WO2020155929A1

WO2020155929A1 - Method for determining rock mass integrity

Info

Publication number: WO2020155929A1
Application number: PCT/CN2019/127420
Authority: WO
Inventors: 刘飞香; 郑大桥; 廖金军; 杜义康; 易达云; 肖正航; 蒋海华; 杜洋; 伍容
Original assignee: 中国铁建重工集团股份有限公司
Priority date: 2019-02-01
Filing date: 2019-12-23
Publication date: 2020-08-06
Also published as: CN109872311A

Abstract

A method for determining rock mass integrity, comprising: step 1, obtaining an image of a tunnel face to be analyzed; step 2: extracting features from the image of said tunnel face to obtain a set of feature parameters of the image of said tunnel face; and step 3: determining the rock mass integrity of a rock mass corresponding to said tunnel face on the basis of the set of feature parameters by using a preset neural network model. In the method, a convolutional neural network is applied to the automatic determination of rock mass integrity, and the final determination result can be obtained only by inputting an image to be tested into a neural network model, without manual participation during the entire identification process, so that excessive dependence on human experience is avoided, and the accuracy of determination of rock mass integrity is improved.

Description

A method for judging the integrity of rock mass

Technical field

The invention relates to the technical field of engineering geological prospecting, in particular to a method for judging the integrity of a rock mass.

Background technique

Surrounding rock is the surrounding rock mass whose stress state changes due to the influence of excavation in underground rock engineering. Rock mass refers to a complex geological body containing geological structural planes such as joints, cracks, bedding, and faults under certain geological conditions. The basic quality of the rock mass is the most basic attribute inherent in the rock mass that affects the stability of the engineering rock mass. The quality of the basic quality of the rock mass depends on the internal factors that constitute the structural characteristics of the rock mass, and the integrity of the rock mass plays a controlling role One of the factors.

The integrity of rock mass refers to the degree of development of various geological interfaces dominated by fissures in the rock mass. It is a comprehensive reflection of the structure of the rock mass. It depends on factors such as the degree of structural plane cutting, the size of the structure, and the bonding state between blocks. General indicators used in rock mass engineering.

Summary of the invention

The present invention provides a method for judging the integrity of a rock mass, the method comprising:

Step 1: Obtain an image of the tunnel face to be analyzed;

Step 2: Perform feature extraction on the tunnel face image to be analyzed to obtain the feature parameter set of the tunnel face image to be analyzed;

Step 3: Based on the characteristic parameter set, use a preset neural network model to determine the rock mass integrity degree of the rock mass corresponding to the face of the tunnel to be analyzed.

According to an embodiment of the present invention, in the second step, the image quality of the tunnel face image to be analyzed is also judged to determine whether the tunnel face image to be analyzed meets a preset condition, wherein: If the preset condition is met, the operation of extracting structural features of the tunnel face image to be analyzed is performed.

According to an embodiment of the present invention, the preset condition includes at least one of the listed items:

The image pixel value is greater than the preset pixel value, the sensitivity is greater than the preset sensitivity threshold, and the image definition is greater than the preset definition threshold.

According to an embodiment of the present invention, the step two includes:

Preprocessing the tunnel face image to be analyzed to obtain a preprocessed image;

Perform feature extraction according to the structural plane of the preprocessed image to obtain the feature parameter set.

According to an embodiment of the present invention, in the step three,

Based on the characteristic parameter set, use a preset neural network model to determine the probability of the integrity of the rock mass corresponding to the tunnel face to be analyzed in various types of rock mass;

The probability of the integrity of the rock mass in various rock masses is compared with a preset decision threshold, and the rock mass integrity of the rock mass corresponding to the face of the tunnel to be analyzed is determined according to the comparison result.

According to an embodiment of the present invention, the preset decision threshold is determined according to the following expression:

Among them, F represents the preset decision threshold, P represents the model accuracy, and R represents the model recall rate.

According to an embodiment of the present invention, in the step three, it is judged whether the probability of the rock mass in the i-th type of rock mass is greater than the preset decision threshold, and if it is greater, the tunnel to be analyzed is determined The rock mass integrity degree of the rock mass corresponding to the tunnel face belongs to the i-th type rock mass integrity degree.

According to an embodiment of the present invention, if there are at least two types of rock mass completeness whose probability of rock mass integrity is greater than the preset decision threshold, then the one with the largest probability is determined as the to-be-analyzed The rock mass integrity degree of the rock mass corresponding to the tunnel face.

According to an embodiment of the present invention, the step of constructing the preset neural network includes:

Step a. Obtain multiple tunnel face images with different degrees of rock mass integrity;

Step b: Extracting rock integrity identification data according to the multiple tunnel face images with different rock integrity degrees;

Step c: Use the rock integrity recognition data to train a preset neural network, and train to obtain the preset neural network model.

According to an embodiment of the present invention, in the step c, the performance evaluation is performed during the training process through the loss function, Roc curve, Auc value, and average inference time, and the training model is determined according to the performance evaluation result.

The rock integrity identification method provided by the present invention applies the convolutional neural network to the automatic identification of the integrity of the rock mass, and it only needs to input the image to be tested into the neural network model to obtain the final identification result. Since the entire identification process does not require human participation in the judgment, excessive dependence on human experience can be avoided, and the accuracy of identification of the integrity of the rock mass can be improved.

At the same time, in the implementation process of this method, only the collected face image is transmitted to the main control room of the rock drilling rig in real time, and the judgment result is obtained in real time through the neural network model, which improves the judgment of the integrity of the rock mass. effectiveness.

Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained through the structures specifically pointed out in the specification, claims and drawings.

Description of the drawings

In order to explain the embodiments of the present invention or the technical solutions in the prior art more clearly, the following will briefly introduce the drawings required in the description of the embodiments or the prior art:

Fig. 1 is a schematic diagram of the implementation process of a rock mass integrity identification method according to an embodiment of the present invention;

Figure 2 is a schematic diagram of the implementation process of constructing a preset neural network model according to an embodiment of the present invention;

Fig. 3 is a schematic diagram of an implementation process of using a preset neural network model to determine the integrity of the rock mass corresponding to the face image of the tunnel to be analyzed according to an embodiment of the present invention.

detailed description

Hereinafter, the implementation of the present invention will be described in detail with reference to the accompanying drawings and embodiments, so as to fully understand how the present invention applies technical means to solve technical problems and achieve the realization process of technical effects and implement them accordingly. It should be noted that, as long as there is no conflict, the various embodiments of the present invention and the various features in each embodiment can be combined with each other, and the technical solutions formed are all within the protection scope of the present invention.

Meanwhile, in the following description, many specific details are set forth for the purpose of explanation to provide a thorough understanding of the embodiments of the present invention. However, it is obvious to those skilled in the art that the present invention may be implemented without the specific details or the specific manner described herein.

In addition, the steps shown in the flowcharts of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical sequence is shown in the flowcharts, in some cases, they can be different Perform the steps shown or described in the order here.

At present, the methods for determining the integrity of rock masses mainly include qualitative evaluation and quantitative indicators. Among them, the qualitative evaluation standard is to measure the integrity of the rock mass from the degree of structural plane development (such as the degree of combination of main structural planes, main structural plane types, corresponding structural types, etc.) according to the geometric macroscopic form of the rock mass.

There are many indicators for quantitative evaluation of rock mass integrity at home and abroad. For example, elastic wave testing method (evaluation indicators based on this method include rock integrity index Kv, etc.), core drilling method (evaluation indicators based on this method include rock quality indicator RQD, unit core cracks, etc.), structural plane Statistical method (evaluation indicators based on this method include rock mass volume joint number Jv, average joint spacing dp, etc.). These methods of evaluating rock mass integrity have their own advantages and disadvantages, and most of the evaluation indicators only reflect the integrity of the rock mass from a certain side.

In the process of tunnel excavation, the exposed rock mass is often inconsistent with the rock mass obtained in the previous exploration, and it is difficult to obtain the integrity of the rock mass in real time. In the field construction, it is more dependent on the existing survey report and the field engineer's own experience to determine, the results obtained in this way are easily affected by subjective factors and experience in actual operation. Therefore, it is important to find a more convenient and accurate method to determine the integrity of the rock mass.

In response to this problem, the present invention provides a new method for judging the integrity of the rock mass. This method does not require manual participation in the process of judging the integrity of the rock mass, thus avoiding excessive reliance on human experience. Improve the accuracy and speed of the judgment of the integrity of the rock mass.

Figure 1 shows a schematic diagram of the implementation process of the rock mass integrity identification method provided by this embodiment.

As shown in FIG. 1, the rock integrity identification method provided by this embodiment first obtains an image of the tunnel face to be analyzed in step S101.

Specifically, when acquiring the image of the tunnel face to be analyzed, it is preferable to install a camera and lighting lamp with a pan-tilt function under the hanging basket of the construction machinery (such as a rock drilling rig), and use the headlight above the cab of the rig Brighten the face of the face with the headlight, and automatically take a picture of the face of the face to obtain a clear, uniform light and dark image of the face of the face without obvious obstructions (such as human shadows, mechanical equipment, etc.).

In this embodiment, in order to ensure the accuracy and reliability of the integrity of the rock mass finally obtained, the method preferably performs image quality identification on the face image of the tunnel to be analyzed in step S101 to determine the tunnel palm to be analyzed. Whether the sub-surface image meets the preset conditions. Among them, if the tunnel face image to be analyzed meets the preset conditions, the method will perform the operation of extracting the structural features of the tunnel face image to be analyzed; and if the tunnel face image to be analyzed does not meet the preset conditions, the method The method will re-acquire the image of the tunnel face to be analyzed.

Specifically, in this embodiment, the method determines in step S101 whether the image pixel value of the image of the tunnel face to be analyzed is greater than the preset pixel value, and determines whether the sensitivity of the image is less than the preset sensitivity threshold, and also It will judge whether the image sharpness is greater than the preset sharpness threshold.

For example, if the image pixel value of the tunnel face image to be analyzed is greater than 1600W pixels, and the image sensitivity is less than 200, and the corresponding image processing algorithm determines that the image definition meets the requirements (that is, greater than the preset definition threshold), then This method can also determine that the tunnel face image to be analyzed meets the preset conditions.

It should be pointed out that in different embodiments of the present invention, the specific values of the preset pixel value, preset sensitivity threshold, and preset definition threshold can be configured to different reasonable values according to actual needs. The present invention is not correct. The specific values of the above parameters are limited.

Of course, in other embodiments of the present invention, according to actual conditions, the method may also only use one or several of the items listed above to determine whether the tunnel face image to be analyzed meets the preset conditions. The invention is not limited to this.

As shown in FIG. 1, after the tunnel face image to be analyzed is obtained, the method preferably extracts structural features of the tunnel face image to be analyzed in step S102 to obtain a feature parameter set.

Specifically, in this embodiment, the method preferably first preprocesses the tunnel face image to be analyzed in step S102 to obtain the preprocessed image.

For example, in this embodiment, when preprocessing the tunnel face image to be analyzed, the method preferably first performs histogram equalization on the tunnel face image to be analyzed to obtain the first image. Subsequently, the method performs high-pass filtering on the first image to obtain the preprocessed image.

The histogram equalization can use the image histogram to adjust the contrast of the image. In the process of histogram equalization, two conditions must be guaranteed: (1) No matter how the pixels are mapped, the original size relationship must remain unchanged, but the contrast will increase; (2) If it is an 8-bit image, the pixel mapping function The value range of should be between 0 and 255, and should not exceed the range.

Specifically, in the histogram equalization process, the mapping method adopted in this embodiment is preferably:

Among them, _Sk represents the cumulative probability of pixels of different gray levels, n represents the sum of pixels in the image, n _j represents the number of pixels with gray level j, and L represents the total number of possible gray levels in the image.

High-pass filtering processing on the tunnel face image can enhance the fuzzy edges of the tunnel face image.

Through analysis, it is found that low-pass filtering can smooth out high-frequency grayscale changes, thereby removing high-frequency random noise, but it often produces blurring effects. Therefore, in this embodiment, in the face crack detection, in order to enhance the contrast, thereby enhancing the edge of the image, the method preferably adopts a high-pass filtering method.

The ideal color low-pass filter template is:

Among them, H(u,v) represents, D ₀ represents the radius of the passband, and D(u,v) represents the distance to the center of the spectrum.

Among them, D(u,v) can be calculated by the following expression:

Among them, M and N represent the size of the spectrum image, and (M/2, N/2) is the center of the spectrum. The ideal high-pass filter is the opposite of the low-pass filter, which is 1 minus the low-pass filter template.

Of course, in other embodiments of the present invention, the method may also adopt other reasonable methods to preprocess the tunnel face image to be analyzed, and the present invention is not limited to this.

In this embodiment, after the preprocessing process is completed, the method performs feature extraction on the obtained preprocessed image, thereby obtaining the required feature parameter set. Specifically, in this embodiment, the method preferably performs grayscale processing on the preprocessed image, thereby converting a three-channel RGB image into a single-channel grayscale image, and by extracting the grayscale value of each pixel The characteristic parameter set of the tunnel face image to be analyzed can be obtained.

Of course, in other embodiments of the present invention, the method may also adopt other reasonable methods to extract the feature data set of the tunnel face image to be analyzed, and the present invention is not limited to this.

As shown in Figure 1, in this embodiment, after obtaining the characteristic parameter set of the tunnel face image to be analyzed, the method uses a preset neural network model to determine the corresponding tunnel face image to be analyzed in step S103. The degree of rock mass integrity of the rock mass.

In this embodiment, the preset neural network model used in step S103 is constructed in advance. Among them, FIG. 2 shows a schematic diagram of the implementation process of constructing a preset neural network model in this embodiment.

As shown in FIG. 2, in this embodiment, when the method constructs the preset neural network model, it is preferable to first obtain multiple face images of different rock mass integrity levels in step S201, and then perform step S202. In step S201, the rock integrity identification data is extracted according to the multiple face images of different rock mass integrity degrees obtained in step S201.

Specifically, in this embodiment, the rock mass integrity degree corresponding to the tunnel face image acquired in step S201 of the method is preferably k, and the pixel value of the tunnel face image is preferably m*n . For example, the value of k can be 5, that is, the rock mass integrity degree of the rock mass corresponding to the tunnel face image is divided into 5 categories: a, b, c, d, and e.

In step S202, by performing quality analysis on the acquired tunnel face images, unqualified images can be eliminated, and it is also necessary to ensure that the data of the k types of tunnel face images are in a preset number ( For example, 1000 sheets) or more. Among them, this method can expand the amount of data by rotating and enlarging the image. The method also extracts the rock mass integrity identification data from the acquired images of the tunnel face with qualified quality in step S202.

In this embodiment, the implementation principle and implementation process of the above step S202 are similar to the related content of the above step S102, so the specific content of step S202 will not be repeated here.

For example, the rock mass integrity identification data obtained in step S202 of the method includes a two-dimensional matrix x as a preset neural network, and the pixels of the face image are m*n, so x∈m*n. At the same time, the above-mentioned rock mass integrity identification data also includes the types of rock mass integrity degrees, which are preferably a, b, c, d, and e. That is, the rock mass integrity identification data can be divided into a, b, c, There are 5 types of d and e, and each type of data set contains all the data. In addition, this method can also divide the five categories of a, b, c, d, and e into positive and negative categories. The specific division method is shown in Table 1.

Table 1

种类划分Classification	正类岩体完整性程度(数据量)Integrity degree of normal rock mass (data volume)	非正类岩体完整性程度(数据量)Integrity degree of non-normal rock mass (data volume)
a类Type a	1级(1000张)Level 1 (1000 sheets)	2、3、4、5级(共4000张)Level 2, 3, 4, 5 (4000 sheets in total)
b类Type b	2级(1000张)Level 2 (1000 sheets)	1、3、4、5级(共4000张)Level 1, 3, 4, 5 (4000 sheets in total)
c类class c	3级(1000张)Level 3 (1000 sheets)	1、2、4、5级(共4000张)Level 1, 2, 4, 5 (4000 sheets in total)
d类d category	4级(1000张)Level 4 (1000 sheets)	1、2、3、5级(共4000张)Level 1, 2, 3, 5 (4000 sheets in total)
e类class e	5级(1000张)Level 5 (1000 sheets)	1、2、3、4级(共4000张)Level 1, 2, 3, 4 (4000 sheets in total)

In addition, in order to evaluate the generalization ability of the model, in this embodiment, the method preferably divides the above-mentioned data set into a training data set and a verification data set at a ratio of 9:1;

In this embodiment, after the rock mass integrity identification data is obtained, the method uses the rock mass integrity identification data to train the preset neural network in step S203, so as to train the required preset neural network model.

Specifically, in this embodiment, the method preferably defines the structure of the neural network model

The main work of neural network structure design is to design the structure of the neuron nodes of each layer and the connection between the layers. There are three main structures in the convolutional neural network, namely the convolutional layer, the pooling layer and the full Connection layer.

The fully connected layer receives vector data as input and outputs the vector. The fully connected layer is used as the last layer of the network to combine the features extracted by the convolutional layer and output the result. The convolution layer receives a multi-channel image as input, and uses a certain size of convolution kernel to perform a convolution operation on the input data. The result of the convolution operation is calculated by the activation function as the output. The pooling layer is a down-sampling layer, which has the function of preserving input characteristics and reducing the input size. At the same time, due to its sampling characteristics, the pooling layer has a certain degree of rotation invariance to the input data.

In the convolutional layer and the fully connected layer, the activation function is used to perform secondary processing on the result of the linear calculation after the linear calculation, so that the neural network has the ability to fit the nonlinear function, and the activation function used by the convolutional layer is the ReLU function , The activation function of the fully connected layer is the Sigmoid function, and its expression is as follows:

ReLU:y(x)=max(0,x) (4)

In this embodiment, the method preferably uses the loss function

To measure the gap between the model output and the expected output. Specifically, the loss function

Preferably, it can be calculated using the following expression:

Among them, y represents the expected output of the model,

Represents the actual output of the model.

In the training process, the method preferably passes the loss function

Roc curve, Auc value and average inference time-consuming performance evaluation, and determine the training model according to the performance evaluation results. Specifically, in this embodiment, the method selects a model with the best overall performance from the performance evaluation result as the training model, and then debugs and selects appropriate training hyperparameters.

For example, this method can set a variety of models with different depths, and then perform performance evaluation through loss function, Roc curve, Auc value, and average inference time. In this embodiment, the method preferably selects the Auc value as the decision threshold, and other parameters as the qualified threshold. That is, the method preferably selects the model with the highest Auc value as the training model on the basis of meeting the qualified threshold.

In this embodiment, after a suitable model is determined, the method will debug the training hyperparameters that are unexpected in the model structure. Specifically, the aforementioned training hyperparameters preferably include: step size, batch size, and number of iterations.

Among them, if the step size is too large, the model will fail to converge. If the step size is too small, the iteration speed will be too slow. Therefore, it is necessary to choose a suitable step size to ensure that the model has a larger step size when the model converges.

If the batch size is too small, the model will have a large difference between the calculated gradient and the real gradient of the function space during the gradient calculation process, which will make the model's iterative oscillation too large or even unable to converge. If the batch size is too large, it will consume too much computing resources for one iteration calculation, which will make the iteration speed too slow or even impossible to train. Therefore, a suitable batch size needs to be selected. In this embodiment, the method preferably configures the batch size to be 10-60 during the actual training process.

Too few iterations will make the model iteratively stop prematurely, at this time the model underfits; while too many iterations will make the model overfit and affect the generalization ability of the model. Therefore, it is also necessary to select an appropriate number of iterations, or to judge the generalization ability of the model in real time during the training process, and to stop training before the model is overfitted.

In this embodiment, the method preferably uses the training set data in the rock integrity identification data to iteratively train the parameters w in the model, so that the loss value L of the training set of the model reaches the global or local minimum.

Specifically, in this embodiment, the method adopted in the iteration of the method is preferably the gradient descent method, and the parameter update process can be shown in the figure expression (7):

Among them, w _i and w _i-1 respectively represent the value of the parameter w after the i-th iteration and after the i-1th iteration, and α represents the step length.

In the gradient descent method, after a suitable step size and enough iterative training, the model loss will converge to a global optimal solution or a local optimal solution.

In this embodiment, the method preferably measures the performance of the trained model on the verification set to determine whether the model meets the task requirements. Among them, if the task requirements cannot be met, the method repeats the above-mentioned model determination process to the parameter w iteration process until the model structure and parameters that meet the task requirements are found.

Of course, in other embodiments of the present invention, the method can also adopt other reasonable ways to construct the foregoing predetermined neural network model according to actual needs, and the present invention is not limited to this.

In this embodiment, as shown in FIG. 3, when the preset neural network model is used to determine the rock mass integrity degree of the rock mass corresponding to the image of the tunnel face to be analyzed, the method is preferably based on step S301. The characteristic parameter set obtained in step S102 is used to determine the probability of the integrity of various rock masses corresponding to the face of the tunnel to be analyzed by using a preset neural network model. Subsequently, in step S302, the method compares the probability of the integrity of the rock mass in various types of rock mass with the preset decision threshold, and in step S303, the rock mass corresponding to the face of the tunnel to be analyzed is determined according to the comparison result The degree of integrity of the rock mass.

In this embodiment, using a preset neural network model, the method can convert a multi-classification problem of rock mass integrity recognition into k two classification problems. The output result y of the aforementioned preset neural network model is a matrix. The matrix has a total of k characteristic parameters. Among them, y _i represents the probability value of i-level rock mass integrity, and the value of i is a natural number from 1 to k. For example, if the rock mass integrity degree of the rock mass corresponding to the tunnel face image is divided into 5 categories: a, b, c, d, and e, the value of i above is a natural number from 1 to 5.

In this embodiment, the method uses a preset decision threshold in step S302 and step S303 to convert the probability value output by the preset neural network model into a decision binary value. In this embodiment, the selection criterion of the decision threshold F is to maximize the harmonic average value of precision and recall, that is, there is:

In this embodiment, the method preferably judges whether the probability y _{i of the} integrity of the rock mass in the i-th type of rock mass is greater than the preset decision threshold F. Among them, if it is greater than, it is determined that the rock mass integrity degree of the rock mass corresponding to the tunnel face to be analyzed belongs to the i-th rock mass integrity degree.

For example, taking a type data set as an example, when the probability y _{1 is} greater than the preset decision threshold F, the output result of this method will be 1, which means the rock mass corresponding to the tunnel face to be analyzed The degree of integrity belongs to the degree of integrity of type a rock mass (that is, the degree of integrity of rock mass of level 1). When y _{1 is} less than or equal to the preset decision threshold F, the output result of this method will be 0, which means that the rock mass integrity degree of the rock mass corresponding to the tunnel face to be analyzed belongs to non-level 1 rock The degree of body integrity.

At the same time, for b, c, d, and e data sets, this method can use the same method to determine the integrity of the rock mass.

Generally speaking, the value of one and only one element in the output matrix y obtained by this method will be greater than the preset decision threshold F, and the value of the element of the sum will be less than or equal to the preset decision threshold F. It should be pointed out that in this embodiment, if the probability of rock mass integrity is greater than the preset decision threshold F, there are at least two types of rock mass completeness (that is, there are at least two elements in the output matrix y. Is greater than the preset decision threshold F), then this method will determine the maximum probability value as the rock mass integrity degree of the rock mass corresponding to the tunnel face to be analyzed.

For example, in the output matrix y, the values of y ₂ and y ₃ are both greater than the preset decision-making threshold F and y ₂ >y _3. At this time, the method will determine the rock mass of the rock mass corresponding to the face of the tunnel to be analyzed The degree of completeness is determined as the degree of integrity of type b rock mass (ie, the degree of integrity of the second level rock mass).

It can be seen from the above description that the rock integrity identification method provided by the present invention applies the convolutional neural network to the automatic identification of the integrity of the rock mass. It only needs to input the image to be tested into the neural network model. The final judgment result can be obtained. Since the entire identification process does not require human participation in the judgment, excessive dependence on human experience can be avoided, and the accuracy of identification of the integrity of the rock mass can be improved.

It should be understood that the embodiments disclosed in the present invention are not limited to the specific structures or processing steps disclosed herein, but should extend to equivalent substitutions of these features understood by those of ordinary skill in the relevant art. It should also be understood that the terms used herein are only used for the purpose of describing specific embodiments and are not meant to be limiting.

The "one embodiment" or "an embodiment" mentioned in the specification means that a specific feature, structure, or characteristic described in conjunction with the embodiment is included in at least one embodiment of the present invention. Therefore, the phrases "one embodiment" or "an embodiment" appearing in various places throughout the specification do not necessarily all refer to the same embodiment.

Although the above examples are used to illustrate the principles of the present invention in one or more applications, for those skilled in the art, without departing from the principles and ideas of the present invention, it is obvious that the form, usage and implementation can be Make various changes to the details without creative work. Therefore, the present invention is defined by the appended claims.

Claims

A method for judging the integrity of a rock mass, characterized in that the method comprises:

Step 1: Obtain an image of the tunnel face to be analyzed;

Step 2: Perform feature extraction on the tunnel face image to be analyzed to obtain the feature parameter set of the tunnel face image to be analyzed;

Step 3: Based on the characteristic parameter set, use a preset neural network model to determine the rock mass integrity degree of the rock mass corresponding to the face of the tunnel to be analyzed.
The method according to claim 1, wherein in the second step, the image quality of the tunnel face image to be analyzed is also judged to determine whether the tunnel face image to be analyzed meets the expected A condition is set, wherein if the preset condition is satisfied, the operation of extracting structural features of the tunnel face image to be analyzed is performed.
The method according to claim 2, wherein the preset condition includes at least one of the listed items:

The image pixel value is greater than the preset pixel value, the sensitivity is greater than the preset sensitivity threshold, and the image definition is greater than the preset definition threshold.
The method according to any one of claims 1 to 3, wherein the step two comprises:

Preprocessing the tunnel face image to be analyzed to obtain a preprocessed image;

Perform feature extraction according to the structural plane of the preprocessed image to obtain the feature parameter set.
The method according to any one of claims 1 to 4, wherein in the step three,

Based on the characteristic parameter set, use a preset neural network model to determine the probability of the integrity of the rock mass corresponding to the tunnel face to be analyzed in various types of rock mass;

The probability of the integrity of the rock mass in various rock masses is compared with a preset decision threshold, and the rock mass integrity of the rock mass corresponding to the face of the tunnel to be analyzed is determined according to the comparison result.
The method according to claim 5, wherein the preset decision threshold is determined according to the following expression:

Among them, F represents the preset decision threshold, P represents the model accuracy, and R represents the model recall rate.
The method according to claim 5 or 6, characterized in that, in the step 3, it is judged whether the probability of the integrity of the rock mass in the i-th type of rock mass is greater than the preset decision threshold, wherein, if greater than, It is determined that the rock mass integrity degree of the rock mass corresponding to the tunnel face to be analyzed belongs to the i-th type rock mass integrity degree.
The method according to claim 7, characterized in that if there are at least two types of rock mass completeness whose probability of the rock mass integrity degree is greater than the preset decision threshold value, then the probability value is determined to be the largest Is the rock mass integrity degree of the rock mass corresponding to the face of the tunnel to be analyzed.
The method according to any one of claims 1 to 8, wherein the step of constructing the preset neural network comprises:

Step a. Obtain multiple tunnel face images with different degrees of rock mass integrity;

Step b: Extracting rock integrity identification data according to the multiple tunnel face images with different rock integrity degrees;

Step c: Use the rock integrity recognition data to train a preset neural network, and train to obtain the preset neural network model.
The method according to claim 9, wherein, in the step c, the performance evaluation is performed during the training process through the loss function, the Roc curve, the Auc value, and the average inference time, and the training model is determined according to the performance evaluation result .