CN116682010A - Surrounding rock classification real-time prediction method based on TBM rock slag image - Google Patents
Surrounding rock classification real-time prediction method based on TBM rock slag image Download PDFInfo
- Publication number
- CN116682010A CN116682010A CN202310643903.0A CN202310643903A CN116682010A CN 116682010 A CN116682010 A CN 116682010A CN 202310643903 A CN202310643903 A CN 202310643903A CN 116682010 A CN116682010 A CN 116682010A
- Authority
- CN
- China
- Prior art keywords
- rock
- surrounding rock
- classification
- image
- surrounding
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 239000011435 rock Substances 0.000 title claims abstract description 240
- 239000002893 slag Substances 0.000 title claims abstract description 59
- 238000000034 method Methods 0.000 title claims abstract description 44
- 230000006870 function Effects 0.000 claims description 25
- 238000012549 training Methods 0.000 claims description 21
- 238000013527 convolutional neural network Methods 0.000 claims description 12
- 238000012545 processing Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 7
- 238000011176 pooling Methods 0.000 claims description 7
- 238000004590 computer program Methods 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 3
- 230000013011 mating Effects 0.000 claims 1
- 239000013589 supplement Substances 0.000 claims 1
- 230000005641 tunneling Effects 0.000 abstract description 11
- 238000010276 construction Methods 0.000 description 8
- 239000000243 solution Substances 0.000 description 8
- 238000010586 diagram Methods 0.000 description 7
- 239000011159 matrix material Substances 0.000 description 6
- 206010057175 Mass conditions Diseases 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 3
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 2
- 238000005520 cutting process Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001502 supplementing effect Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000009412 basement excavation Methods 0.000 description 1
- 238000005422 blasting Methods 0.000 description 1
- 238000004422 calculation algorithm Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005553 drilling Methods 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 239000010438 granite Substances 0.000 description 1
- 239000003673 groundwater Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000002347 injection Methods 0.000 description 1
- 239000007924 injection Substances 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/176—Urban or other man-made structures
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Analysis (AREA)
Abstract
The scheme relates to a surrounding rock classification real-time prediction method based on TBM rock residue images, relates to the technical field of TBM surrounding rock classification prediction and automatic discrimination in tunnel engineering, and is used for judging surrounding rock classification of a face rock mass in advance and providing suggestions for subsequent support and tunneling parameter adjustment in real time; meanwhile, the man-made subjectivity of surrounding rock classification judgment is reduced, the working efficiency of geological engineers is improved, and technical support is provided for automatic surrounding rock classification judgment. The method comprises the following steps: in the current circulation section, a TBM rock slag image is acquired at fixed time, and a trained surrounding rock classification prediction model is utilized to judge a surrounding rock class classification result; in the process of the circulation section, according to the judgment result of a plurality of rock residue images obtained by the current circulation section, the probability Pi of the current circulation section belonging to various surrounding rocks is calculated in real time, and the surrounding rock classification corresponding to the highest probability is selected as a surrounding rock classification recognition output result.
Description
Technical Field
The scheme relates to the technical field of tunnel engineering TBM surrounding rock classification prediction and automatic discrimination, in particular to a surrounding rock classification real-time prediction method based on TBM rock slag images.
Background
Compared with the traditional drilling and blasting method, the TBM tunneling machine method has the advantages of high construction speed, high mechanical informatization degree, small ecological disturbance and the like, and is widely applied to the construction of long tunnels. In the TBM tunneling process, a cutter shield is closely attached to a tunnel face, engineering technicians cannot observe rock mass conditions in front of the tunnel face, and only rock mass exposed behind the shield (about 6m away from the tunnel face) can be used for estimating the rock mass in front, so that certain information lag exists.
The surrounding rock classification is used as a comprehensive evaluation index of rock mass conditions, the rock mass hardness degree, the rock mass integrity, the underground occurrence environment and other conditions can be evaluated, the TBM field surrounding rock classification result is determined by observing the field rock mass condition by a geological engineer, and certain subjectivity and uncertainty exist. In addition, the rock slag images under the same surrounding rock classification condition have higher similarity, are difficult to distinguish manually, and are challenging to identify by human eyes.
Disclosure of Invention
In order to solve the problems in the prior art, the purpose of the scheme is to provide a surrounding rock classification real-time prediction method based on TBM rock slag images, judge surrounding rock classification of the rock mass of the face in advance, and provide suggestions for subsequent support and tunneling parameter adjustment in real time; meanwhile, the working efficiency of geological engineers is improved, the man-made subjectivity of surrounding rock classification judgment is reduced, and technical support is provided for automatic surrounding rock classification judgment. In order to achieve the technical purpose, the technical scheme is as follows.
In a first aspect, the present disclosure provides a method for real-time prediction of surrounding rock classification based on TBM slag images, the method comprising the steps of:
in the current circulation section, a TBM rock slag image is acquired at fixed time, and a trained surrounding rock classification prediction model is utilized to judge a surrounding rock class classification result;
in the process of the circulation section, according to the judgment result of a plurality of rock residue images acquired by the current circulation section, calculating the probability P of the current circulation section belonging to various surrounding rocks in real time i Selecting the surrounding rock classification corresponding to the highest probability as a surrounding rock class classification recognition output result;
the surrounding rock classification prediction model is constructed by adopting a deep convolutional neural network, TBM rock slag images are taken as input, feature extraction is carried out on the TBM rock slag images, and adjacent rock slag information of a plurality of continuous rock slag images is utilized to output the surrounding rock classification grade of the current circulation section.
In one embodiment of the foregoing technical solution, the probability P that the current circulation segment belongs to the i-class surrounding rock i :
P i =N i /N
In the formula (1): p (P) i Representing the probability that the current picture belongs to i-type surrounding rock; n (N) i The number of images of which the model is judged to be the classification result of the i-type surrounding rock in the image of the circulation segment is represented; n represents the total number of images acquired for the loop at the current time.
In one embodiment of the foregoing technical solution, the surrounding rock classification prediction model includes a convolution layer, an activation function, a pooling layer, a full connection layer, and a softmax function; wherein: the convolution layer, the activation function and the pooling layer are used for extracting surrounding rock image features, and the full-connection layer and the softmax function are used for calculating the prediction category of the surrounding rock image;
the training of the surrounding rock classification prediction model comprises the following steps:
taking the preprocessed rock residue image training set as input of a surrounding rock classification prediction model, and outputting the prediction category of the surrounding rock image;
and calculating the error between the predicted surrounding rock category and the real category by adopting the cross entropy loss function, and self-adaptively adjusting the learning rate and the model parameter according to the value of the cross entropy loss function until the training stop condition is met.
In one embodiment of the above technical solution, the rock residue image acquisition device is located on a rear supporting belt conveyor behind the TBM shield.
In one embodiment of the above technical solution, the rock residue image acquisition device is composed of a camera, a light supplementing lamp and a fixing bracket.
In one embodiment of the foregoing technical solution, the rock residue image preprocessing includes: s1, if the rock residue image is used for training, adopting a data enhancement technology of horizontal and vertical overturn for the rock residue image, and executing S2; otherwise, directly executing S2; s2, processing the image into a uniform size.
In one embodiment of the foregoing solution, the deep convolutional neural network is a ResNet, VGG, squeezeNet model or other convolutional neural network.
In one embodiment of the foregoing technical solution, the removing data includes: images of fault shooting, images of belt idling, images of rock residue coverage below 30% of shooting area and images of external interference.
In a second aspect, the present disclosure proposes a real-time prediction system for surrounding rock classification based on TBM slag images, including a memory and a processor, where the memory stores a computer program that can be loaded by the processor and execute any one of the methods described above.
In a third aspect, a computer readable storage medium storing a computer program capable of being loaded by a processor and performing any one of the methods described above is provided.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings that are needed in the description of the embodiments will be briefly described below, it being obvious that the drawings in the following description are only some embodiments of the present application, and that other drawings may be obtained according to these drawings without inventive effort to a person skilled in the art.
FIG. 1 is a schematic diagram of a shooting position of an image acquisition device in an embodiment;
FIG. 2, a schematic representation of the surrounding rock grade in one embodiment;
FIG. 3 is a schematic diagram of anomaly data in one embodiment;
FIG. 4 is a schematic diagram of model convergence of a surrounding rock classification prediction model constructed by ResNet18 in one embodiment;
FIG. 5 is a schematic view of model convergence of a surrounding rock classification prediction model constructed by VGG19bn in a specific embodiment;
FIG. 6 is a schematic view of model convergence of a surrounding rock classification prediction model constructed by adopting SquezeNet v1.0 in a specific embodiment;
FIG. 7 is a schematic diagram of a classification task confusion matrix in one embodiment;
FIG. 8, a schematic diagram of model accuracy in one embodiment;
fig. 9 is a schematic diagram of a surrounding rock classification prediction flow in an embodiment.
Detailed Description
Term interpretation:
classification of surrounding rock: the tunnel engineering is based on the concept of engineering analogy, the surrounding rock is classified by indexes such as hardness degree, rock integrity and the like of the surrounding rock to form a surrounding rock classification system, the surrounding rock is divided into I-V grades according to the railway tunnel design specification (TB 10001-2016/J449-2016), wherein the I-grade surrounding rock is the most stable, the rock mass gradually deteriorates along with the increase of the grades, and engineering technology and designers determine corresponding construction methods and structural designs according to the surrounding rock classification. The surrounding rock classification provides important reference values for tunnel construction, payment schemes, engineering cost and the like.
TBM slag: in the tunneling process of the TBM (Tunnel Boring Machine ), rock slag formed by cutting rock mass of the face by the cutterhead is conveyed out of the tunnel along with a transmission system through a belt conveyor.
The tunnel boring machine has the advantages of high boring speed, small disturbance to surrounding rock, good tunnel forming, small influence on ecological environment and the like, and is widely applied to large tunnel project construction including traffic, water conservancy and mines. In the TBM tunneling process, a cutter shield is tightly attached to a tunnel face, engineering technicians cannot observe rock mass conditions in front of the tunnel face, and can observe and judge the rock mass in front only through the rock mass exposed behind the shield (approximately 5m away from the tunnel face), so that certain subjectivity and uncertainty exist, and certain information lag exists compared with the rock mass information of the tunnel face of the cutter which is breaking the rock. In recent years, the TBM realizes construction data acquisition and storage in the tunneling process by carrying a plurality of sensors, the TBM construction data provides possibility for application prediction of a machine learning method, and in many scholars research, the TBM high-frequency mechanical data is used as input to predict surrounding rock classification, and research is mainly developed for the problems of characteristic parameter selection and data imbalance.
The scheme provides a new surrounding rock classification prediction method, a surrounding rock classification prediction model constructed by a deep convolutional neural network is utilized to fully mine rock slag information of a current rock slag image and adjacent rock slag information of a TBM rock slag image, the surrounding rock classification prediction effect is improved, the rock mass information of a face is perceived in real time, suggestions are provided for supporting measures and tunneling parameter adjustment behind a shield in real time in advance, the working efficiency of a geological engineer is improved, the artificial subjectivity of surrounding rock classification judgment is reduced, and references are provided for automatic surrounding rock classification judgment. The method comprises the following steps:
in the current circulation section, a TBM rock slag image is acquired at fixed time, and a trained surrounding rock classification prediction model is utilized to judge a surrounding rock class classification result;
ending the current circulation segment, and calculating probability P of the current circulation segment belonging to various surrounding rocks according to the determination result of the plurality of Zhang Yanzha images of the circulation segment i Selecting surrounding rock classification corresponding to the highest probability as a surrounding rock grade identification output result;
the surrounding rock classification prediction model is constructed by adopting a deep convolutional neural network, TBM rock slag images are taken as input, feature extraction is carried out on the TBM rock slag images, and adjacent rock slag information of a plurality of continuous rock slag images is utilized to output the surrounding rock classification grade of the current circulation section.
Taking a main cave slag image of a TBM tunnel of a Gaoli mountain as an example, the technical scheme for carrying out surrounding rock grading prediction by implementing the method is clearly and completely described with reference to the accompanying drawings.
Engineering overview
Tunnel positive tunnel TBM excavation diameter of Darui railway Gaogong mountain tunnel is 9m, maximum pushing distance of oil cylinder is 1.8m, and construction length of positive tunnel is 10.9km. The lithology of the TBM positive hole stratum is single and mainly takes granite as a main material. The surrounding rock classification along the line is mainly classified by III-class and IV-class surrounding rock, and the length of the surrounding rock classification along the line is shown in Table 1.
TABLE 1
(II) acquisition device
In the tunneling process of the TBM, rock slag formed by cutting rock mass of the face by the cutterhead is conveyed out of the tunnel along with a transmission system through a belt conveyor. According to the positions and functions of the belt conveyors, the belt conveyors are divided into (1) a main belt conveyor, (2) a rear matched belt conveyor, (3) a continuous belt conveyor and (4) a slag turning belt conveyor, the image acquisition equipment is arranged as far as possible in consideration of the real-time prediction analysis requirement, and because (1) the main belt conveyor has not been fully separated from the rock slag mud water and the arrangement space is limited, the image acquisition equipment is arranged at the position of the rear matched belt conveyor in comprehensive aspects, as shown in fig. 1.
The image acquisition equipment comprises a camera, a light supplementing lamp and a fixed support, and continuous photographing of the equipment under stable illumination is ensured when the belt runs at high speed.
Based on the image acquisition equipment, 2953 rock slag images under various surrounding rock classification conditions are acquired together, wherein the surrounding rock images comprise four different surrounding rock categories of IIIa level, IVb level, va level and Vb level, and the number of the original rock slag images of each level of surrounding rock is shown in table 2. An example of a slag image under different surrounding rock classifications is shown in fig. 2. ( And (3) injection: in order to further distinguish the quality of surrounding rocks, a certain type of rock mass is often divided into an a part and a b part according to standard scoring, wherein a represents a part with better quality in the rock mass, for example, va-level surrounding rocks are V-level surrounding rocks with better quality )
TABLE 2
The acquired original image is stored in an original image database.
(III) data processing
The original image has low correlation between the image information and surrounding rock classification caused by partial factor interference, which is unfavorable for the task of surrounding rock classification prediction, so that the original image needs to be subjected to data processing. Four types of data that need to be culled are listed in fig. 3.
(1) And (3) fault shooting: during the working process of the acquisition equipment, the screen is in an abnormal working state;
(2) belt idle: when the TBM is used for slag discharge, in order to ensure that the rock slag on the belt is completely removed, the belt is longer than the actual slag discharge time during operation, so that part of photos are only taken on the belt and no effective rock slag information exists;
(3) scattered blocks or groundwater cover: when TBM just starts tunneling, the belt runs in advance, and a large amount of water and scattered blocks are accompanied, so that effective rock residue information is very little, and images of shooting areas with the rock residue coverage range lower than 30% are removed uniformly;
(4) external interference: in the working process of the acquisition equipment, human interference irrelevant to rock slag information exists, and images containing external interference are removed in order to ensure the accuracy of the model.
After the above processing, an effective slag image dataset is obtained, and the number of effective slag images under each level of surrounding rock is shown in table 3.
TABLE 3 Table 3
And storing the obtained effective rock residue image as an original image into an effective image database, wherein data in the effective image database can be used as a training sample training model.
(IV) model creation, training and application
The scheme adopts a deep convolutional neural network model to conduct surrounding rock classification prediction. In the implementation mode, three deep convolutional neural networks of ResNet18, VGG19bn and SqueezeNet v1.0 are selected respectively to construct surrounding rock classification prediction models one by one, the surrounding rock classification prediction models are all performed on the same machine based on Pytorch1.12.1 under a python environment, a CPU is Intel (R) Core (TM) i9-119002.50GHz, a GPU is GeForce RTX 3090Ti, a RAM is 64GB, and the models are trained and applied under the GPU.
The method comprises the steps of constructing a surrounding rock classification prediction model, wherein the surrounding rock classification prediction model comprises a convolution layer, an activation function, a pooling layer, a full connection layer and a softmax function; wherein: the convolution layer, the activation function and the pooling layer are used for extracting surrounding rock image features, and the full connection layer and the softmax function are used for calculating the prediction category of the surrounding rock image. The surrounding rock classification prediction model takes TBM rock slag images as input, performs feature extraction on the TBM rock slag images, and outputs the surrounding rock classification grade of the current circulation section by utilizing adjacent rock slag information of a plurality of continuous rock slag images.
Parameter setting of a surrounding rock classification prediction model: the initial learning rate lr of the model was set to 0.0001, the batch size was 128, and the pre-trained model under the load-on-batch environment was set.
Forward propagation of surrounding rock classification prediction model: and taking the preprocessed training set picture as the input of the convolutional neural network, extracting surrounding rock image characteristics through operations such as a convolutional layer, an activation function, a pooling layer and the like, and calculating the prediction category of the surrounding rock image through the full-connection layer and the softmax function. Then calculating the error between the predicted surrounding rock class and the real class by adopting a cross entropy loss function, wherein the cross entropy loss function is as follows:
wherein: n represents the batch size, M represents the surrounding rock class number, y ic Numbers representing the true class of surrounding rock (the four types of surrounding rock correspond to 0, 1, 2 and 3 respectively), p ic The representative model belongs to category y for sample i ic Is used for the prediction probability of (1).
Back propagation and parameter updating of surrounding rock classification prediction model: setting the initial learning rate lr of the model to be 0.0001, adopting a Adam (Adaptive Moment Estimation) optimization algorithm, and adaptively adjusting the learning rate and model parameters according to the value of the cross entropy loss function to improve the classification performance of the model; iterative training: the forward propagation and parameter updating process is repeated until the maximum iteration number of 100 turns is reached.
In forward propagation, the rock residue image preprocessing includes: data processing and label calibration. The data processing comprises adopting a data enhancement technology of horizontal and vertical overturn to obtain more surrounding rock pictures, and the diversity of samples can be increased and the generalization capability of the model can be improved through the data processing. The picture size is then cut out uniformly to 224 x 224 to ensure that the model is able to handle the same size input image. And the label calibration is to perform grade calibration on each image according to the actual surrounding rock classification information. The preprocessed image is stored in the active image database. Randomly selecting 80% of pictures as training sets for parameter updating of the model, and the remaining 20% as test sets for evaluating the performance of the surrounding rock grading model.
As shown in fig. 4, 5 and 6, the convergence effects of the three models are shown in fig. 4, 5 and 6, the ResNet18 model and the VGG19 model bn, squeezeNet model are increased along with epoch, the Loss function is continuously reduced, the model is finally converged to a certain stable value, the Loss corresponds to the ACC effect, the accuracy ACC is continuously improved while the Loss is reduced, the model is normally converged, and the fitting problem is avoided. Wherein: definition of the accuracy ACC is that the sample ratio of all samples which are correctly judged, that is, the greater the number of TP and TN, the higher the ACC value:
where TP, FP, FN, TN is defined in the confusion matrix of fig. 8.
Fig. 7 is a diagram of a classification task confusion matrix, taking a classification task as an example, the classification task is divided into 4 regions according to the correspondence between the real labels and the prediction results: a sample set with a positive predicted value and a positive true value is defined as TP (True Positive); a sample set with a positive predicted value and a negative true value is defined as FP (False Positive); defining a sample set with a negative sample predicted value and a positive true value as FN (False Negative); a sample set with negative predicted and real values is defined as TN (True Negative). When the label is n-dimensional, the confusion matrix is an n multiplied by n matrix, and in the confusion matrix, the more diagonal area samples are, the better the model prediction effect is.
The results of the model accuracy ACC of the above three models are shown in fig. 8, where the model res net18 achieves the highest accuracy 0.986,SqueezeNet v1.0 model, which is not much different from res net 18. The VGG19bn accuracy is only 0.915, which is obviously lower than that of other two models, and the prediction performance is poor.
In order to test the real-time prediction speed of the model, the model corresponding to the highest accuracy of each model in training is stored, all the rock residue images are identified and predicted, and the average time consumption of each image in prediction is calculated as shown in table 4. The average time consumption of the three model predictions is relatively close to about 70ms, and the requirements of predicting images in real time in the TBM tunneling process can be met. Where VGG19bn is the longest time consuming and the SqueezeNet v1.0 model is the shortest time consuming, the possible reason that lightweight SqueezeNet v1.0 fails to exhibit significant advantages in speed is the computational effort required for the experimental environment GPU to be far higher than Yu Shanzhang images. It should be noted that in the three model training processes, VGG19bn resulted in the slowest training speed due to the large model size, resNet18 times, and the lightweight SquezeNet v1.0 training speed was the fastest. In contrast, the SquezeNet v1.0 has the characteristics of small model parameter scale and high training and operation speed.
TABLE 4 Table 4
According to the analysis, the ResNet18 model, the VGG19bn model and the SqueezeNet v1.0 model can successfully predict surrounding rock classification tasks, and compared with the ResNet18 model and the SqueezeNet v1.0 model, the ResNet18 model and the SqueezeNet v1.0 model have higher accuracy, the ACC is as high as 0.986, partial misjudgment exists in the prediction results in Va and Vb level ranges, and cross-level misjudgment exists in the VGG19bn prediction results, so that the influence on surrounding rock classification results is larger. In addition, in the aspects of training time length and average time consumption of prediction of each image, the SquezeNet v1.0 model has high training and calculating speed by virtue of the characteristics of a lightweight model, and achieves the effect of improving the calculating speed as much as possible on the premise of not greatly reducing the model precision to a certain extent. Surrounding rock classification prediction is recommended by using ResNet18 and SqueezeNet v1.0 models, and the SqueezeNet v1.0 model may have more obvious speed advantages under the condition of limited calculation power.
According to experimental results, although the rock slag images under the same surrounding rock grade have higher similarity, the rock slag images under different surrounding rock grades have certain discrimination degree, and the deep learning method is adopted to excavate TBM rock slag information to classify surrounding rocks, so that classification results are reliable, and the working efficiency of geological engineers is improved to a certain extent.
When the trained model is used for actual classification, the rock slag image preprocessing is only needed to cut the image size into the input image size which can be processed by the model. When the surrounding rock classification prediction is performed in real time, 1 rock slag image is captured every 1 minute by taking a circulating section as a unit, and each rock slag image is obtained in real time through a surrounding rock classification prediction modelThe surrounding rock classification result Set corresponding to the image, and the probability that the current circulation segment belongs to the i-type surrounding rock is defined as P i ,P i The calculation is shown in formula (3). Taking the classification probability P of surrounding rock i The highest classification is used as the surrounding rock classification result of the current circulation section, and the final surrounding rock classification result Class is shown in the formula (4).
P i =N i /N (3)
In the formula (1): i is the surrounding rock grade number; p (P) i Representing the probability that the current picture belongs to i-type surrounding rock; n (N) i The number of images of which the model is judged to be the classification result of the i-type surrounding rock in the image of the circulation segment is represented; n represents the total number of images acquired for the loop at the current time.
Class=Max(P i ,P II ,P III ,P IV ,P V ) (4)
In summary, as shown in fig. 9, the process of the surrounding rock classification prediction flow is that the TBM surrounding rock classification is predicted through a convolutional neural network, the rock mass condition of the face is judged in advance through rock slag, the surrounding rock classification information of the face is obtained, the image in the circulation section is identified by taking the circulation section as a unit, then the surrounding rock classification to which the circulation section belongs is comprehensively judged based on probability, and the accidental caused by the single image prediction result can be avoided.
And acquiring a rock slag image in real time in the process, acquiring a corresponding classification result of judging the surrounding rock grade in real time through a model, and covering the last result with the current result. In another embodiment, the surrounding rock grade classification result corresponding to each rock residue image can be obtained by simultaneously inputting the rock residue images acquired in one circulation section by taking the circulation section as a unit. The method can be further improved, probability calculation is designed into a model, and a final surrounding rock grade identification result is directly output through the model.
From the above description of the embodiments, it will be apparent to those skilled in the art that the method of the present disclosure may be implemented by software plus necessary general purpose hardware, and of course may be implemented by special purpose hardware including an application specific integrated circuit, a special purpose CPU, a special purpose memory, a special purpose component, etc. Generally, functions performed by computer programs can be easily implemented by corresponding hardware, and specific hardware structures for implementing the same functions can be varied, such as analog circuits, digital circuits, or dedicated circuits. However, in more cases for the present disclosure, a software program implementation is a better implementation.
Reference throughout this specification to "one embodiment," "another embodiment," "an embodiment," and so forth, means that a particular feature, structure, or characteristic described in connection with the embodiment is included in at least one embodiment of the present application as broadly described. The appearances of the phrase in various places in the specification are not necessarily all referring to the same embodiment. Further, when a particular feature, structure, or characteristic is described in connection with any embodiment, it is intended that such feature, structure, or characteristic be implemented within the scope of the application.
Claims (10)
1. The surrounding rock classification real-time prediction method based on TBM rock slag images is characterized by comprising the following steps of:
in the current circulation section, a TBM rock slag image is acquired at fixed time, and a trained surrounding rock classification prediction model is utilized to judge a surrounding rock class classification result;
in the process of the circulation section, calculating the probability of the current circulation section belonging to various surrounding rocks in real time according to the judgment result of a plurality of rock residue images acquired by the current circulation section, and selecting the surrounding rock classification corresponding to the highest probability as a surrounding rock class classification recognition output result;
the surrounding rock classification prediction model is constructed by adopting a deep convolutional neural network, TBM rock slag images are taken as input, feature extraction is carried out on the TBM rock slag images, and adjacent rock slag information of a plurality of continuous rock slag images is utilized to output the surrounding rock classification grade of the current circulation section.
2. The method according to claim 1, characterized in that the probability P that the current circulation segment belongs to class i surrounding rock i :
P i =N i /N
In the formula (1): p (P) i Representing the probability that the current picture belongs to i-type surrounding rock; n (N) i The number of images of which the model is judged to be the classification result of the i-type surrounding rock in the image of the circulation segment is represented; n represents the total number of images acquired for the loop at the current time.
3. The method according to claim 1, characterized in that:
the surrounding rock classification prediction model comprises a convolution layer, an activation function, a pooling layer, a full connection layer and a softmax function; wherein: the convolution layer, the activation function and the pooling layer are used for extracting surrounding rock image features, and the full-connection layer and the softmax function are used for calculating the prediction category of the surrounding rock image;
the training of the surrounding rock classification prediction model comprises the following steps:
taking the preprocessed rock residue image training set as input of a surrounding rock classification prediction model, and outputting the prediction category of the surrounding rock image;
and calculating the error between the predicted surrounding rock category and the real category by adopting the cross entropy loss function, and self-adaptively adjusting the learning rate and the model parameter according to the value of the cross entropy loss function until the training stop condition is met.
4. The method of claim 1, wherein the rock residue image acquisition device is located on a rear mating belt conveyor behind the TBM shield.
5. The method of claim 1, wherein the rock residue image acquisition device is comprised of a camera, a light supplement lamp, and a fixed bracket.
6. The method of claim 1, wherein the rock slag image preprocessing comprises:
s1, if the rock residue image is used for training, adopting a data enhancement technology of horizontal and vertical overturn for the rock residue image, and executing S2; otherwise, directly executing S2;
s2, processing the image into a uniform size.
7. The method of claim 1, wherein the deep convolutional neural network is a ResNet, VGG, squeezeNet model or other convolutional neural network.
8. The method according to claim 1, characterized in that: before the preprocessing of the rock slag image, carrying out data rejection processing, wherein the rejection data comprises the following steps: images of fault shooting, images of belt idling, images of rock residue coverage below 30% of shooting area and images of external interference.
9. Surrounding rock classification real-time prediction system based on TBM rock slag image, which is characterized in that: comprising a memory and a processor, said memory having stored thereon a computer program capable of being loaded by the processor and performing the method according to any of claims 1 to 8.
10. A computer-readable storage medium, characterized by: a computer program stored which can be loaded by a processor and which performs the method according to any one of claims 1 to 8.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310643903.0A CN116682010B (en) | 2023-06-01 | 2023-06-01 | Surrounding rock classification real-time prediction method based on TBM rock slag image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202310643903.0A CN116682010B (en) | 2023-06-01 | 2023-06-01 | Surrounding rock classification real-time prediction method based on TBM rock slag image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN116682010A true CN116682010A (en) | 2023-09-01 |
CN116682010B CN116682010B (en) | 2024-01-23 |
Family
ID=87784845
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310643903.0A Active CN116682010B (en) | 2023-06-01 | 2023-06-01 | Surrounding rock classification real-time prediction method based on TBM rock slag image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN116682010B (en) |
Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109886329A (en) * | 2019-02-18 | 2019-06-14 | 中国铁建重工集团有限公司 | Rock crusher level detection method, detection system and heading equipment |
CN109933577A (en) * | 2019-03-08 | 2019-06-25 | 山东大学 | Prediction technique and system can be tunneled based on TBM rock-machine dynamic state of parameters interaction mechanism tunnel |
CN110232419A (en) * | 2019-06-20 | 2019-09-13 | 东北大学 | A kind of method of side slope rock category automatic identification |
CN110245695A (en) * | 2019-05-30 | 2019-09-17 | 华中科技大学 | A kind of TBM rock slag order of magnitude recognition methods based on convolutional neural networks |
CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
CN112182973A (en) * | 2020-09-30 | 2021-01-05 | 山东大学 | TBM (Tunnel boring machine) operation parameter optimization method considering cutter vibration and rock slag geometric information |
CN112990227A (en) * | 2021-02-08 | 2021-06-18 | 中国铁建重工集团股份有限公司 | Face geology detection method |
CN113112446A (en) * | 2020-03-05 | 2021-07-13 | 成都理工大学 | Tunnel surrounding rock level intelligent judgment method based on residual convolutional neural network |
CN115272750A (en) * | 2022-07-07 | 2022-11-01 | 上海工业自动化仪表研究院有限公司 | TBM surrounding rock identification method and device based on optimized VGG16 network |
CN116066121A (en) * | 2023-02-20 | 2023-05-05 | 中铁工程装备集团有限公司 | TBM tunneling tunnel surrounding rock real-time grading method and shield tunneling machine |
US20230144184A1 (en) * | 2021-11-11 | 2023-05-11 | Shandong University | Advanced geological prediction method and system based on perception while drilling |
-
2023
- 2023-06-01 CN CN202310643903.0A patent/CN116682010B/en active Active
Patent Citations (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109886329A (en) * | 2019-02-18 | 2019-06-14 | 中国铁建重工集团有限公司 | Rock crusher level detection method, detection system and heading equipment |
CN109933577A (en) * | 2019-03-08 | 2019-06-25 | 山东大学 | Prediction technique and system can be tunneled based on TBM rock-machine dynamic state of parameters interaction mechanism tunnel |
CN110245695A (en) * | 2019-05-30 | 2019-09-17 | 华中科技大学 | A kind of TBM rock slag order of magnitude recognition methods based on convolutional neural networks |
CN110232419A (en) * | 2019-06-20 | 2019-09-13 | 东北大学 | A kind of method of side slope rock category automatic identification |
CN111368909A (en) * | 2020-03-03 | 2020-07-03 | 温州大学 | Vehicle logo identification method based on convolutional neural network depth features |
CN113112446A (en) * | 2020-03-05 | 2021-07-13 | 成都理工大学 | Tunnel surrounding rock level intelligent judgment method based on residual convolutional neural network |
CN112182973A (en) * | 2020-09-30 | 2021-01-05 | 山东大学 | TBM (Tunnel boring machine) operation parameter optimization method considering cutter vibration and rock slag geometric information |
CN112990227A (en) * | 2021-02-08 | 2021-06-18 | 中国铁建重工集团股份有限公司 | Face geology detection method |
US20230144184A1 (en) * | 2021-11-11 | 2023-05-11 | Shandong University | Advanced geological prediction method and system based on perception while drilling |
CN115272750A (en) * | 2022-07-07 | 2022-11-01 | 上海工业自动化仪表研究院有限公司 | TBM surrounding rock identification method and device based on optimized VGG16 network |
CN116066121A (en) * | 2023-02-20 | 2023-05-05 | 中铁工程装备集团有限公司 | TBM tunneling tunnel surrounding rock real-time grading method and shield tunneling machine |
Non-Patent Citations (3)
Title |
---|
A.-S. CAPELLE-LAIZE; C. FERNANDEZ-MALOIGNE;: "《TBM for color image processing: a quantization algorithm》", 《YEAR: 2006 | CONFERENCE PAPER | PUBLISHER: IEEE》 * |
于国新;白明洲;许兆义: "《铁路隧道围岩分级的知识挖掘》", 《铁道学报》 * |
何新成;: "基于LIBSVM算法在隧道围岩分级上的应用", 公路, no. 04 * |
Also Published As
Publication number | Publication date |
---|---|
CN116682010B (en) | 2024-01-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110569843B (en) | Intelligent detection and identification method for mine target | |
CN111881811B (en) | Self-accretion attention pooling mechanism-based rock classification method for tunneling equipment | |
CN111832607B (en) | Bridge disease real-time detection method based on model pruning | |
CN112347943A (en) | Anchor optimization safety helmet detection method based on YOLOV4 | |
CN113780719A (en) | Comprehensive evaluation method for mine geological environment | |
Zhang et al. | Deep learning‐based automatic detection of muck types for earth pressure balance shield tunneling in soft ground | |
CN115408925A (en) | Rock mass parameter prediction method and device for tunnel construction | |
CN115422822A (en) | Tunnel rock mass parameter prediction method and device | |
CN113255580A (en) | Method and device for identifying sprinkled objects and vehicle sprinkling and leaking | |
Chiffard et al. | Adaptive niche-based sampling to improve ability to find rare and elusive species: Simulations and field tests | |
CN115481565A (en) | Earth pressure balance shield tunneling parameter prediction method based on LSTM and ant colony algorithm | |
CN116682010B (en) | Surrounding rock classification real-time prediction method based on TBM rock slag image | |
Pan et al. | Intelligent rock fracture identification based on image semantic segmentation: methodology and application | |
CN117272139A (en) | Shield tunneling machine attitude prediction method combining characteristic correlation and time correlation | |
CN117423157A (en) | Mine abnormal video action understanding method combining migration learning and regional invasion | |
CN109766780A (en) | A kind of ship smog emission on-line checking and method for tracing based on deep learning | |
CN112905832B (en) | Complex background fine-grained image retrieval system and method | |
CN104463137A (en) | Anomaly facial image detection method and system based on characteristic space decomposition | |
Lee et al. | Accelerating multi-class defect detection of building façades using knowledge distillation of DCNN-based model | |
CN115359091A (en) | Armor plate detection tracking method for mobile robot | |
CN114581666A (en) | Intelligent rock sample identification method and system based on image deep learning | |
CN112883889A (en) | Method suitable for detecting violation behaviors in power construction scene | |
CN113033289A (en) | Safety helmet wearing inspection method, device and system based on DSSD algorithm | |
Tundia et al. | Aiding Irrigation Census in Developing Countries by Detecting Minor Irrigation Structures from Satellite Imagery. | |
Valencia et al. | Blasthole Location Detection Using Support Vector Machine and Convolutional Neural Networks on UAV Images and Photogrammetry Models |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |