of the restored geological hazard risk data are classified using preset restoration effect evaluation rules to obtain geological ecological restoration evaluation results.

2. The geological ecological restoration assessment method for geological disaster prevention and control according to claim 1, characterized in that, Collect geological information, multidimensional remote sensing images, and ultrasonic detection data of the target karst landform study area, including: A vertical camera, a tilting camera, and an ultrasonic acquisition device are fixedly installed on the drone; the vertical camera is perpendicular to the main platform of the drone; the angle between the tilting camera and the vertical camera is 45°. The target karst landform study area was divided into several square blocks of the same size using GIS to obtain the original grid; Set the data collection trajectory and flight altitude; the requirements for the data collection trajectory include: traversing each of the four square blocks in the four directions of east, south, west, and north respectively; The UAV is controlled to fly according to the flight altitude and the acquisition trajectory. Each time the flight distance reaches half the distance corresponding to the square block, data is acquired using the vertical camera, the tilt camera, and the ultrasonic acquisition device to obtain the multidimensional remote sensing image and the ultrasonic detection data. The multidimensional remote sensing image includes: a vertical multidimensional remote sensing image and an azimuth multidimensional remote sensing image. The overlap between adjacent vertical multidimensional remote sensing images or azimuth multidimensional remote sensing images is 45%.

3. The geological ecological restoration assessment method for geological disaster prevention and control according to claim 1, characterized in that, Based on the geological information, the occurrence risk of each type of geological hazard is calculated using a pre-constructed risk assessment formula to obtain pre-statistical geological hazard risk data, including: Construct a collapse risk probability formula; the collapse risk probability formula is as follows: in, H a =ω1×H a1 +ω2×H a2 +ω3×H a3 ;f ak =1.2×V s -200; P1 represents the probability of collapse risk; a0, a1, a2, and a3 are the basic regression coefficient, the first regression coefficient, the second regression coefficient, and the third regression coefficient, respectively; R1 is the collapse risk value; Δh is the monthly variation of the groundwater level; L t ω is the distance from the fault; f ω h ω Ha ω n These are, respectively, the weights for bearing capacity, water depth, activity intensity, and porosity; f ak H represents the natural bearing capacity of the soil and rock foundation; h represents the depth of the groundwater level; H a The comprehensive index of human activity intensity; n represents the porosity of the soil and rock mass; ω1, ω2, and ω3 are the weights for mining activities, underground engineering, and the proportion of arable land, respectively; H a1 H a2 H a3 These are the impact coefficients for mining activities, underground engineering, and arable land proportion, respectively; V s Shear wave velocity of the rock and soil mass; A landslide risk probability formula is constructed; the landslide risk probability formula is: P2=1-exp(-λ×R2); where, P2 represents the probability of landslide risk; λ is the Poisson correction coefficient; R2 is the landslide risk value; ω α ω Hsl ω Fs ω P5 ω NDVI These are the weights for slope, height, stability, cumulative rainfall, and vegetation index, respectively; α is the slope; H sl F represents the slope height. s P5 represents the stability coefficient of the soil and rock mass; P5 represents the cumulative rainfall over 5 consecutive days; NDVI represents the normalized difference in vegetation index; γ i The natural unit weight of the rock and soil mass; h i φ represents the thickness of the soil and rock mass. i γ is the internal friction angle of the rock and soil; sat The saturated unit weight of the soil and rock mass; i The length of the sliding surface; Construct a debris flow risk probability formula; the debris flow risk probability formula is as follows: Where R3 represents the probability of debris flow risk; ω Ws ω I1 ω J These are the weights for sediment source, maximum rainfall, and longitudinal gradient, respectively; W s I1 represents the debris flow source volume; I1 represents the maximum hourly rainfall; J represents the longitudinal gradient of the valley. The geological information is calculated using the collapse risk probability formula, the landslide risk probability formula, and the debris flow risk probability formula to obtain the pre-statistical geological hazard risk data; the pre-statistical geological hazard risk data includes: collapse probability data, landslide probability data, and debris flow probability data.

4. The geological ecological restoration assessment method for geological disaster prevention and control according to claim 1, characterized in that, The multimodal proximity sensing model includes: an input layer, a feature extraction layer, a proximity sensing embedding layer, a multimodal cross-attention fusion layer, and a multi-target output layer connected in sequence.

5. The geological ecological restoration assessment method for geological disaster prevention and control according to claim 1, characterized in that, Based on the original geological hazard risk data, the post-remediation geological hazard risk data is classified according to preset remediation effect evaluation rules to obtain geological ecological restoration evaluation results, including: The original geological hazard risk data and all the restored geological hazard risk data are classified according to the type of hazard to obtain single-indicator statistical data; the data in the single-indicator statistical data are arranged in chronological order. Extract the difference between the first and last data points in the single-indicator statistical data to obtain the risk reduction before and after the repair. Linear fitting is performed on the single-indicator statistical data to obtain a fitted curve, and the slope of the fitted curve is extracted to obtain the slope of the risk change trend. A pre-constructed dynamic weighting formula is used to weight the risk reduction before and after the repair and the slope of the risk change trend to obtain a comprehensive evaluation score; the expression of the dynamic weighting formula is: Wherein, S is the comprehensive evaluation score; ΔR is the risk reduction before and after the repair; K is the adjustment constant; R0 is the first data in the single indicator statistics; and k is the slope of the risk change trend. The comprehensive evaluation score is classified according to the preset scoring interval division standard to obtain the geological and ecological restoration evaluation result.

6. The geological ecological restoration assessment method for geological disaster prevention and control according to claim 2, characterized in that, The pre-statistical geological hazard risk data, the multi-dimensional remote sensing images, and the ultrasonic detection data are preprocessed to obtain raw training data, including: The multidimensional remote sensing image is cropped according to the original grid division boundary to obtain a standard-sized image; Use the coordinates corresponding to the center of the standard-size image as the alignment label of the standard-size image; The ultrasonic detection data is denoised using wavelet transform and corrected using the propagation velocity parameters of karst rock mass to obtain calibrated detection data. The coordinates corresponding to the calibration probe data center are used as the alignment labels of the calibration probe data. The standard-sized images and the calibration probe data with the same alignment labels are fused to obtain the alignment data. The original training data is obtained by fusing the aligned data and the pre-statistical geological disaster risk data.

7. A geological ecological restoration assessment method for geological disaster prevention and control according to any one of claims 4 and 6, characterized in that, The multimodal proximity sensing model is trained using the original training data to obtain a geological hazard risk prediction model, including: The input layer receives the original training data and assigns a block index to each group of aligned data. In the feature extraction layer, ResNet-50 with embedded dilated convolution and residual attention is used to extract features from the standard-size image to obtain image features, and MLP with embedded BN and Dropout is used to extract features from the calibration detection data to obtain numerical features. In the proximity-aware embedding layer, the block indices of the eight neighboring blocks adjacent to the current block are matched, and the image features of the neighboring blocks are fused with the image features of the current block to obtain image-aware enhanced features. The numerical features of the neighboring blocks are fused with the numerical features of the current block to obtain numerical-aware enhanced features. The image perception enhancement feature is set as the query vector, and the numerical perception enhancement feature is set as the key vector and the value vector. Based on the cross-attention mechanism, the multimodal cross-attention fusion layer is used to calculate the weights of the query vector, the key vector, and the value vector and perform weighted fusion to obtain cross-modal fusion features. The multi-objective output layer is used to perform multi-objective probability mapping on the cross-modal fusion features to obtain the model prediction results.

8. A geological ecological restoration assessment system for geological disaster prevention and control, characterized in that, include: The research data acquisition module is used to collect geological information, multi-dimensional remote sensing images, and ultrasonic detection data of the target karst landform research area. The type setting module is used to set the type of geological disaster; The types of geological hazards include: collapse hazards, landslide hazards, and debris flow hazards; The initial assessment module is used to calculate the occurrence risk of each type of geological hazard based on the geological information using a pre-built risk assessment formula, and obtain pre-statistical geological hazard risk data; The preprocessing module is used to preprocess the pre-statistical geological disaster risk data, the multi-dimensional remote sensing image, and the ultrasonic detection data to obtain the original training data. The model building module is used to train the multimodal proximity perception model using the original training data to obtain a geological disaster risk prediction model; The pre-repair risk determination module is used to determine the target karst landform monitoring area and the original geological hazard risk data of the target karst landform monitoring area before geological and ecological restoration; The post-repair information acquisition module is used to repeatedly acquire the multi-dimensional remote sensing images and ultrasonic detection data of the target karst landform monitoring area at fixed intervals after geological and ecological restoration, so as to obtain post-repair environmental information; the fixed intervals include any one of 10 days, 15 days, and 20 days. The risk prediction module is used to sequentially input the remediated environmental information into the geological hazard risk prediction model to obtain remediated geological hazard risk data. The effect evaluation module is used to classify the effect of the restored geological disaster risk data according to the original geological disaster risk data using preset restoration effect evaluation rules, and obtain the geological ecological restoration evaluation results.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the processor; wherein the memory stores instructions executable by the processor, the instructions being executed by the processor to enable the processor to perform a geological ecological restoration assessment method for geological disaster prevention and control as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute a geological ecological restoration assessment method for geological disaster prevention and control, as described in any one of claims 1 to 7.

Images