A method and system for predicting rock pile growth and development
Through deep learning model, the edge of rock piles is identified and combined with the improved cold and frozen weathering model for prediction, the problem of low statistical efficiency and accuracy of rock piles is solved, timely and accurate prediction of rock pile growth and development is achieved, and the engineering is improved Operational security of facilities.
Patent Information
- Application Number
- CN202410249396.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-03-05
AI Technical Summary
In the prior art, the statistical efficiency and accuracy of rock piles are low, which is difficult to meet the statistical needs of large-scale mapping. There is no relevant technology to pay attention to and monitor the growth and development of rock piles, resulting in the inability to predict their growth and development in a timely and accurate manner. The process poses a threat to the operational safety of engineering facilities.
The satellite image recognition model based on deep learning is used to extract the edge profile of rock piles. Combined with the improved cold-frozen weathering and rock strength models, the long-term temperature inversion sequence is introduced to calculate the cold-frozen weathering intensity by analyzing the relationship between cold-frozen weathering intensity and altitude. For weathering and rock deterioration intensity, critical algorithms are used to predict long-term scales.
It improves the accuracy and efficiency of rock pile recognition and extraction, reveals the rock pile formation mechanism, extends the service life of the engineering facilities, improves operational stability, and provides prevention and control measures to reduce manual intervention and errors.
Smart Images

Figure CN118155086B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of rock pile prediction, and more specifically to a method and system for predicting the growth and development of rock piles in alpine areas. Background Art
[0002] In high-altitude mountainous areas, the traditional method for counting rock piles mainly uses traditional manual inspections, which has low statistical efficiency and accuracy, and can only be used for small-scale statistics on certain sections of the road, making it difficult to meet the statistical needs of large-scale mapping.
[0003] As the temperature continues to rise in recent years, the temperature fluctuation has gradually increased, resulting in the continuous expansion of the original rock pile fragmentation range, increasing the difficulty of rock pile statistics, and forming new high-risk disaster areas. At present, there is no relevant technology or method to pay attention to and monitor this process. In addition, the existing technology cannot form a system that integrates the identification and extraction of rock piles, the analysis of formation methods, and the prediction of growth and development, resulting in the inability to effectively predict and control the growth and development of rock piles, making the threats to the operation safety of engineering facilities such as roads and bridges increasingly aggravated.
[0004] In summary, the statistical efficiency and accuracy of existing technologies for rock piles are low, which makes it difficult to meet the statistical needs of large-scale mapping. There is no relevant technology to pay attention to and monitor this process, resulting in the inability to make timely and accurate predictions on the growth and development process of rock piles, which has a serious impact on the operational safety of engineering facilities. Summary of the invention
[0005] In response to the problems existing in the above-mentioned fields, the present invention proposes a method and system for predicting the growth and development of rock piles, which can solve the technical problems that the statistical efficiency and accuracy of rock piles in the existing technology are low, it is difficult to meet the statistical needs of large-scale mapping, and there is no relevant technology to pay attention to and monitor this process, resulting in the inability to make timely and accurate predictions on the growth and development of rock piles, which has a serious impact on the operational safety of engineering facilities.
[0006] In order to solve the above technical problems, the present invention discloses a method for predicting the growth and development of a rock mass, comprising the following steps:
[0007] Collect rock pile image data in the study area, extract the edge contour height of the rock pile through the recognition model used for rock pile image recognition, and obtain the rock pile image recognition result of the study area;
[0008] According to the results of rock pile image recognition, the strength ratio parameter is added to the rock strength degradation calculation equation, and the plateau rock strength value is corrected to obtain an improved rock strength degradation curve relationship. By analyzing the relationship between frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering curve relationship to obtain an improved frost weathering curve relationship.
[0009] The improved rock strength degradation curve relationship comprises the following steps:
[0010] The strength ratio parameter θ is added to the original rock strength degradation calculation equation to correct the plateau rock strength value. The calculation formula is:
[0011]
[0012] Where h is the rock depth; h0 is the depth corresponding to the symmetric center of the curve; b is the shape function coefficient; S0(t) is the shear strength parameter of rock and soil at time t; S H is the lower limit strength value of rock weathering; θ is the rock strength ratio parameter;
[0013] The improved frost weathering curve relationship equation comprises the following steps:
[0014] By analyzing the relationship between the intensity of frost weathering and altitude, it is found that with the increase of temperature, the total frost weathering intensity at the same altitude also shows an upward trend; under the same total frost weathering intensity, the altitude range also expands;
[0015] A long-term temperature inversion sequence is introduced into the rock strength degradation curve relationship. Through this temperature inversion sequence, the freezing weathering of rocks in the past few years is calculated to obtain the average weathering rate.
[0016] The relationship between the temperature and depth of the frost weathering basic curve is:
[0017]
[0018] Where t is the time in days, T(z,t) is the temperature at different times and depths; T mα is the annual average temperature; T α is half of the annual temperature variation; α is the thermal diffusion coefficient; P y Refers to the cycle of temperature changes;
[0019] Improved frost weathering curve relationship:
[0020]
[0021] In the formula, F ciis the frost weathering intensity, which refers to the integral of the temperature gradient over time in the frost weathering window; D is the distance from 0℃ to -5℃; D * is the length index;
[0022] According to the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship, the frost weathering intensity, rock degradation intensity and time effect factor are calculated;
[0023] The calculation of the frost weathering intensity, rock degradation intensity and time effect factor comprises the following steps:
[0024] According to the relationship between the temperature and depth of frost weathering and the improved relationship between the frost weathering curve, by calculating the temperature lapse rate and the altitude where the annual average temperature is 0℃, it is found that the formation of condensed ice is directly related to the intensity of frost weathering.
[0025] The improved frost weathering curve relationship is used to simulate the frost weathering in a year at a fixed T mα Under the condition of temperature at different depths every day, judge whether it is a situation where condensed ice can grow;
[0026] According to the frost weathering intensity, rock degradation intensity and time effect factor, the key algorithm is used to expand the time series, analyze the change trend of frost weathering intensity-altitude over time, and make a long-term prediction of the change trend of the rock mass in the study area to obtain the prediction results;
[0027] The method of obtaining the prediction result comprises the following steps:
[0028] By calculating the intensity of frost weathering and rock degradation in the study area, the specific reasons and related data affecting the rock fragmentation and formation of adverse geological phenomena in the study area are obtained, and the dimensionality reduction evolution rate of the rock mass in the study area, that is, the time effect factor, is obtained;
[0029] Based on the calculated frost weathering intensity, rock degradation intensity and time effect factor, key algorithms are used to expand the time series and make long-term predictions on the changing trends of the rock mass in the study area.
[0030] Preferably, the image data of the rock mass in the study area collected includes satellite images and remote sensing data; wherein the satellite images are obtained by using field photos or Landsat satellite products, and the remote sensing data are obtained by using field photos.
[0031] Preferably, obtaining the rock pile image recognition result of the study area comprises the following steps:
[0032] Construct training set data based on the collected rock pile image data in the research area;
[0033] A recognition model for rock pile image recognition is established, namely, a satellite image rock pile recognition model based on deep learning; wherein, the satellite image rock pile recognition model uses the semantic segmentation network DeepLab as the skeleton, introduces the void ratio in the convolution operation of the void convolution; introduces a multi-scale feature fusion strategy, and obtains image features from multiple scales by adopting multiple parallel branch networks, combining void convolutions of different sizes and pooling operations; introduces an encoder-decoder structure to extract the edge contour height of the rock pile in the satellite image;
[0034] Specifically, the semantic segmentation network DeepLab is used as the skeleton, by introducing the dilation rate in the convolution operation of the dilated convolution, the separation step is 1, and the convolution kernel size is 3, which is used to capture the detailed information in the satellite image and process the small objects and edges in the image;
[0035] Multi-scale feature fusion is introduced, and the introduced multi-scale information is used as another important part of the semantic segmentation network. By adopting multiple parallel branch networks, combined with different sizes of hole convolution and pooling operations, the field of view is 512+256+128+64, respectively, to obtain image features from multiple scales;
[0036] The encoder-decoder structure is introduced. The encoder part uses the convolutional neural network ResNet as the basic model to extract image features. The decoder part restores the spatial resolution of the segmentation result by upsampling and fusing multi-scale features.
[0037] The semantic segmentation network smoothes and refines the segmentation results by using the conditional random field (CRF) post-processing technology.
[0038] Preferably, constructing the training set data comprises the following steps:
[0039] Perform image preprocessing on satellite images, set the network input image channel size to 512×512, set the semantic segmentation category to 2, and the semantic segmentation category includes rock piles and ordinary mountains;
[0040] Inject category labels into the processed satellite images used for training, obtain the original image files and annotated images, and record the TXT files of the training and test file names.
[0041] Preferably, the training process of the satellite image rock pile recognition model based on deep learning comprises the following steps:
[0042] By customizing the number of cyclic training, the number of training packages is 2, changing the training parameters and the number of training times, and obtaining the model corresponding to the optimal network parameters after the training, the trained satellite image rock pile recognition model is obtained, wherein the network parameters include pixel accuracy, average pixel accuracy, average intersection-over-combination ratio and frequency-weighted intersection-over-combination ratio;
[0043] The trained deep learning-based satellite image rock pile recognition model was used for testing to obtain the rock pile contour height information in the collected rock pile images.
[0044] Preferably, a rock mass growth prediction system is also included, comprising:
[0045] The rock pile image information acquisition module is used to collect rock pile image data in the study area, extract the rock pile edge contour height through the recognition model for rock pile image recognition, and obtain the rock pile image recognition result in the study area;
[0046] The rock strength correction module is used to add the strength ratio parameter to the rock strength degradation calculation equation according to the rock pile image recognition result, correct the plateau rock strength value, and obtain the improved rock strength degradation curve relationship; by analyzing the relationship between the frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering action curve relationship to obtain the improved frost weathering action curve relationship; based on the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship, the frost weathering intensity, rock degradation intensity and time effect factor are calculated;
[0047] The change trend prediction module is used to expand the time series based on the frost weathering intensity, rock degradation intensity and time effect factor through key algorithms, analyze the change trend of frost weathering intensity-altitude over time, and make long-term predictions on the change trend of the rock mass in the study area to obtain prediction results;
[0048] The improved rock strength degradation curve relationship comprises the following steps:
[0049] The strength ratio parameter θ is added to the original rock strength degradation calculation equation to correct the plateau rock strength value. The calculation formula is:
[0050]
[0051] Where h is the rock depth; h0 is the depth corresponding to the symmetric center of the curve; b is the shape function coefficient; S0(t) is the shear strength parameter of rock and soil at time t; S H is the lower limit strength value of rock weathering; θ is the rock strength ratio parameter;
[0052] The improved frost weathering curve relationship equation comprises the following steps:
[0053] By analyzing the relationship between the intensity of frost weathering and altitude, it is found that with the increase of temperature, the total frost weathering intensity at the same altitude also shows an upward trend; under the same total frost weathering intensity, the altitude range also expands;
[0054] A long-term temperature inversion sequence is introduced into the rock strength degradation curve relationship. Through this temperature inversion sequence, the freezing weathering of rocks in the past few years is calculated to obtain the average weathering rate.
[0055] The relationship between the temperature and depth of the frost weathering basic curve is:
[0056]
[0057] Where t is the time in days, T(z,t) is the temperature at different times and depths; T mα is the annual average temperature; T α is half of the annual temperature variation; α is the thermal diffusion coefficient; P y Refers to the cycle of temperature changes;
[0058] Improved frost weathering curve relationship:
[0059]
[0060] In the formula, F ci is the frost weathering intensity, which refers to the integral of the temperature gradient over time in the frost weathering window; D is the distance from 0℃ to -5℃; D * is the length index;
[0061] The calculation of the frost weathering intensity, rock degradation intensity and time effect factor comprises the following steps:
[0062] According to the relationship between the temperature and depth of frost weathering and the improved relationship between the frost weathering curve, by calculating the temperature lapse rate and the altitude where the annual average temperature is 0℃, it is found that the formation of condensed ice is directly related to the intensity of frost weathering.
[0063] The improved frost weathering curve relationship is used to simulate the frost weathering in a year at a fixed T mα Under the condition of temperature at different depths every day, judge whether it is a situation where condensed ice can grow;
[0064] The method of obtaining the prediction result comprises the following steps:
[0065] By calculating the intensity of frost weathering and rock degradation in the study area, the specific reasons and related data affecting the rock fragmentation and formation of adverse geological phenomena in the study area are obtained, and the dimensionality reduction evolution rate of the rock mass in the study area, that is, the time effect factor, is obtained;
[0066] Based on the calculated frost weathering intensity, rock degradation intensity and time effect factor, key algorithms are used to expand the time series and make long-term predictions on the changing trends of the rock mass in the study area.
[0067] Compared with the prior art, the present invention has the following beneficial effects:
[0068] The present invention breaks through the technical defects of low statistical efficiency and accuracy of traditional statistical methods for rock piles. This method can reduce and lower human intervention and errors, and improve the accuracy and efficiency of identification and extraction by constructing a satellite image rock pile identification model based on deep learning; by predicting the changing trend of the rock pile in the study area on a long-term scale, it can reveal the formation mechanism of the rock pile and prevent it, extend the service life of the project, and improve its operational stability. In the prediction process, this method improves the original frost weathering model and rock strength model, calculates the frost weathering intensity and rock strength of the study area, analyzes the changing trend of frost weathering intensity-altitude over time, and finally obtains the specific reasons and related data that affect the rock crushing and formation of adverse geological phenomena in the area, and obtains the dimensionality reduction evolution rate of the rock pile in the area, which is conducive to taking preventive measures. BRIEF DESCRIPTION OF THE DRAWINGS
[0069] Figure 1 It is a schematic diagram of the overall method flow of the present invention;
[0070] Figure 2 It is a schematic diagram of the extraction results of the satellite image rock pile identification model based on deep learning of the present invention;
[0071] Figure 3 A relationship diagram between the cold weathering intensity and the altitude of the present invention;
[0072] Figure 4 It is a time-temperature inversion sequence diagram of the present invention;
[0073] Figure 5 The relationship diagram between the formation of condensed ice and the intensity of frost weathering of the present invention is shown in FIG.
[0074] Figure 6 The network architecture of the satellite image rock pile identification model based on deep learning of the present invention;
[0075] Figure 7 It is a schematic diagram of the variation trend of the frost weathering intensity under the long time series of the present invention;
[0076] Figure 8 It is the rock degradation strength model curve of the present invention. DETAILED DESCRIPTION
[0077] The following will be combined with the attached embodiment of the present invention Figure 1-8 , the technical solutions in the embodiments of the present invention are clearly and completely described. It should be understood that the terms described in the present invention are only used to describe specific implementation methods and are not used to limit the present invention.
[0078] like Figure 1 As shown, the present invention discloses a method for extracting features, analyzing formation modes and predicting growth and development of surrounding rock piles during construction or operation of an engineering project in a high-cold mountainous area. The method comprises the following steps:
[0079] S1: Collect photographic images and remote sensing images;
[0080] S2: Preprocess the images, including cropping and labeling, to construct the training set;
[0081] S3: Construct a satellite image rock pile recognition model based on deep learning, using the semantic segmentation network DeepLab as the skeleton, and introduce the void ratio in the convolution operation of the void convolution; introduce a multi-scale feature fusion strategy, by using multiple parallel branch networks, combining void convolutions of different sizes and pooling operations, to obtain image features from multiple scales; introduce an encoder-decoder structure to extract the edge contour height of the rock pile in the satellite image;
[0082] Use the training set to train the satellite image rock pile recognition model based on deep learning, and change the training parameters and the number of training times;
[0083] S4: Evaluate the accuracy of the training results and use the model with the best training results to conduct follow-up research;
[0084] S5: inputting the image of the study area, i.e., the road section to be extracted and analyzed, into the model of step S4 to automatically extract image information;
[0085] S6: Calculate the frost weathering intensity, rock degradation intensity, and time effect factor. Remote sensing satellite images or photos collected on site can be used to automatically identify and extract the rock pile, and calculate the frost weathering intensity, rock degradation intensity, and dimensionality reduction evolution rate of the area to achieve the purpose of prediction and prevention.
[0086] Specifically, the image in step S1 can come from a remote sensing satellite image or a photo taken on site.
[0087] In steps S2-S4, the input image is processed according to the procedure, including cropping and labeling, to form a training set and after multiple rounds of iterative training, the model with the best training effect is selected for subsequent research. The satellite image rock pile recognition model is trained using the training set; the remote sensing image of the study area is input into the trained satellite image rock pile recognition model to extract the image information of the rock pile in the study area, such as Figure 2 Shown are the extraction results of the rock pile identification model from satellite images.
[0088] In steps S5 and S6, the method will use the model with the best training effect in step S4 for subsequent extraction and prediction, and will calculate the frost weathering intensity, rock degradation intensity and dimensionality reduction evolution rate of the study area. If the temperature data of previous years is input, the changing trend of frost weathering intensity under long-term series can also be simulated, which is conducive to prediction and prevention.
[0089] According to the results of rock pile image recognition, the strength ratio parameter is added to the rock strength degradation calculation equation, and the plateau rock strength value is corrected to obtain an improved rock strength degradation curve relationship. By analyzing the relationship between frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering curve relationship to obtain an improved frost weathering curve relationship.
[0090] Based on the extracted image information of the rock pile in the study area and combined with the temperature data of the study area over the years, the frost weathering intensity, rock degradation intensity and time effect factor are calculated through the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship; and the key algorithm is used to expand the time series, analyze the changing trend of frost weathering intensity-altitude over time, and make a long-term prediction of the changing trend of the rock pile in the study area.
[0091] The relationship between the temperature and depth of the frost weathering basic curve is:
[0092]
[0093] Where t is the time in days, T(z,t) is the temperature at different times and depths; T mα is the annual average temperature; T α is half of the annual temperature variation; α is the thermal diffusion coefficient; P y Refers to the cycle of temperature change, which is 365 days.
[0094] Improved frost weathering curve relationship:
[0095]
[0096] In the formula, F ciis the frost weathering intensity, which refers to the integral of the temperature gradient over time in the frost weathering window; D is the distance from 0℃ to -5℃; D * is the length index, set to 50cm.
[0097] Firstly, the relationship between the intensity of frost weathering and altitude was calculated, such as Figure 3 As shown in Figure 2, with the increase of temperature, the total frost weathering intensity also shows an upward trend at the same altitude (up to 12℃·cm -1 ), and the altitude range with the same total frost weathering intensity level also expands. According to the gray dotted line in the figure (take = 8℃·cm -1 ), the total frost weathering intensity peak appears in a wider altitude range, from the initial [4300m, 4700m] to [3800m, 5350m], which further leads to the increase of the multi-year average frost weathering intensity.
[0098] A long-term temperature inversion sequence (45,000 years ago to the present) is introduced into the relationship between the cold weathering curve. Through this temperature inversion sequence, Figure 4 As shown in the figure, the frost weathering of rocks in the past few years can be calculated, and the average weathering rate can be obtained to be (3.4~5.67)×10 -8 a -1 .
[0099] Improved rock strength degradation curve relationship:
[0100]
[0101] Where h is the rock depth; h0 is the depth corresponding to the symmetric center of the curve; b is the shape function coefficient; S0(t) is the shear strength parameter of rock and soil at time t; S H is the lower limit strength value of rock weathering; θ is the strength ratio parameter, and θ=0.8 is taken based on experience.
[0102] By calculating the intensity of frost weathering and rock degradation in the study area, the specific causes and related data affecting rock fragmentation and the formation of adverse geological phenomena in the study area were obtained, and the dimensionality reduction evolution rate of the rock mass in the study area, namely the time effect factor, was obtained.
[0103] like Figure 5 As shown, it simulates a fixed T mα Under the condition of temperature at different depths every day, we can judge whether it is a situation where condensed ice can grow. The relationship between the formation of condensed ice and depth is as follows: mα When it is a positive number, look at the time period of one year, when t takes a value and T (z,t)In the freezing weathering window, check whether there is a positive temperature gradient. If so, add 1 to the number of days for the growth of condensed ice. Similarly, when T MA When it is negative, look at the time period of one year, when t takes a value and T (z,t) In the freezing weathering window, is there a negative temperature gradient? If so, the number of days for the growth of condensed ice is increased by 1 (but one thing needs to be noted in this case, the growth of condensed ice requires liquid water, so in this case, it is necessary to add a rule when t takes this value, and there is T in the range of 1-1000cm at depth z. (z,t) For example, in the figure, t = 91 (T MA is a negative number), there is T in the range of 1-1000cm on that day (z,t) When z=400, it is in the frost weathering window and there is a negative temperature gradient. At this time, the number of days of partial ice growth at z=400 is recorded as 1. Therefore, the formation of partial ice is directly related to the intensity of frost weathering.
[0104] The present invention also discloses a rock mass growth prediction system, comprising:
[0105] The rock pile image information acquisition module is used to collect rock pile image data in the study area, extract the rock pile edge contour height through the recognition model for rock pile image recognition, and obtain the rock pile image recognition result in the study area;
[0106] The rock strength correction module is used to add the strength ratio parameter to the rock strength degradation calculation equation according to the rock pile image recognition result, correct the plateau rock strength value, and obtain the improved rock strength degradation curve relationship; by analyzing the relationship between the frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering action curve relationship to obtain the improved frost weathering action curve relationship; based on the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship, the frost weathering intensity, rock degradation intensity and time effect factor are calculated;
[0107] The change trend prediction module is used to expand the time series through key algorithms according to the frost weathering intensity, rock degradation intensity and time effect factors, analyze the change trend of frost weathering intensity-altitude over time, make long-term predictions on the change trend of the rock mass in the study area, and obtain prediction results.
[0108] The rock pile growth and development prediction method and system proposed in the present application can effectively identify and extract rock piles in remote sensing images or photos, and timely calculate and predict their growth and development laws and trends, so as to provide targeted and effective prevention and control plans. The method is easy to operate, accurate and efficient, low-cost, can effectively reduce human errors, and is real-time and targeted.
[0109] Example
[0110] The embodiment of the present invention provides a method for extracting features, analyzing formation modes, and predicting growth and development of surrounding rock piles during construction or operation of an engineering project in a high-cold mountainous area, comprising the following steps:
[0111] Step 1: Collect photographic images and remote sensing images using field photos or Landsat satellite products.
[0112] Step 2: Preprocess the collected images, mainly by using the computer server to segment, crop and annotate the images to form a training set for training the deep learning model.
[0113] Step three is the stage of training, distinguishing, and using deep learning models, which also belongs to the recognition and extraction module in the system.
[0114] During this stage, in order to facilitate rock pile identification and data collection and statistics, a satellite image rock pile identification model based on deep learning was constructed. The semantic segmentation model DeepLab was used as the skeleton, integrating void convolution, multi-scale feature fusion and encoder-decoder structure to extract the edge contour height of rock piles in satellite images, which is conducive to data statistics and predictive analysis of rock pile distribution and area.
[0115] The specific network structure of this model is as follows Figure 6 As shown in the figure, the core of the satellite image rock pile recognition model based on deep learning is the dilated convolution technology. By introducing the dilated rate in the convolution operation of the dilated convolution, with a separation step of 1 and a convolution kernel size of 3, the effective field of view of the convolution layer can be increased while maintaining the size of the receptive field, thereby better capturing the detailed information in the satellite image and better processing the small objects and edges in the image.
[0116] Multi-scale information is introduced as another important part of the network. The semantic segmentation network uses multiple parallel branch networks, combined with different sizes of hole convolution and pooling operations, with fields of view of 512+256+128+64, to obtain image features from multiple scales. This multi-scale feature fusion strategy helps to improve the precision and accuracy of the segmentation results and overcome the information loss and ambiguity that are prone to occur at a single scale.
[0117] An encoder-decoder structure is introduced, in which the encoder part uses the convolutional neural network ResNet as the basic model to extract image features; the decoder part restores the spatial resolution of the segmentation result by upsampling and fusing multi-scale features.
[0118] Finally, the semantic segmentation network further improves the segmentation results by using post-processing techniques such as conditional random fields (CRF). CRF can smooth and refine the segmentation results by considering the contextual relationship between pixels, thereby obtaining more accurate segmentation boundaries.
[0119] The specific processing steps are:
[0120] 1. Perform image preprocessing on satellite images, set the network input image channel size to 512*512, and set the semantic segmentation category to 2 (rock piles and ordinary mountains);
[0121] 2. Creation of training data set: Inject category labels into training images to obtain original image files and annotated images, as well as TXT files recording training and test file names.
[0122] 3. Train the semantic segmentation network. The number of cyclic training cycles is customized. The number of training packages is 2. After the training, the network parameters are obtained. The specific evaluation indicators are shown in Table 1.
[0123] Table 1. Evaluation indicators of deep learning network extraction accuracy
[0124]
[0125] 4. Use the trained deep learning-based satellite image rock pile recognition model for verification, and the rock pile contour information in the obtained image is convenient for the following calculation and analysis.
[0126] As shown in Table 2, the accuracy of rock mass extraction and identification results along some sections of G219 was evaluated using the model.
[0127] Table 2. Test results of deep learning network extraction accuracy
[0128]
[0129]
[0130] Step 4: According to the basic meteorological data of the study area, the temperature lapse rate and the altitude where the annual average temperature is 0°C are automatically calculated. By improving the original frost weathering model and rock strength model, the frost weathering intensity and rock strength of the area are calculated. Figure 7 The results show the temporal variation trend of frost weathering intensity and altitude from 420 ka to the present. The frost weathering intensity shows a single peak with altitude, which is always maintained at 8-9℃·cm -1The maximum values appear in the range of 3800m to 5200m. Due to the rapid rise of global temperature in 16ka and the maintenance of relatively high temperature for a period of time, the annual temperature variation increased during this period, resulting in the cold weathering intensity in this area jumping to 11-12.5℃·cm -1 , but the curve change trend is still basically consistent with other time scales.
[0131] Figure 8 It is the rock degradation strength model curve, which is used to analyze the influence of single factor change on S0(t) under time and depth effect. The range of S0(t) is 20-320kPa. When other factors remain unchanged, as S0(t) continues to deteriorate, the strength change transition interval gradually increases and the change amplitude increases, and finally approaches 280kpa and remains stable. This phenomenon proves that weathering is directly related to time.
[0132] Finally, the specific reasons and related data that affect the rock fragmentation and formation of adverse geological phenomena in the region are obtained, and the dimensionality reduction evolution rate of the rock mass in the region, that is, the time effect factor, is obtained, which is conducive to the adoption of prevention and control measures. The key algorithm is used to extend the time series, so that the model can show the change trend of the cold weathering intensity-altitude over time from 420ka to the present, so as to study the dimensionality reduction evolution rate of the rock in the study area on a long time scale, and thus provide prevention and control measures suitable for various projects in the study area.
[0133] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.
[0134] In addition, unless otherwise specified, all technical and scientific terms used in the present invention have the same meanings as commonly understood by those skilled in the art to which the present invention belongs. All documents mentioned in this specification are incorporated by reference to disclose and describe the methods related to the documents. In the event of any conflict with any incorporated document, the content of this specification shall prevail.
Claims
1. A method for predicting the growth and development of a rock mass, characterized in that: The following steps are involved: Collect rock pile image data in the study area, extract the edge contour height of the rock pile through the recognition model used for rock pile image recognition, and obtain the rock pile image recognition result of the study area; According to the results of rock pile image recognition, the strength ratio parameter is added to the rock strength degradation calculation equation, and the plateau rock strength value is corrected to obtain an improved rock strength degradation curve relationship. By analyzing the relationship between frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering curve relationship to obtain an improved frost weathering curve relationship. The improved rock strength degradation curve relationship comprises the following steps: The strength ratio parameter θ is added to the original rock strength degradation calculation equation to correct the plateau rock strength value. The calculation formula is: Where h is the rock depth; h0 is the depth corresponding to the center of the curve; b is the shape function coefficient; S0(t) is the shear strength parameter of rock and soil at time t; S H is the lower limit strength value of rock weathering; θ is the rock strength ratio parameter; The improved frost weathering curve relationship equation comprises the following steps: By analyzing the relationship between the intensity of frost weathering and altitude, it is found that with the increase of temperature, the total frost weathering intensity at the same altitude also shows an upward trend; under the same total frost weathering intensity, the altitude range also expands; A long-term temperature inversion sequence is introduced into the rock strength degradation curve relationship. Through this temperature inversion sequence, the freezing weathering of rocks in the past few years is calculated to obtain the average weathering rate. The relationship between the temperature and depth of the frost weathering basic curve is: Where t is the time in days, T(z,t) is the temperature at different times and depths; T mα is the annual average temperature; T α is half of the annual temperature variation; α is the thermal diffusion coefficient; P y Refers to the cycle of temperature changes; Improved frost weathering curve relationship: In the formula, F ci is the frost weathering intensity, which refers to the integral of the temperature gradient over time in the frost weathering window; D is the distance from 0℃ to -5℃; D * is the length index; According to the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship, the frost weathering intensity, rock degradation intensity and time effect factor are calculated; The calculation of the frost weathering intensity, rock degradation intensity and time effect factor comprises the following steps: According to the relationship between the temperature and depth of frost weathering and the improved relationship between the frost weathering curve, by calculating the temperature lapse rate and the altitude where the annual average temperature is 0℃, it is found that the formation of condensed ice is directly related to the intensity of frost weathering. The improved frost weathering curve relationship is used to simulate the frost weathering in a year at a fixed T mα Under the above conditions, the temperature corresponding to different depths every day is used to determine whether the condensation ice can grow; According to the frost weathering intensity, rock degradation intensity and time effect factor, the key algorithm is used to expand the time series, analyze the change trend of frost weathering intensity-altitude over time, and make a long-term prediction of the change trend of the rock mass in the study area to obtain the prediction results; The method of obtaining the prediction result comprises the following steps: By calculating the intensity of frost weathering and rock degradation in the study area, the specific reasons and related data affecting the rock fragmentation and formation of adverse geological phenomena in the study area are obtained, and the dimensionality reduction evolution rate of the rock mass in the study area, that is, the time effect factor, is obtained; Based on the calculated frost weathering intensity, rock degradation intensity and time effect factor, key algorithms are used to expand the time series and make long-term predictions on the changing trends of the rock mass in the study area.
2. The method for predicting rock mass growth and development according to claim 1, characterized in that: The image data of the rock pile body collected in the study area includes satellite images and remote sensing data; wherein the satellite images are obtained by using photos taken on the spot or Landsat satellite products, and the remote sensing data are obtained by using photos taken on the spot.
3. The method for predicting rock mass growth and development according to claim 2, characterized in that: The method of obtaining the rock pile image recognition result in the study area comprises the following steps: Construct training set data based on the collected rock pile image data in the study area; A recognition model for rock pile image recognition is established, namely, a satellite image rock pile recognition model based on deep learning; wherein, the satellite image rock pile recognition model uses the semantic segmentation network DeepLab as the skeleton, introduces the void ratio in the convolution operation of the void convolution; introduces a multi-scale feature fusion strategy, and obtains image features from multiple scales by adopting multiple parallel branch networks, combining void convolutions of different sizes and pooling operations; introduces an encoder-decoder structure to extract the edge contour height of the rock pile in the satellite image; Specifically, the semantic segmentation network DeepLab is used as the skeleton, by introducing the dilation rate in the convolution operation of the dilated convolution, the separation step is 1, and the convolution kernel size is 3, which is used to capture the detailed information in the satellite image and process the small objects and edges in the image; Multi-scale feature fusion is introduced, and the introduced multi-scale information is used as another important part of the semantic segmentation network. By adopting multiple parallel branch networks, combined with different sizes of hole convolution and pooling operations, the field of view is 512+256+128+64, respectively, to obtain image features from multiple scales; The encoder-decoder structure is introduced. The encoder part uses the convolutional neural network ResNet as the basic model to extract image features. The decoder part restores the spatial resolution of the segmentation result by upsampling and fusing multi-scale features. The semantic segmentation network smoothes and refines the segmentation results by using the conditional random field (CRF) post-processing technology.
4. The method for predicting rock mass growth and development according to claim 3, characterized in that: The construction of training set data includes the following steps: Perform image preprocessing on satellite images, set the network input image channel size to 512×512, set the semantic segmentation category to 2, and the semantic segmentation category includes rock piles and ordinary mountains; Inject category labels into the processed satellite images used for training, obtain the original image files and annotated images, and record the TXT files of the training and test file names.
5. The method for predicting rock mass growth and development according to claim 4, characterized in that: The training process of the satellite image rock pile recognition model based on deep learning includes the following steps: By customizing the number of cyclic training, the number of training packages is 2, changing the training parameters and the number of training times, and obtaining the model corresponding to the optimal network parameters after the training, the trained satellite image rock pile recognition model is obtained, wherein the network parameters include pixel accuracy, average pixel accuracy, average intersection-over-combination ratio and frequency-weighted intersection-over-combination ratio; The trained deep learning-based satellite image rock pile recognition model was used for testing to obtain the rock pile contour height information in the collected rock pile images.
6. A rock mass growth prediction system, characterized in that: include: The rock pile image information acquisition module is used to collect rock pile image data in the study area, extract the rock pile edge contour height through the recognition model for rock pile image recognition, and obtain the rock pile image recognition result in the study area; The rock strength correction module is used to add the strength ratio parameter to the rock strength degradation calculation equation according to the rock pile image recognition result, correct the plateau rock strength value, and obtain the improved rock strength degradation curve relationship; by analyzing the relationship between the frost weathering intensity and altitude, a long-term temperature inversion sequence is introduced into the frost weathering action curve relationship to obtain the improved frost weathering action curve relationship; based on the improved rock strength degradation curve relationship and the improved frost weathering action curve relationship, the frost weathering intensity, rock degradation intensity and time effect factor are calculated; The change trend prediction module is used to expand the time series based on the frost weathering intensity, rock degradation intensity and time effect factor through key algorithms, analyze the change trend of frost weathering intensity-altitude over time, and make long-term predictions on the change trend of the rock mass in the study area to obtain prediction results; The improved rock strength degradation curve relationship comprises the following steps: The strength ratio parameter θ is added to the original rock strength degradation calculation equation to correct the plateau rock strength value. The calculation formula is: Where h is the rock depth; h0 is the depth corresponding to the center of the curve; b is the shape function coefficient; S0(t) is the shear strength parameter of rock and soil at time t; S H is the lower limit strength value of rock weathering; θ is the rock strength ratio parameter; The improved frost weathering curve relationship equation comprises the following steps: By analyzing the relationship between the intensity of frost weathering and altitude, it is found that with the increase of temperature, the total frost weathering intensity at the same altitude also shows an upward trend; under the same total frost weathering intensity, the altitude range also expands; A long-term temperature inversion sequence is introduced into the rock strength degradation curve relationship. Through this temperature inversion sequence, the freezing weathering of rocks in the past few years is calculated to obtain the average weathering rate. The relationship between the temperature and depth of the frost weathering basic curve is: Where t is the time in days, T(z,t) is the temperature at different times and depths; T mα is the annual average temperature; T α is half of the annual temperature variation; α is the thermal diffusion coefficient; P y Refers to the cycle of temperature changes; Improved frost weathering curve relationship: In the formula, F ci is the frost weathering intensity, which refers to the integral of the temperature gradient over time in the frost weathering window; D is the distance from 0℃ to -5℃; D * is the length index; The calculation of the frost weathering intensity, rock degradation intensity and time effect factor comprises the following steps: According to the relationship between the temperature and depth of frost weathering and the improved relationship between the frost weathering curve, by calculating the temperature lapse rate and the altitude where the annual average temperature is 0℃, it is found that the formation of condensed ice is directly related to the intensity of frost weathering. The improved frost weathering curve relationship is used to simulate the frost weathering in a year at a fixed T mα Under the above conditions, the temperature corresponding to different depths every day is used to determine whether the condensation ice can grow; The method of obtaining the prediction result comprises the following steps: By calculating the intensity of frost weathering and rock degradation in the study area, the specific reasons and related data affecting the rock fragmentation and formation of adverse geological phenomena in the study area are obtained, and the dimensionality reduction evolution rate of the rock mass in the study area, that is, the time effect factor, is obtained; Based on the calculated frost weathering intensity, rock degradation intensity and time effect factor, key algorithms are used to expand the time series and make long-term predictions on the changing trends of the rock mass in the study area.
