KR102344741B1 - Method for analysis of blind ends in tunnels using artificial intelligence and Method for predicting front blind ends - Google Patents

Abstract

Provided, in the present invention, is a method comprising: a step of allowing a blind end surface image information input part for learning (110) to load the blind end surface image information for learning prepared in advance for learning, allowing the analysis information input part (120) for learning to load the analysis information for learning corresponding to the blind end surface image information, and setting the learning information by mapping the loaded blind end surface image information for learning and the analysis information for learning; and a step of constructing a blind end surface automatic analysis system by repeatedly performing the above step for each of a plurality of pre-prepared blind end surface images for learning. Therefore, the present invention is capable of significantly shortening a construction time of the blind end surface automatic analysis system.

Description

Method for analysis of blind ends in tunnels using artificial intelligence and Method for predicting front blind ends using artificial intelligence

The present invention relates to a method for automatically analyzing a tunnel scene using artificial intelligence and a method for predicting a front end scene using the same.

Construction of tunnel construction is, for example, made in the order of charging and blasting, ventilation, blocking treatment, observing and analyzing blind ends, determining support and excavation patterns, installing support materials, etc., and charging and It returns to the blasting stage and is repeated multiple times for a specified period.

Here, the accuracy of the scene observation and analysis is very important. A safe and effective follow-up operation, that is, the next stage of excavation inside the makjang, is possible only when the analysis of carcinoma, rock quality, joint, and groundwater identified in the makjang is accurately performed.

The speed of observation and analysis of the scenes is also important. In case of blasting more than twice a day in blast excavation, the analysis of the scene must be completed within 15 minutes. This is because it takes a lot of time to charge, blast, ventilate, handle bugs, and install support materials.

Various methods for this are proposed in the prior art.

Japanese Patent Application Laid-Open No. 2019-023392A proposes a system for evaluating the curtains using machine learning. Although the present prior art does not present a specific machine learning method, when a general machine learning method (Softmax classification, SVM, etc.) is used, the accuracy of the final evaluation result is 30 to 40%, so there is a big problem that the accuracy is low. In addition, it takes a considerable amount of time to construct the system according to machine learning and to infer the output layer according to the analysis of the scene image in the field, so the possibility of actual field application is very low.

Japanese Patent No. 4094770B2 proposes a bedrock determination system. Although artificial intelligence such as fuzzy logic is used, there is a problem that bedrock sample extraction is required and that bedrock analysis equipment must be provided at the tunnel construction site because it uses real bedrock data confirmed by an excavator rather than image analysis.

European Patent Registration No. 1176393B1 proposes a system for predicting the joint joint using machine learning. A three-dimensional camera needs to be provided for each site, and there is a problem in that it is not possible to determine carcinomas and types other than joints.

As such, there have been attempts to automatically analyze the scene using artificial intelligence in the prior art. At the tunnel construction site, scene analysis using artificial intelligence is not applied.

Currently, a program that does not use artificial intelligence and supports manual analysis by a user directly classifying an area on an image captured by a camera is often used even today. However, this is only a simple program in which a correction function is added after loading an image, and when a non-expert uses the program, the accuracy is lowered, so it is difficult to guarantee the quality and the analysis time is also increased.

In addition, in order to use artificial intelligence, it is necessary to build an automatic analysis system using the prepared learning scenes information. However, since it is necessary for an expert to manually prepare the learning scenes information used in the process of constructing such an automatic analysis system, it is also necessary to shorten this process.

(Patent Document 1) JP 2019-023392A

(Patent Document 2) JP 2018-071165A

(Patent Document 3) KR 10-2009-0020359A

(Patent Document 4) JP 4094770B2

(Patent Document 5) EP 1176393B1

(Patent Document 6) JP 2019-065648A

The present invention has been devised to solve the above problems.

The important things in analyzing the scene image information are accuracy, speed, and possibility of modification by the user. According to the prior art, accuracy was low in the method using artificial intelligence, and there was also a problem in speed. It could speed up the inference time, but in this case the accuracy would be lower. In some cases, a user interface that can be modified by the user was provided, but it was difficult for non-experts to access it. The present invention is to solve these problems, and proposes a method in which accuracy, speed, and possibility of user's modification can all be provided.

In addition, in the prior art, there was a inconvenience in that the operator (expert) manually input the scene information for learning, which leads to a bigger problem in that an error is generated when a person directly inputs and judges directly. The present invention intends to propose a method that can more easily build an automatic analysis system for scenes by solving these problems.

An embodiment of the present invention for solving the above problems, learning to use the scene image information as an input layer and digital face mapping (DFM) analysis information to the scene image information as an output layer A method of performing analysis on input image information of a scene through an automatic analysis system using the scene, - Here, the analysis information includes information about carcinoma, information about cancer quality, and information about joints- (A -1) The learning scene image information input unit 110 loads the learning scene image information prepared in advance for learning, and the learning analysis information input unit 120 loads the learning analysis information corresponding to the closing scene image information, and the By mapping the loaded scene image information for learning and the analysis information for learning, setting learning information (S110); And (B) by repeatedly performing the step (A-1) for each of the plurality of pre-prepared scenes for learning, building an automatic analysis system for scenes (S200); including, wherein the learning information set in step (S110) of (A-1) is generated by detecting a cluster range of the same target in a scene and expressing a target region, and the target is carcinoma, cancer, and joint Including, (T1) expressing the correction region of the target region using the reference line connecting the specific points included in the target region and the height of each specific point, and in the order of the excavation depth of the final correction region a first manner generated using a rate of change; (T2) a second method in which the first and second target areas are expressed in the order of the excavation depth of the makjang surface, and the first and second target areas are generated using a center point distance; or (T3) a third method in which the target region is divided into multiple equal parts to express the divided region, and is generated using the rate of change of a specific point included in the divided region in the order of the excavation depth of the final scene; A method is provided, using any one of them.

In addition, in the first method, (T-11) the first point P1 of the first height, the second point P2 of the second height, and the third height at the boundary line b1 of the target area SZ after confirming the third point P3 of , forming a reference line SL connecting the second point P2 and the third point P3 ( S11 ); (T-12) a first height H1 between the boundary line b1 passing through all of the first to third points P1, P2, and P3 and the reference line SL, and a second height of the target area When the ratio of the height H2 is equal to or greater than the threshold value, by deleting the target range in the direction of the first point P1 with respect to the reference line SL as a boundary, the correction area of the target area is expressed (S12); and (T-13) confirming the rate of change of previous correction areas in the order of the excavation depth of the final scene, and predicting the image information and analysis information of the final scene in front using the rate of change, thereby generating the learning information (S13) ); may include

In addition, in the step (T-13), vertical lines passing through the midpoint of each of the final scenes are arranged in the order of the excavation depth, and the high and low points of the correction region are displayed on the vertical lines, but the high and low points are the correction The upper boundary line and the lower boundary line of the region are intersection points where they intersect the reference line, respectively, and the vertical axis view may be generated by connecting the high and low points of each vertical line to track the change of the correction region by image tracking.

In addition, in the step (T-12), the first height is a length between intersection points where a vertical line passing through the midpoint of the film scene intersects a boundary line and a reference line, respectively, and the second height is the vertical line of the target area It may be the length between intersections crossing the boundary line.

In addition, in the second method, (T-21) if the distance between the first central point of the first target region and the second central point of the second target region is less than a threshold value, the first and second target regions are homogeneous regions setting to (S21); and (T-22) confirming the rate of change of the previous homogeneous areas in the order of the excavation depth of the last scene, and predicting the target area of the front end face using the rate of change (S22); may include

In addition, in the step (T-22), the high and low points of the first target area are displayed on a first vertical line passing through the first center point, and the second target area is displayed on a second vertical line passing through the second center point. Displaying the high and low points of the area, arranging the first and second vertical lines in the order of the depth of excavation, so as to track the change of the first and second target area by image tracking, homogeneous for each of the first and second vertical lines A vertical axis can be created by connecting the high and low points of the region, respectively.

In addition, the third method includes: (T-31) extracting one end point and the other end point of the divided region, respectively (S31); and (T-32) in the order of excavation depth, by confirming the rate of change of one end point and the other end point for each division area of the previous target areas of the same quality, and predicting the one end point and the other end point for each division area of the target area of the front end scene. Predicting the target area of the front scene (S32); may include

In addition, the division region divides the target region in the horizontal direction, and in step (T-32), the one end and the other end point are the highest and lowest points of each divided region, and the one and the other end points, respectively The rate of change of may include a rate of change of the distance between the one end point and the other end point.

In addition, (T-30) before expressing the divided area by dividing the target area into multiple equal parts, the center points of the previous target areas are detected in the order of the excavation depth, and if the distance between the center points of the previous target areas is less than the threshold value, setting the corresponding previous target areas to be homogeneous with each other (S30); may further include.

In addition, in the present invention, in the above-described method, after step (A-1), (A-2) the learning information dividing unit 130 crops the learning information as a preset unit to learn one learning. Generating a plurality of divided learning information for each information (S120); (A-3) converting the format conversion unit 210 into a learning format by loading the plurality of divided learning information (S130); and (A-4) performing, by the learning performing unit 230, a plurality of learning times for each single scene image information using the format-converted plurality of divided learning information (S140); Further comprising, after the (B) step, (C) when the scene image information is input to the input unit 310, the inference unit 330 infers the analysis information using the automatic analysis system for the scene scene, and outputs The unit 350 outputs it (S300); may further include.

In addition, after the step (B), (B-1) the validation unit 240, using a plurality of other learning scenes image information to verify the validity of the automatic scene analysis system built in the step (B) performing (S210); Further comprising, the learning end scene image information and analysis information corresponding thereto are prepared N pieces, the learning end scene image information performed in the step (B) is A (N>A), the (B- Other learning scenes image information performed in step 1) may be NA pieces.

In addition, the step (C) includes: (C-11) inputting the scene image information to the input unit 310 (S311); (C-12) A step in which a tunnel boundary node is input on the inputted scene image information to the image correction unit 320, the input tunnel boundary node is connected, and only the connected area is extracted, thereby modifying the closing scene image information (S312); and (C-13) the inference unit 330 infers the analysis information using the modified automatic scene analysis system, and the output unit 350 outputs it (S313); may include

In addition, the step (C), (C-21) the step of inputting the scene image information to the input unit 310 (S321); (C-22) the image correction unit 320 dividing the input scene image information into a plurality of regions using a watershed technique (S322); (C-23) Inferring, by the inference unit 330, analysis information for each region through the automatic scene analysis system by using each of a plurality of regions of the input scene image information as an input layer (S323) ; and (C-24) the inference unit 330 integrates the analysis information for each region to infer the digital face-mapped analysis information for the input scene image information, and the output unit 350 outputs it to (S324); may include

On the other hand, the present invention is a method for predicting a front curtain scene through the prediction system 10 using the above-described method, and after step (C), (C-1) inferred by the inference unit 330 The analysis information of the scene output from the output unit 350 is input to the prediction system 10 (S410); (C-2) generating a target region image by inputting the analysis information of the scene scene to the scene scene analysis module 12 provided in the prediction system 10 (S420); (C-3) processing, by the scene processing module 13 provided in the prediction system 10, the target region image expressed by the scene analysis module 12 in a preset manner (S430); and (C-4) predicting the front scene using the image processed by the scene processing module 13 through the scene prediction module 14 provided in the prediction system 10, but Predicting the front curtain scene using any one of the first to third methods (S440); It provides a forward curtain scene prediction method further comprising a.

The use of the foundation according to the present invention has the following advantages.

The present invention has the following advantages.

First, it is very easy to construct an automatic analysis system for scenes. The automatic scene analysis system constructed according to the present invention uses artificial intelligence and is built through learning. In the state in which the system is built, the scene is automatically analyzed, but in the process of building, it is necessary for the operator to directly input learning information, and this learning information is determined manually by the operator. However, according to the present invention, since the learning information is automatically generated by a specific method, risks such as mistakes made by the operator and subjective judgment criteria are removed, so that reliability can be further improved. In addition, since the operator's direct input and direct judgment process is omitted, the construction time of the automatic scene analysis system can be significantly accelerated.

Second, high-accuracy analysis information can be checked even by non-experts. Through learning, the last information is analyzed with high accuracy. As will be described later, the accuracy according to the prior art using artificial intelligence was only 30 to 40%, but the accuracy of 70% or more was verified in the embodiment actually implemented according to the present invention. A number of training sets verified by experts are used, and when real users are used, the verification results are recycled as learning information to further increase the accuracy.

Third, the analysis is quick. In order to proceed with the next stage of charging, blasting, and ventilation, the analysis of the scene must be completed within 15 minutes as described above. One of the reasons for using artificial intelligence is for such a rapid analysis, but in the prior art, it was difficult to implement due to the large amount of computation of artificial intelligence and reasoning itself. The present invention was implemented to enable rapid analysis in various ways. By dividing and learning the image, using the watershed technique, modifying and using the inception algorithm, and setting the optimal parameters, rapid calculation was implemented.

Fourth, user modification is free. Although the results of the primary analysis are provided by artificial intelligence, experts in the field can freely modify them while checking them. By adopting the node method for the free movement of cancer types expressed in polygons and joints expressed in rock quality and line segments, users can intuitively edit, which further reduces the time required to derive final analysis results.

In addition, the RQD (rock quality index) is automatically calculated, a report format is available, and a user interface is provided to input the joint strike slope and joint correction score.

There is no need to prepare separate equipment on site that is large, expensive, or difficult to maintain. If only one information processing device (eg, cell phone, tablet, notebook, etc.) having a camera and communication function is provided in the field, the method according to the present invention can be sufficiently performed.

1 schematically shows a system for carrying out a method according to the invention;
2 shows a flow chart for carrying out the method according to the invention;
3 shows an example of the scene image information for learning and the analysis information for learning.
4 and 5 show examples of a method of dividing learning information.
6 exemplarily shows a network of the learning method used in the present invention.
7 shows an example of a method of modifying the scene image information in the present invention.
8 shows an example of a watershed technique used to correct the scene image information in the present invention.
9A to 9C show examples of canny edge detection techniques used in the present invention.
10 shows an example in which analysis information, which is a result inferred according to the present invention, is output.
11 shows an example of various RQD radiation center correction methods used in the present invention.
12 shows an example of a joint correction method by a user in the present invention.
13 shows an example in which the joint strike slope and joint joint correction score are input by the user in the present invention.
14 shows an example of a polygon modification method by a user in the present invention.
15 is an image showing an image window output through the output means of the prediction system according to the present invention.
16 is a diagram schematically illustrating a state in which an attribute value of a dark region is checked according to the forward prediction method according to the present invention.
17 is a diagram illustrating a state in which a correction area is expressed in a dark area according to the forward prediction method according to the present invention.
18 is an image showing the vertical axis of the correction region expressed according to the forward prediction method according to the present invention.
19 is a diagram schematically illustrating a dark area for prediction through the forward prediction method according to the present invention.
FIG. 20 is an image showing an example for predicting an anterior dark mass of the dark mass of FIG. 19 .
21 is a diagram schematically illustrating a state in which an attribute value of a dark region is checked according to the forward prediction method according to the present invention.
22 to 24 are diagrams schematically illustrating a state in which a vertical line of a dark area is displayed according to the forward prediction method according to the present invention.
25 is an image showing a vertical axis view corresponding to the surface of each excavation depth according to the forward prediction method according to the present invention.
26 is a diagram schematically illustrating a state in which an attribute value of a dark area is checked according to the forward prediction method according to the present invention.
27 is a view showing a change in the position of one end point of the previous divided regions according to the forward prediction method according to the present invention.

Hereinafter, "learning" refers to when an information processing device such as a computer applies input information to the input layer, it undergoes predetermined training to automatically infer the corresponding output information to build an algorithm that outputs it to the output layer. It refers to the data processing technique that is performed and is a kind of artificial intelligence. Machine learning may be included. Learning consists of training using a plurality of training sets in which an input layer and an output layer are determined. Learning is performed by using a predetermined algorithm as the subject of the information processing device, and when the algorithm is built after performing a number of learning, it is possible to automatically perform inference in a predetermined manner according to the information input by the user. do. The algorithm used here may vary. Below, as a learning algorithm to be used, inception is proposed as an example.

Hereinafter, "digital face mapping (DFM)" means mapping digitized information on a photographed image. By the DFM according to the present invention, information on carcinoma, cancer quality, joint, etc. is digitally mapped on a scene image taken by a user with a camera or the like. In other words, when a user inputs an image of a scene using the method according to the present invention, analysis information is automatically inferred and output is superimposed on the image, and this analysis result is a DFM result. As shown in the upper part of FIG. 10 , the joints are shown as line segments, or as shown in the lower part of FIG. 10 , regions of carcinoma (eg, granite, etc.) and rock quality (eg, severe weathering, etc.) may be separated and shown.

Description of the system

In this method, analysis information on the inputted scene image information through an automatic analysis system using the learning that uses the scene image information as an input layer and the analysis information digital face mapped to the scene image information as an output layer It relates to a method of automatically outputting . Here, the analysis information includes information on carcinoma, information on cancer quality, and information on joints.

An automatic scene analysis system for performing the method according to the present invention will be described with reference to FIG. 1 .

A system in which the method according to the present invention is performed includes: a learning information generating module 100 for generating learning information; a learning module 200 for constructing an automatic scene analysis system by performing learning based on learning information; DFM module 300 for automatically inferring and outputting information such as carcinoma, cancer quality, joint, etc. by analyzing the image of the scene input by the user based on the built-up automatic analysis system; a user modification module 400 that provides a function that a user can modify based on the inferred analysis information; and a learning information addition module 500 that converts the final modified analysis information and the scene image information back into learning information to enable learning.

The learning information generation module 100 includes a scene image information input unit 110 for learning, an analysis information input unit 120 for learning, a learning information division unit 130 , and a learning information database 190 . Here, the learning information generating module 100 is configured to receive learning information from the prediction system 10 . The prediction system 10 automatically maps the learning scene image information and the learning analysis information in a manner to be described later, which will be described separately later.

The learning module 200 includes a format conversion unit 210 , a learning parameter setting unit 220 , a learning execution unit 230 , and a validity verification unit 240 .

The DFM module 300 includes an input unit 310 and an image correction unit 320 . It includes an inference unit 330 , a joint extraction unit 340 , an output unit 350 , an RQD calculation unit 360 , and a DFM database 390 .

Each function is described in more detail below.

Preparation of learning information

A method of preparing learning information for performing the method according to the present invention will be described with further reference to FIG. 2 .

Prepare the image information of the scene for learning as the input layer and the analysis information for learning as the output layer that analyzes it. It is preferable that the analysis information for learning is information verified by an expert. The scene image information for learning and the analysis information for learning are mapped to each other.

If there are N pieces of image information for learning scenes and corresponding analysis information for learning, some of them (A pieces) are used for learning and the rest (N-A pieces) are used for verification (N>A).

In the real implementation system, 900 out of 1000 learning scenes image information was used for learning, and the remaining 100 were used for verification.

The learning scene image information input unit 110 loads the learning scene image information prepared in advance for learning, and the learning analysis information input unit 120 loads the learning analysis information corresponding to the closing scene image information. The prepared scene image information for learning and the analysis information for learning are referred to as learning information (FIG. 3).

It is important not to learn it as it is, but to learn it by dividing it. This is to learn a lot of learning information quickly by reducing the amount of computation. To this end, the learning information dividing unit 130 generates a plurality of divided learning information for each single scene image information by cropping the prepared learning information as a preset unit.

In the actual implementation system, it was divided into pixel units of 30X30 and stored (FIG. 4). As learning information, a plurality of scene image information and corresponding analysis information are mapped and acquired, and the divided information is stored in the learning information database 190 as an individual file (FIG. 5).

Meanwhile, the preparation of learning information according to the present invention will be separately described later.

Learning method

The learning module 200 performs learning using the divided learning information prepared in the learning information generation module 100 .

First, the format conversion unit 210 loads a plurality of divided learning information and converts it into a learning format.

In the actual implementation embodiment, the learning information was converted to .tfrecord so that learning can proceed using TF-Slim of tensorflow, which includes the width, height, depth, and label of the image identified in the scene image information. (label), raw image (image raw), etc. are included.

Next, the learning performing unit 230 uses the format-converted plurality of divided learning information, but while converting a learning rate and a training step for a predetermined number of times, learning a plurality of times for each divided learning information carry out In this process, the learning parameter setting unit 220 may set the optimal parameter so that the amount of calculation is minimized. The optimal parameters will be described below.

Through this process, learning is performed several times for each piece of divided learning information, and learning is performed several times for each piece of learning information in which a plurality of divided learning information are gathered.

In the actual implementation embodiment, the learning method used by the learning performing unit 230 is Inception. In a deep learning network, the deeper and the wider the layer, the better the performance, but the amount of computation also increases exponentially, so it is not suitable for learning progress for pumice scenes such as the present invention. Inception reduces the number of connections between nodes in order to reduce the amount of computation, and makes the matrix operation a dense operation. The inception algorithm is adopted in consideration of the specificity of the image analysis of the scene as in the present invention (the size of the image itself is wide, the identification of the boundary between carcinoma and rock quality is important, and the segmentation of line segments such as joints is important).

Inception is a method of reducing the number of parameters in 3X3 and 5X5 convolutional layers by reducing the image channel by 1X1 convolution operation. Accordingly, while configuring a network deeper than CNN (Convolutional Neural Networks), the number of parameters is reduced and the amount of computation is reduced.

Specifically, in general inception, (1) 1X1 convolution for the previous layer, (2) 1X1 convolution and 3X3 convolution, (3) 1X1 convolution and 5X5 convolution, and (4) 3X3 max pooling and 1X1 convolution perform 4 types of operations and combine them into one to proceed with learning.

In the present invention, while using inception, the need to perform 7X7 convolution operation in the first layer (stem layer) is replaced with 3x3 convolution operation 3 times, the optimizer is set to RMSProp, and the last layer (fully connected layer) ), batch normalization (BN) was applied. In addition to this, factorization is applied to further reduce the amount of computation and time by improving the kernel, performing 2 3X3 convolution operations instead of 1 5X5 convolution operation, and 2 3X1 convolution operations instead of 1 3X3 convolution. improved. As a result, it was confirmed that the calculation amount was reduced by about 2.78 times. 6 schematically shows a network used in this way.

Meanwhile, the learning parameter setting unit 220 sets the final parameter according to the result of performing learning while changing the learning rate and the training stage. The parameters set here can be divided into optimization parameters and learning parameters.

As an optimizer parameter, weight_decay, Optimizer, opt_epsilon, rmsprop_momentum, rmsprop_decay, etc. may be mentioned.

weight_decay means the weight for optimizing the loss function of the model. Here, it was set to 0.00004 according to the optimization result.

opt_epsilon is the smallest value to prevent the case where the calculated value becomes 0. Here, it was set to 1 according to the optimization result.

rmsprop_momentum is the momentum value of the RMSPropOptimizer. Here, it was set to 0.9 according to the optimization result.

rmsprop_decay is the rate of decrease of the probability applied for each update. Here, it was set to 0.9 according to the optimization result.

As a learning parameter, learning_rate, max_number_of_steps, learning_rate_decay_type, end_learning_rate, learning_rate_decay_factor, num_epochs_per_decay, replicas_to_aggregate, moving_average_decay, etc. may be mentioned.

learning_rate is the initial learning rate. Here, it was set to 0.0001 according to the optimization result.

max_number_of_steps is the maximum number of training steps. Here, it was set to 1000 according to the optimization result.

learning_rate_decay_type is a parameter that specifies how to decrease the learning rate. These include fixed, exponential, and polynomial. Here, it was set to fixed according to the optimization result.

end_learning_rate is the minimum learning progress rate used when polynomial is set in the learning rate reduction method. Here, it was set to 0.0001 according to the optimization result.

learning_rate_decay_factor is the learning decay rate when exponential is set in the learning rate reduction method. Here, it was set to 0.94 according to the optimization result.

num_epochs_per_decay is the number of epochs after the learning rate is reduced. Here, it was set to 2 according to the optimization result.

replicas_to_aggregate is the number of gradients to collect before updating the parameter. Here, it was set to 1 according to the optimization result.

moving_average_decay is the decay rate used for the moving average. Here, it is set to None according to the optimization result.

In this way, learning is performed several times for each divided learning information in one cut-scene image information for learning, and this learning is performed on a plurality of cut-scene image information for learning, thereby constructing an automatic final scene analysis system.

The validation unit 240 confirms the built system by validating the validity by using other unused scene image information among the prepared learning information.

As a result of checking in the actual implementation example, it was confirmed that the accuracy was 70% or more, and thus the accuracy was relatively high. It showed a big difference compared to the case where a learning algorithm other than Inception was used.

[Table 1]

Acquisition of analysis information using DFM

Now, the user can automatically acquire analysis information such as carcinoma, rock quality, joint, etc. by inputting the image information of the scene using the built-up automatic analysis system for the scene.

First, makjangmyeon image information is input to the input unit 310 . The user can directly obtain and input the scene image information by shooting with a camera while using an information processing device equipped with a communication function and a camera, or select an image stored in the memory in the information processing device and use the input unit 310 It can also be entered through In the field, only information processing equipment is sufficient, and no separate equipment is required.

The image correction unit 320 corrects the image to be input in several ways.

The input image taken by the user further includes information about boundary objects such as surrounding walls as well as the curtain screen (see the lower left of FIG. 7), and the present invention provides a correction function for extracting the image for the curtain screen . That is, an image is input (top of FIG. 7 ) and a tunnel interface node may be input on the scene image information input by the image correction unit 320 (bottom left of FIG. 7 ), and the input tunnel interface node is connected and only the connected area is extracted (bottom right of FIG. 7), so that the scene image information is modified. Most of the prior art selects the method of correcting the image rather than the method of selecting a part from the image. In this case, distortion may occur in the image, and the distortion in the scene analysis can seriously affect the next blasting process, The present invention selects an area extraction method by node.

In addition, as shown in FIG. 8 , the image correction unit 320 may divide the input scene image information into a plurality of regions by using a watershed technique. This reduces the amount of computation when inputting the scene image information input as the input layer, thereby enabling rapid computation and inference in the reasoning unit 330 to be described later. As described in the prior art, speed is very important in image analysis of a scene.

Next, the inference unit 330 infers the analysis information using a system constructed for each divided area, and infers the digital face-mapped analysis information for the input scene image information by integrating the analysis information. That is, the inference unit 330 inputs the scene image information to the input layer of the built-up automatic scene analysis system, and when checking the output layer by artificial intelligence, it is inferred as analysis information.

On the other hand, in another embodiment of the present invention, the joint extraction may be made separately. The joint extraction unit 340 extracts a line segment from the input scene image information using a canny edge detection technique (FIG. 9a). Next, the joint extraction unit 340 may remove noise by performing a predetermined test for each extracted line segment (FIG. 9B). Also, if the length of the line segment is less than or equal to a predetermined pixel value, it can be removed from the extraction result. It may also be (Fig. 9c). Now, the remaining line segments will be output as joints.

When all the analysis information is arranged through this process, the output unit 350 outputs the analysis information as a result (FIG. 10). The result of this method may be stored in the DFM database 390, may undergo a user modification process to be described later, and may be used again as learning information to increase accuracy. It may be output in the form of a report as shown in the lower right of FIG. 10 .

In another embodiment of the present invention, the RQD (rock quality designation; rock quality index) may be calculated and further output. The RQD calculation can be performed by any well-known method, but the important thing is to automatically set the RQD radiation center. That is, the RQD calculation unit 360 includes the steps of setting the center of the RQD radiation line, and calculating and outputting the RQD using the center of the RQD radiation line.

As a method for the RQD calculation unit 360 to set the center of the RQD radiation, the present invention provides four methods. First, it is possible to set the center of the RQD radiation line as the center of the points constituting the joint among the analysis information (top left of FIG. 11 ). Second, the center of the RQD radiation line may be set as the center of the center point of the line segments constituting the joint in the analysis information (top right of FIG. 11 ). That is, it is set based on the densely populated area. Third, the center of the RQD radiation line can be set as the center of the longest line segment among the line segments constituting the joint among the analysis information (lower left of FIG. 11 ). Fourth, the center of the RQD radiation line can be set as the center of the horizontal radiation line that intersects the joint most in the analysis information among a plurality of horizontal radiation lines set on the scene image information (bottom right of FIG. 11 ).

Each method may be necessary depending on the scene situation and the site situation. Accordingly, in the present invention, all of these methods may be performed automatically or may be performed by a user selecting a specific method.

user edit module

The present invention provides a function that allows an expert who has confirmed the analysis information to be automatically output by learning to correct carcinoma, cancer quality, joint, and the like. In particular, by adopting a convenient and intuitive method, experts can quickly make corrections and confirm analysis information. To this end, the user modification module 400 provides a function that the user can modify based on the analysis information automatically inferred by the output unit 350 .

The output unit 350 outputs the joint as a line segment, and the user correction module 400 may output the joint by adding a joint node on the joint. The output joint can be added, moved, and deleted by the user correction module 400 , and the output joint node can also be added, moved, and deleted by the user correction module 400 . When the joint node is moved, the joint shape including the moved joint node is changed accordingly (FIG. 12). A joint strike slope (FIG. 13(A)) and a joint correction score may be additionally input (FIG. 13(B)).

In addition, the output unit 350 outputs the area where the carcinoma and the rock quality are distinguished as a polygon and, if necessary, also divides the groundwater into a polygon and outputs it. have. The output polygonal node can be added, moved, and deleted by the user modification module 400, and when the polygonal node is moved, the polygonal shape including the moved polygonal node is changed accordingly (FIG. 14).

Add learning information

The analysis information that has been more reliably corrected by the experts through the user modification process can be used as learning information together with the learning scene image information used in the initial learning. In particular, since it has undergone a user modification process, a training set with excellent data accuracy is used. This increases the analysis accuracy.

The learning information addition module 500 may input the analysis information modified by the user and the corresponding scene image information into the analysis information input unit 120 for learning and the scene image information input unit 110 for learning, respectively, and the input The information is utilized as learning information through the above-described steps.

Creation of learning information through the prediction system

Hereinafter, a process in which the learning information provided to the above-described learning information generating module 100 is automatically generated using the prediction system 10 will be described.

Referring to FIG. 1 , the prediction system 10 according to the present embodiment includes: an actual measured scene information DB 11 for storing and managing information about a dark area of an actual measured scene by excavation depth; a mak scene analysis module 12 for defining a rocky region by analyzing the ground information of the mak scene; a mak scene processing module 13 that calculates the dark region image expressed by the mak scene analysis module 12 according to the set value to check the attribute value and process; a scene prediction module 14 for predicting the dark area of the front scene by calculating the dark area image and attribute values with the dark quality area information of the actual scene found in the actual scene information DB 11; includes The editing status of the prediction system 10 is displayed through the output means 20 while the respective modules 11, 12, 13 are processing.

The measured scene information DB 11 stores and manages the information on the dark area of the scene measured for each depth of the excavation tunnel. In general, after the excavation tunnel is excavated at a specified interval, the makjang surface corresponding to the incision is measured and analyzed. The information on the Makjangmyeon confirmed at this time is diverse, such as joints, carcinomas, and rock quality, which are key information that can be used to analyze the ground.

In the following description, the target of information about the makjangmyeon is represented by 'rock quality'. However, the target measured by the operator for Makjangmyeon is not limited to the rock quality, and includes all types of investigations necessary for the analysis of carcinomas, joints, and other ground. In addition, the actual measurement scene information DB 11 stores and manages reports such as face mapping, cancer determination result information, and rock grade (RMR), which are basic data of rock quality area information.

The makjang scene analysis module 12 defines a rocky region by analyzing the ground information of the makjang scene. The ground of the final scene confirmed after excavation at the site is analyzed by a professional worker, and detailed information for mapping is collected using equipment such as digital camera light laser scanning.

The ground information collected in this way is analyzed through the setting process of the mak scene analysis module 12, and a specific rocky area is identified and detected within the mak scene. The dark zone according to the present invention includes a position value for visually confirming the range of the designated target. Here, the position value is a code for 2D image processing, and a pixel coordinate is set according to the code. In addition, since codes corresponding to the same dark areas are grouped and processed collectively, the pixels of the output means 20 corresponding to the codes are executed and the corresponding dark areas are visualized as graphic images.

Therefore, the scene analysis module 12 generates a dark area image based on the position value and displays it through the output means 20 . In addition, through the analysis of digital face mapping data (XML), point information such as joint and carcinoma is extracted and expressed as a web screen graph. In addition, the dark quality region may also include specification information of its own rock quality.

The mak scene processing module 13 calculates the dark area image expressed by the mak scene analysis module 12 according to the set value to check the attribute value and process it. As described above, since the dark region includes the position value, it is possible to determine attribute values for the position and length of the boundary line of the dark region image, the midpoint of the dark region image, and the like. In addition, various points and lines belonging to the dark region image may be additionally drawn based on the attribute value. A more detailed description of the scene processing module 13 will be described in detail while describing the following examples.

The final scene prediction module 14 calculates the dark quality region image and the attribute value with the dark quality region information of the actual final scene retrieved from the actual measured scene information DB 11 to predict the dark region of the front scene. The front makjangmyeon can be predicted based on a section of an unexcavated tunnel, or on the basis of the information on the rocky area of the actual makjangmyeon collected so far. Prediction of the front end scene will be described in detail by explaining the following examples.

The specific operation of the above-described prediction system 10 will be described in detail along with the description of the forward prediction method of the present embodiment.

15 is an image showing an image window output through the output means of the prediction system according to the present invention, and FIG. 16 is a schematic diagram showing a state of confirming the attribute value of the dark area according to the forward prediction method according to the present invention. It is a drawing.

15 and 16, the ground trend-based forward prediction method according to the measurement data of the final scene of the tunnel according to this embodiment detects the grouping range of the same rock quality in the last scene (TZ) of the excavation tunnel and detects the rocky region (SZ). ) a first step of expressing; a second step of confirming the first point Pl of the first height, the second point PZ of the second height, and the third point P3 of the third height at the boundary line b1 of the rocky region SZ; a third step of forming a reference line SL connecting the second point P2 and the third point P3; The first height Hl and H3 between the boundary lines BL1 and BL2 and the reference line SL passing through the first point Pl to the third point P3, and the second height TH of the dark area SZ , H2) is greater than or equal to the threshold value, a fourth step of expressing the correction region SZ' of the dark matter region SZ by deleting the dark quality range in the direction of the first point Pl with the reference line SL as a static system; a fifth step of confirming the rate of change of the previous correction areas in the order of the excavation depth of the makjangmyeon (TZ), and predicting the dark area of the front makjangmyeon; includes

Each step is described in more detail.

1st stage, dark area expression stage

The makjang scene analysis module 12 detects the grouping range of the same rock quality in the makjang face TZ of the excavation tunnel to express the dark quality region SZ. As described above, the makjangmyeon analysis module 12 detects the rocks in the makjangmyeon TZ based on the detailed information collected for mapping using the information analyzed by the professional worker in the field and each gun equipment.

Subsequently, the makjang scene analysis module 12 detects a cluster range of the same dark quality among the detected dark materials to define a dark area, that is, a dark area (SZ). The dark zone according to the present invention includes a position value for visually confirming the range of the designated target. Therefore, the scene analysis module 12 may generate a dark area image based on the position value and display it through the output means 20 . For example, as shown in FIG. 15 , the scene analysis module 12 includes a graphic image SZMl of the first window Wl, a live-action composite image SZM2 of the second window W2, and a third window. The attribute value display image SZM3 of (W3) is displayed through the output means 20 .

2nd step, attribute value detection step

The mak scene processing module 13 calculates the dark area image expressed by the mak scene analysis module 12 according to the set value to check the attribute value and process it.

As described above, since the dark area SZ includes a position value, the dark area SZ image can be displayed as a boundary line b1 in the film scene TZ image as shown in Fig. 16(a). . Therefore, as shown in (b) of FIG. 16 , on the boundary line b1 of the dark area (SZ) image, the first point (Pl) of the first height, the second point (PZ) of the second height, and the third point of the third height The point P3 may be identified, and a vertical line c1 may be generated based on the midpoint CP of the image of the scene TZ as shown in FIG. 16( c ).

In addition, the makjang scene processing module 13 may additionally draw various points and lines based on the attribute values of the dark region (SZ) image.

For reference, the first point Pl to the third point P3 may be selected in the order of highest or lowest height, or leftmost or rightmost in the dark matter region SZ image. In this embodiment, the highest In the order of height, the first point Pl to the third point P3 are checked.

Step 3, Baseline Formation Step

The scene processing module 13 forms a reference line SL connecting the second point P2 and the third point P3 of the dark region SZ image as shown in FIG. 16B . Here, the reference line SL is a reference for expressing the correction region SZ' (refer to FIG. 17).

When the reference line SL is formed, the scene processing module 13 is configured to intersect the boundary line b1 with the vertical line c1 at intersections CP1 and CP3 as shown in FIG. You can check the intersection (CP2) that intersects with . However, since the intersection points CP1 to CP3 are selectively used according to the method of generating the correction region, it will be described in detail in the fourth step.

17 is a diagram illustrating a state in which a correction area is expressed in a dark area according to the forward prediction method according to the present invention.

4th step, correction area expression step

The scene processing module 13 has a first height Hl and H3 between the boundary lines BL1 and BL2 passing through the first point Pl to the third point P3 and the reference line SL, and a dark area If the ratio of the second height (TH, H2) of (SZ) is equal to or greater than the threshold value, the dark area (SZ") in the direction of the first point Pl is deleted with the reference line SL as the boundary, and the dark area SZ is The correction area SZ' is expressed, where the threshold value may be 1/4, and in addition, the threshold value may be variously changed according to the scale of the excavated tunnel and the rocky area and other ground conditions.

In the first embodiment, the first height Hl between the boundary line BL1 and the reference line SL passing through the first point Pl to the third point P3 is, as shown in FIG. , is the distance between the intersection points CPI and CPZ where the boundary line BL1 and the reference line SL intersect the vertical line c1, respectively. In addition, the second height HZ of the dark zone SZ is the distance between the intersection points CPI and CP3 where the boundary lines BL1 and BL3 intersect the vertical line c1, respectively. That is, the first height Hl and the second height HZ of the first embodiment are defined based on the vertical line c1.

The second height (HZ) compared to the first height (Hl) confirmed in this way is compared with the threshold value, and if the ratio is equal to or greater than the threshold value, the dark area SZ in the direction of the first point (Pl) with the reference line (SL) as the boundary. ) is considered to be non-uniform and is deleted. Therefore, the correction area SZ' is expressed in the dark area SZ image except for the deleted area SZ". However, if the ratio is less than the threshold value, the entire dark matter region (SZ) image is maintained as it is.

In the second embodiment for expressing the correction region SZ', the first height H3 between the static boundary line BL1 and the reference line SL via the first point Pl to the third point P3 is, As shown in FIG. 5B , it is the height of the region SZ″ surrounded by the boundary lines BL1 and BLZ and the reference line SL.

In addition, the second height TH of the dark area SZ is the height of the dark area SZ image. That is, in the posture of the dark area (SZ) image in the scene (TZ) image, it is the height of the area (SZ") surrounded by the boundary lines BL1 and BLZ and the reference line SL and the dark area SZ image.

The third height H3 compared to the second height TH confirmed in this way is compared with a threshold value, and if the ratio is equal to or greater than the threshold value, the dark area SZ in the direction of the first point Pl with the reference line SL as the boundary ") is regarded as non-uniform and is deleted. Therefore, the correction area SZ' is expressed in the area of the image of the dark area SZ except for the deleted area SZ". However, if the ratio is less than the threshold value, the entire dark matter region (SZ) image is maintained as it is.

18 is an image showing the vertical axis of the correction area expressed according to the forward prediction method according to the present invention, and FIG. 19 is a diagram schematically showing the dark area for prediction through the forward prediction method according to the present invention, FIG. 20 is an image showing an example for predicting the dark area in front of the dark area of FIG. 19 .

Step 5, Prediction of Cancerous Areas

The dark scene prediction module 14 calculates the dark quality region (SZ) image and attribute values with the dark quality region information of the actually measured final scene retrieved from the actual measured scene information DB 11 to predict the dark region of the front scene.

More specifically, the rate of change of the previous correction areas is confirmed in the order of the excavation depth of the makjang face (TZ), and the dark area (SZ) of the front face is predicted. Here, the actual measured scene information DB 11 stores and manages the information on the rock quality of the actual measured scene by excavation depth. The information on the dark area of the actual makjangmyeon includes a longitudinal view of the makjangmyeon (TZ) image along with the frontal view of the makjangmyeon (TZ) image, and the makjangmyeon (TZ) image is classified according to the excavation depth of the tunnel. Therefore, the vertical view is arranged along the excavation depth as shown in FIG. 20 and is graphically displayed as a panoramic image.

The vertical axis will be described in more detail.

As shown in the drawing of FIG. 18(a), the first height Hl is confirmed based on the vertical line c1 of the makjangmyeon TZ image, and only the vertical line c1 section is displayed in the form of a one-dimensional line. However, in the l-dimensional line-shaped makjangmyeon TZ section, the second height H2 section corresponding to the dark zone SZ and the first height Hl corresponding to the area with possibility of deletion (SZ") ) section is set.

The first height (Hl) section wave and the second height (H2) section are set in the form of a one-dimensional line, the makjangmyeon (TZ) section is arranged by excavation depth as shown in the vertical axis view of Fig. 18 and Fig. 20 (b), A vertical view showing the change according to the excavation depth of the rocky region SZ in the Makjang plane TZ is formed.

That is, the vertical axis diagram is created by connecting the high point (HP) and the low point (LP) of the vertical lines, which are in the form of one-dimensional lines, respectively, so as to track the change of the dark matter region (SZ) by image tracking.

On the other hand, in the diagram of FIG. 18(b), if the first height Hl compared to the second height H2 is equal to or greater than the threshold value, only the correction region SZ' is deleted by deleting the region corresponding to the first height Hl. Shown are the TZ images and their vertical axis. As can be seen from the vertical view, the first height Hl as much as the deleted area SZ" is deleted. As a result, the shape of the vertical view of FIG. 18(a) and the vertical axis of the drawing of FIG. 18(b) For reference, in the case of the vertical axis of the drawing of Fig. 18(b), the high point of each vertical line in the one-dimensional line form is to track the change of the correction region SZ' by image tracking. (HP) and trough (LP) are respectively connected to form a ordinate.

As a result, the scene prediction module 14 can track and confirm the change trend of the rocky area SZ or the correction area SZ' by excavation depth by forming a vertical view.

As shown in FIG. 19 , the scene prediction module 14 sets the correction regions SZl and SZ2 in the section of the vertical line c1 in the form of a one-dimensional line through the following embodiment.

First, in the first embodiment, as shown in Fig. 19(a), the high point HP is the average position of the highest point MH1 and the lowest point ML1 in the upper boundary line L1 of the correction regions SZl and SZ2. , the low point LP is the average position of the highest point MH2 and the lowest point ML2 on the lower boundary line L2 of the correction regions SZl and SZ2. The average positions identified in this way are set as the high point (HP) and the low point (LP) at the corresponding positions of the vertical line (c1) section, and are used to form the vertical axis.

Next, in the second embodiment, as shown in the diagram of FIG. 19(b), the upper boundary line L1 and the lower boundary line L2 of the correction region respectively have a high point HP and a low point LP and a vertical line c1 and a vertical line c1. These are the intersections (Cl, C2) that intersect respectively. The average positions identified in this way are set as a high point (HP) and a low point (LP) at the corresponding positions of the section of the vertical line c1, which is a one-dimensional line, and used to form the vertical axis.

Subsequently, the makjangmyeon prediction module 14 checks the rate of change of the previous correction regions in the order of the excavation depth of the makjangmyeon TZ, and predicts the dark area of the front makjangmyeon.

Since the vertical axis view can confirm the change trend of the rocky area according to the excavation depth, the rocky area of the unexcavated front end face can be predicted based on this.

The first embodiment predicts the dark area of the front end face by reflecting the same rate of change as the rate of change of the recent excavation depth end face. Therefore, in the vertical view of the drawing of FIG. 20(b), the first connection line LL1 connecting the high and low points of the recent depth of excavation, respectively, is extended as much as the designed excavation depth using the second connection line LL2. , predict the rocky area of the unexcavated front end face.

The second embodiment calculates the final rate of change by weighting the degree of change of the connecting lines connecting the high and low points of each of the scenes in the vertical view according to the excavation depth.

To explain this in more detail, as shown in the vertical axis of the drawing of FIG. 20(b) , a vertical line (c1) section, which is a one-dimensional line shape corresponding to the makjang surface, is formed according to the excavation depth, and a high point is in the vertical line (c1) section (HP) and trough (LP) are respectively constituted. Also, the high point HP and the low point LP in the vertical line c1 formed in this way lead to the first connection line LL1, respectively. However, a difference occurs in the degree of change of the first connection line LL1 according to the excavation depth. Therefore, the final change rate for prediction of the second connection line LL2 is calculated by changing the weight for each corresponding gradient. The present embodiment increases the weight as it approaches the front end scene, so that the degree of change of the first connection line LL1 in the recent last scene can give a relatively large weight to the rate of change of the second connection line LL2. In this embodiment, the weights were set to 0.4, 0.3, 0.2, and 0.1 in the order of the degree of change close to the front end scene.

After that, when the front makjangmyeon is actually measured through the forward excavation, a composite area is expressed in the new makjangmyeon and stored in the measured makjang scene information DB 11 .

21 is a diagram schematically illustrating a state in which an attribute value of a dark region is checked according to the forward prediction method according to the present invention.

The first step of detecting the cluster range of the same dark quality in the first scene TZ1 of the excavation tunnel to express the first dark area SZ1 and setting the first central point 21 of the first dark area SZ1 ; In the second film scene (SZ2), which is the last scene of the excavation depth next to the first film scene (TZ1), a cluster range of the same dark material is detected to express the second dark mass region (SZZ), and the second dark mass region (SZZ) is a second step of setting a second central point (22); a third step of determining that the first dark quality region SZ1 and the second dark quality region SZ2 are homogeneous regions when the distance between the first central point 21 and the second central point 22 is less than a threshold value; A fourth step of confirming the change rate of the previous homogeneous areas in the order of the excavation depth of the Makjangmyeon, and predicting the dark area of the front Makjangmyeon; includes

Each step is described in more detail.

Step 1, first dark matter region attribute value detection step

The mak scene analysis module 12 detects the cluster range of the same dark quality in the first mak scene TZ1 of the excavation tunnel to express the first dark mass region SZ1, and the mak scene processing module 13 is the first dark mass region A first central point 21 of (SZ1) is set.

Since the detailed description in which the mak scene analysis module 12 detects the cluster range of the same rock quality in the first mak scene TZ1 of the excavation tunnel and displays the first murky region SZ1 has been described above, the description thereof will be omitted here. do.

When the first scene TZ1 and the first dark region SZ1 are displayed, the scene processing module 13 converts the image of the first dark region SZ1 expressed by the scene analysis module 12 according to the set value. The operation is performed to check and process the first central point 21, which is an attribute value. Since the first dark area SZ1 includes a position value, the image of the first dark area SZ1 is displayed as the first boundary line BL21 in the first scene TZ1 image as shown in FIG. 21(a). can do.

In addition, the makjang surface processing module 13 may additionally draw various points and lines based on the attribute values of the first dark region SZ1 image. In the present invention, a first vertical line CL1 (refer to FIG. 22 ) passing through the first central point 21 of the first dark matter region SZ1 may be generated.

2nd step, 2nd dark matter region attribute value detection step

The mak scene analysis module 12 detects the cluster range of the same dark quality in the second mak scene TZ2 of the excavation tunnel to express the second dark quality region SZ2, and the mak scene processing module 13 is the second dark quality region A second central point 22 of (SZ2) is set.

Since the detailed description in which the mak scene analysis module 12 detects the cluster range of the same rock quality in the second mak scene TZ2 of the excavation tunnel and displays the second murky region SZ2 has been described above, the description thereof will be omitted here. do.

When the second scene TZ2 and the second dark region SZ2 are displayed, the scene processing module 13 converts the image of the second dark region SZ2 expressed by the scene analysis module 12 according to the set value. The operation is performed to check and process the second central point 22, which is an attribute value. Since the second dark area SZ2 includes a position value, the image of the second dark area SZ2 is displayed as a second boundary line BL22 in the image of the second scene TZ2 as shown in FIG. 21(b). can do.

In addition, the makjang scene processing module 13 may additionally draw various points and lines based on the attribute values of the dark region (SZ) image. In the present invention, a second vertical line CL2 (refer to FIG. 22 ) passing through the second central point 22 of the second dark matter region SZ2 may be generated.

In the same manner as in the process of the first and second steps, the final scene analysis module 12 detects the grouping range of the same rock quality in the third and fourth scenes (TZ3, TZ4) of the excavation tunnel, as shown in Fig. 21(c). ) and (d), the subsequent dark regions (SZ31, SZ32, SZ33, SZ41, SZ42) are expressed. In addition, the scene processing module 13 sets the center points 231 to 233, 241, and 242 of the dark regions SZ31, SZ32, SZ33, SZ41, and SZ42, respectively.

Step 3, confirmation of homogeneity of dark areas

The scene processing module 13 measures the distance dt between the first central point 21 and the second central point 22 (see FIG. 11 ), and if the distance dt is less than a threshold value, the first dark zone SZ1 ) and the second dark region SZ2 are set to be homogeneous regions. Here, the threshold value may be variously changed according to the scale of the excavated tunnel and the rocky area and other ground conditions. The threshold value of this embodiment is 1/4 of the difference between the maximum and minimum values of the distances between two dark areas to be compared. That is, the maximum and minimum values are checked by checking the distances between the first and second dark areas SZ1 and SZ2, and 1/4 of the difference between the confirmed maximum and minimum values is the threshold value. . Therefore, if the maximum value of the distance between the first dark matter region (SZ1) and the second dark matter region (SZ2) is 12 and the minimum value is 4, '2', which is 1/4 of the difference, is the first dark matter region (SZ1) and It serves as a criterion for determining whether the second dark region SZ2 is a homogeneous region.

On the other hand, the makjang surface processing module 13 is the center point 231 to 233 of each of the subsequent dark areas (SZ31, SZ32, SZ33) and the second dark area (SZ2) in the drawing of FIG. ) to determine whether the distance between them is less than the threshold value. In addition, in the diagram of (d) of FIG. 21 , the distance between the center points 241 and 242 of each of the subsequent dark areas SZ41 and SZ42 and the center points 231 to 233 of each of the dark areas SZ31, SZ32, SZ33 is measured and the threshold Check whether the value is less than or not.

As described above, the makjangmyeon processing module 13 checks whether there is homogeneity between the dark areas in the makjangmyeon (TZl to TZ4) according to the excavation depth, and proceeds with the subsequent procedure.

22 to 24 are diagrams schematically showing a state in which a vertical line of a rocky area is displayed according to the forward prediction method according to the present invention, and FIG. It is an image showing the vertical axis.

Step 4, Prediction of Cancerous Areas

The last scene prediction module 13 checks the rate of change of the previous homogeneous regions in the order of the excavation depth of the last scenes (TZl to TZ4) and predicts the dark regions of the front scenes.

To this end, the scene processing module 12 searches for the previous dark regions (SZl, SZ2, SZ31, SZ32, SZ33, SZ41, SZ42) in the actual scene scene information DB 11, and the dark regions identified as homogeneous regions ( Check the high and low points of SZl, SZ2, SZ31, SZ32, SZ33, SZ41, SZ42) respectively.

First, the vertical line CL1 and the high points a1 and b1 and the low points a2 and b2 will be described with reference to FIG. 22 .

The mak scene processing module 12 is first detected in the image of the first dark area (SZ1) because the excavation depth is relatively small among the first dark area (SZ1) and the second dark area (SZ2), which are confirmed to be homogeneous with each other. The first vertical line CL1 passing through the first central point 21 of the image of the first dark matter region SZ1 is displayed. At this time, among the intersections between the boundary line BL21 of the first dark zone SZ1 and the first vertical line CL1, a point at a high position is called a high point a1, and a point at a low position is called a low point a2.

On the other hand, since the second dark area SZ2 is the same as the first dark area SZ1, the mak scene processing module 12 reflects the first vertical line CL1 to the second dark area SZ2 as it is, and the second An intersection point between the boundary line BL21 of the dark region SZ2 and the first vertical line CL1 is identified. Among the intersection points, a high point is called a high point b1, and a low point is called a low point b2.

Referring to FIG. 23 , the first to third vertical lines CL1 , CL2 , CL3 , the high points b1 , c1 , d1 , and e1 , and the low points b2 , c2 , d2 and e2 are respectively described.

Since it is confirmed that the 31st dark region SZ31 and the second dark region SZ2 are homogeneous with each other, the Makscene processing module 12 reflects the first vertical line CL1 to the 31st dark region SZ31 as it is. An intersection point between the boundary line BL231 and the first vertical line CL1 of the 31st dark region SZ31 is checked. Among the intersection points, a high point is called a high point c1, and a low point is called a low point c2.

On the other hand, the makjang scene processing module 12 confirms that the 31st dark area (SZ31) and the 32nd dark area (SZ32) are of different dark quality, so that the 32nd dark area (SZ32) passes through the center point 232 of the image. A second vertical line CL2 is indicated. At this time, among the intersections between the boundary line BL232 and the second vertical line CL2 of the 32nd dark zone SZ32, a point with a high position is called a high point d1, and a point with a low position is called a low point d2.

In addition, in the makjang scene processing module 12, it is confirmed that the 31st dark region SZ31, the 32nd dark region SZ32, and the 33rd dark region SZ33 are different from each other, so that the 33rd dark region SZ33 A third vertical line CL3 passing through the center point 233 of the image is displayed. At this time, among the intersections between the boundary line BL233 of the 33rd dark region SZ33 and the third vertical line CL3, a point at a high position is called a high point e1, and a point at a low position is called a low point e2.

The second to third vertical lines CL2 and CL3, the high points d1, e1, f1, and g1, and the low points d1, e1, f1, and g1 will be described with reference to FIG. 13, respectively.

Since it is confirmed that the 32nd dark area SZ32 is the same as the 41st dark area SZ41, the Makscene processing module 12 reflects the second vertical line CL2 to the 41st dark area SZ41 as it is. Thus, an intersection point between the boundary line BL241 of the 41st dark region SZ31 and the second vertical line CL2 is checked. Among the intersection points, a point with a high position is called a high point d1, and a point with a low position is called a low point d2.

Subsequently, the makjang scene processing module 12 confirms that the 33rd dark region SZ33 and the 42nd dark region SZ42 are homogeneous with each other, so the third vertical line CL3 is drawn to the 42nd dark region SZ42. By reflecting as it is, the intersection point between the static boundary line BL242 and the third vertical line CL3 of the 42nd dark zone SZ42 is checked. Among the intersection points, a point at a high position is called a high point (e1), and a point at a low position is called a low point (e2).

As described above, when both the high and low points are confirmed for each dark area, the mak scene prediction module 14 arranges the first vertical line CL1 and the second vertical line CL2 in the order of the excavation depth of the excavation tunnel to form the first dark area. Changes in (SZ1) and the second dark matter region (SZ2) are tracked by image tracking. To this end, as shown in FIG. 25 , the scene prediction module 14 performs the high points (a1, b1, c1, d1, e1, f1, g1) and low points of the homogeneous region for each of the first and second vertical lines (CL1, CL2). (a2, b2, c2, d2, e2, f2, g2) are respectively connected to create a ordinate.

The vertical axis of the present embodiment will be described in more detail.

As shown in Fig. 22, the scene prediction module 14 is located on the first vertical line CL1 passing through the first central point 21 of the first dark region SZ1, the high point of the first dark region SZ1 ( a1) and the low point a2 are marked, and the high point b1 and the low point ( b2) is indicated. Subsequently, the scene prediction module 14 arranges the first vertical line CL1 and the second vertical line Cl2 in the order of the excavation depth of the excavation tunnel, as shown in FIG. 25 , and the first and second vertical lines ( For each CL1 and CL2, a vertical view is generated by connecting the high points al and bl and the low points a2 and b2 of the homogeneous region to the first connecting line LL1, respectively. However, in the case where different vertical lines, such as the second vertical line CL2 and the third vertical line CL3, are configured on the same makjang plane and are vertical lines of a heterogeneous dark region, the 'TZ3' and 'TZ4' makjangmyeon sections of FIG. As shown, the second vertical line CL2 and the third vertical line CL3 may be disposed to overlap each other.

The structure of the vertical axis view shown in FIG. 25 will be further described.

In the Makjangmyeon section of 'TZl', one type of rocky region passing through the first vertical line CL1 is identified. Accordingly, it can be confirmed that there is one high point a1 and one low point a2 through which the first vertical line CL1 passes.

Also, in the section of 'TZ2' Makjangmyeon, one type of dark region passing through the first vertical line CL1 is identified. Accordingly, it can be confirmed that there is one high point b1 and one low point b2 through which the first vertical line CL1 passes. At this time, since the vertical line of the 'TZ2' Makjangmyeon section is the same first vertical line (CL1) as the vertical line of the 'TZl' Makjangmyeon section, the dark area of the 'TZl' Makjangmyeon section and the dark area of the 'TZ2' Makjangmyeon section are the same. to be.

However, in the Makjangmyeon section of 'TZ3', three types of dark regions are identified that pass through the first vertical line CL1, the second vertical line CL2, and the third vertical line CL3, respectively. Accordingly, it can be confirmed that there is one high point c1, d1, e1 and one low point c2, d2, e2 through which the first to third vertical lines CL1 to CL3 respectively pass. At this time, since the vertical line of the 'TZ3' Makjangmyeon section is the same first vertical line (CL1) as the vertical wave of the 'TZ2' Makjangmyeon section, some of the dark areas of the 'TZ3' Makjangmyeon section are the same as the dark areas of the 'TZ2' Makjangmyeon section. is homogeneous Meanwhile, in the 'TZ3' Makjangmyeon section, two more new dark areas are identified.

In addition, two types of dark areas are identified in the 'TZ4' Makjangmyeon Gukan passing through the second vertical line CL2 and the third vertical line CL3, respectively. Accordingly, it can be confirmed that the high points f1 and g1 and the low points f2 and g2 through which the second to third vertical lines CL2 to CL3 pass are one each. At this time, since the vertical line of the 'TZ4' makjangmyeon section is the same second vertical line (CL2) and the third vertical line (CL3) as the vertical wave of the 'TZ3' makjangmyeon section, the dark area of the 'TZ4' makjangmyeon section and the 'TZ3' makjang A part of the rocky area of the Myon section is homogeneous.

The vertical axis diagram of this embodiment described above specifies the distribution of dark areas of the mak scenes (TZl to TZ4) for each depth of excavation, and also shows the change trend of the dark regions of the mak scenes (TZl to TZ4). Therefore, based on the change trend, it is possible to predict the type, shape, distribution, and the like of the rocky area constituted in the front makjang surface.

The scene prediction module 14 connects the high and low points of the homogeneous dark region for each dark scene (TZl to TZ4) of the change trend specified in the vertical diagram, respectively, with the first connection line LL1, followed by The degree of change of the second connection line LL2 to be connected is predicted.

The first embodiment predicts the rocky area of the front end face by reflecting the same rate of change as the rate of change wave of the recent excavation depth face. Accordingly, in the vertical view of FIG. 25, the first connection line LL1 connecting the high and low points of the recent excavation depth of the recent excavation surface is extended as much as the designed excavation depth using the second connection line LL2, and the unexcavated front Predict the dark area of the makjangmyeon.

The second embodiment calculates the final rate of change by weighting the degree of change of the connecting lines connecting the high and low points of each of the scenes in the vertical view according to the excavation depth.

To explain this in more detail, as shown in the vertical axis view of FIG. 25 , a section of first to fourth vertical lines CL1 to CL4 that is a one-dimensional line corresponding to the makjang surface is formed according to the depth of excavation, and the first to fourth vertical lines CL1 are formed. to CL4), the high points a1, b1, c1, d1, e1, f1, g1 and the low points a2, b2, c2, d2, e2, f2, g2 are respectively configured. In addition, the high points a1, b1, c1, d1, e1, f1, g1 and the low points a2, b2, c2, d2, e2, f2, g2 within the first to fourth vertical lines CL1 to CL4 formed in this way are respectively leads to the connecting line. However, depending on the depth of excavation, there is a difference in the degree of change of the connecting line. Therefore, the final change rate for prediction of the second connection line LL2 is calculated by changing the weight for each corresponding gradient. The present embodiment increases the weight as it approaches the front end scene, so that the degree of change of the first connection line LL1 in the recent last scene can give a relatively large weight to the rate of change of the second connection line LL2. In this embodiment, the weights were set to 0.4, 0.3, 0.2, and 0.1 in the order of the degree of change close to the front end scene.

Thereafter, when the front makjangmyeon is actually measured through the forward excavation, the dark area is expressed in the new makjangmyeon and stored in the measured makjang scene information DB 11 .

27 is a diagram schematically illustrating a state in which an attribute value of a dark region is checked according to the forward prediction method according to the present invention.

A first step of detecting a grouping range of the same rock quality in the makjang surface of the excavation tunnel, expressing the dark zone (SZ), and confirming whether it is the same as the dark zone of the previous excavation depth makjang surface; a second step of dividing the dark zone (SZ) into multiple equal parts at regular intervals to express the divided zone; a third step of extracting one end point and the other end point of the divided region, respectively; Check the rate of change of each end point (SP) and other end point (FP) for each division (DZ) of the previous rocky areas of the same homogeneity in the order of the excavation depth of the Makjangmyeon, and It includes a fourth step of predicting the rocky area of the front makjang surface by predicting the other end point.

Each step is described in more detail.

1st stage, dark area expression stage

The mak scene analysis module 12 detects the grouping range of the same rock quality on the makjang face of the excavation tunnel and displays the dark zone SZ. As described above, the makjangmyeon analysis module 12 detects the rocks in the makjangmyeon TZ based on the detailed information collected for mapping by using the information analyzed by the professional worker in the field and each gun equipment. In addition, the scene analysis module 12 sets the center point 21 of the dark area (SZ).

Subsequently, the makjang scene analysis module 12 detects a cluster range of the same dark quality among the detected dark materials to define a dark area, that is, a dark area (SZ). The dark zone (SZ) according to the present invention includes a position value for visually confirming the range of the designated target.

Therefore, the scene analysis module 12 may generate a dark area image based on the position value and display it through the output means 20 .

To explain this in more detail, when the makjangmyeon (TZ) and the dark region (SZ) are expressed, the makjangmyeon processing module 13 sets the dark region (SZ) expressed by the makjangmyeon analysis module 12 to the position value. According to the calculation, the center point 21 and the boundary line b1, which are attribute values of the dark matter region SZ image, are confirmed. Based on the thus-confirmed central point 21 and boundary line b1, the dark area SZ is displayed as a dark area SZ image in the makjangmyeon TZ image as shown in FIG. 27(a).

On the other hand, the shape of the boundary line b1 of the rocky area SZ changes according to the change of the rocky area for each depth of excavation. Therefore, it is possible to check the change behavior of the image of the dark matter region SZ by checking the change in each section of the boundary line b1. In addition, the position of the center point 21 of the rocky region SZ also changes according to the change of the rocky region for each depth of excavation. Therefore, it is possible to check the position change and change direction of the central point 21 of the rocky region for each depth of excavation.

In addition, the makjang scene processing module 13 may additionally draw various points and lines based on the attribute values of the dark region (SZ) image.

When the makjang scene analysis module 12 expresses the dark region SZ in the makjangmyeon TZ, the makjang scene processing module 13 determines whether it is the same as the dark region configured in the makjang scene of the previous excavation steam. The homogeneity of the dark areas constructed on the final surface of different excavation depths is determined by detecting the center points of the dark areas to be compared, and if the distance between the center points of the previous dark areas is less than a threshold value, the previous dark areas are considered to be homogeneous with each other. This is done by setting

Here, the threshold value may be variously changed according to the scale of the excavated tunnel and the rocky area and other ground conditions. The threshold value of this embodiment is 1/4 of the difference between the highest and lowest values of the distances between two dark areas to be compared. That is, the maximum and minimum values are confirmed by checking the distances between the first and second dark areas, and a value of 1/4 of the difference between the confirmed maximum and minimum values is the threshold value. Therefore, if the maximum value of the distance between the first dark matter region and the second dark matter region is 12 and the minimum value is 4, '2', which is 1/4 of the difference, is the homogeneous region between the first dark matter region and the second dark matter region. as a criterion for judging whether or not

2nd step, division area expression step

The makjang scene processing module 13 divides the dark area SZ into multiple equal parts at regular intervals as shown in FIG. 27(b) to express the divided area.

As described above, when the scene analysis module 12 detects the dark area (SZ) in the scene (TZ) and images the dark area (SZ), the scene processing module 13 designates the image of the dark area (SZ) By dividing by intervals, independent images are vivid. At this time, the division of the dark area (SZ) image may be varied in the horizontal direction or the longitudinal direction, but in this embodiment, the dark area (SZ) image is divided in the horizontal direction.

The scene processing module 13 flexibly sets the number of divisions according to the area of the dark area SZ itself or the length in the division direction, as well as the area of the dark area SZ image or the length in the division direction. In general, the scene processing module 13 divides the dark area (SZ) image into 10 divisions or 20 divisions.

Through the division of the dark area (SZ) image, the mak scene prediction module 14 can grasp the change of the dark area (SZ) image for each excavation depth by section, not by the entire dark area (SZ) image, and the section unit Changes for each section can be predicted according to the change data identified by Through this, the makjang scene prediction module 14 can precisely predict the rocky area for each point.

Step 3, check one/other point for each division

The scene processing module 13 checks one end point SP and the other end point FP of the divided area DZ generated by dividing the dark matter area SZ image. Here, the one end point SP and the other end point FP are edge points of the segmented region DZ, and thus are a point of the boundary line b1 of the dark matter region SZ image.

In this embodiment, since the scene processing module 13 divides the dark area SZ image in the horizontal direction, one end SP and the other end point FP of the segmented area DZ are the highest points of the segmented area DZ, respectively. and the lowest point.

27 is a view showing a change in the position of one end point of the previous divided regions according to the forward prediction method according to the present invention.

Step 4, Prediction of Cancerous Areas

The final scene prediction module 14 confirms the rate of change of one end point and the other end point for each divided region of the previous rocky regions of the same quality in the order of the excavation depth of the final scene, and the Predict the other end point to predict the rocky area of the front makjang surface.

27 shows a part of the dark areas (SZ10 to (SZ40) confirmed to be identical to each other in the previous film scenes. b) The figure is an image of the dark area (SZ30) of the third full depth, (c) is an image of the dark area (SZ20) of the second full depth, (d) is an image of the dark area (SZ10) of the previous depth to be.

When explaining with reference to the third divided regions DZ31 to DZ34 in the divided regions of the images of the dark regions SZ10 to SZ40, one point of the third divided regions DZ31 to DZ34 of the images of the dark regions SZ10 to SZ40 (SP1 to SP4) have coordinate values of (x1, y1) to (x4, y4), respectively. The scene prediction module 14 checks the point change of one end point (SP1 to SP4) of the corresponding divided area (DZ31 to DZ34) of the dark area (SZ10 to SZ40) image and grasps the degree of change, and thus the change rate Based on this, one end point of the corresponding division of the front curtain is predicted.

The prediction process of the scene prediction module 14 will be described as an example.

The first embodiment predicts the rocky area of the front makjang face by reflecting the same rate of change as the rate of change wave of the makjang face for each depth of recent excavation. Therefore, the scene prediction module 14 confirms the degree of change of the corresponding divided regions DZ31 and DZ32 in the image of the dark region (DZ10) of the previous depth and the image of the dark region (SZ20) of the second full depth, and the According to the degree of change, the position of the end point of the divided area is predicted within the dark area of the front end of the scene. Here, the prediction of the location of one end point for each divided area (DZ31 to DZ34) of the scene prediction module 14 is the change of the mid island point 21 and the boundary line b1 of the corresponding dark area SZ along with the degree of change of the one end point. This is done taking into account the direction and behavior of change.

In the second embodiment, the final change rate is calculated by weighting the change in the position of the one end of the divided area according to the excavation depth. To explain this in more detail, the scene prediction module 14 changes the corresponding divided regions (DZ33, DZ34) in the fourth full-depth dark region (SZ40) image and the third full-depth dark region (SZ30) image. check In addition, the scene prediction module 14 confirms the degree of change of the corresponding divided regions (DZ32, DZ33) in the image of the dark region (SZ30) of the third previous Kimi and the image of the dark region (SZ20) of the second full depth. In addition, the scene prediction module 14 confirms the degree of change of the corresponding divided area DZ31 DZ32 in the second full depth dark area SZ20 image and the immediately previous dark dark area SZ10 image.

As a result, a difference occurs in the degree of change of the separate points of one end (SP1 to SP4) of the corresponding divided areas (DZ31 to DZ34) according to the excavation depth. Therefore, the final change rate is calculated by changing the weight for each corresponding gradient, and one end point of the corresponding divided region of the dark area of the front end face is predicted according to the final change rate. In this embodiment, the weight is increased as it approaches the front end scene, so that the end scene prediction module 14 is divided into the second full depth dark area (SZ20) image and the dark mass area (SZ10) image of the immediately previous depth ( DZ31, DZ32) allow a relatively large weight to be given to the rate of change of the one-point position of the front segment. In this embodiment, the weights were set to 0.4, 0.3, 0.2, and 0.1 in the order of the degree of change close to the front end scene.

Here, the prediction of the position of one end point for each divided area (DZ31 to DZ34) of the scene prediction module 14 is, as in the first embodiment, the center point 21 and boundary line of the corresponding dark area SZ along with the degree of change of one end point. It is made in consideration of the change direction and form of change in (b1).

Thereafter, when the front makjangmyeon is actually measured through the forward excavation, the dark area is expressed in the new makjangmyeon and stored in the measured makjang scene information DB 11 .

The process of predicting the final scene using an automatic final scene analysis system and prediction system

With reference to FIG. 1, the process of predicting the front scene by combining the automatic analysis system and the prediction system using the aforementioned artificial intelligence will be described. Since the description of the configuration of the systems has been described above, the overlapping description will be omitted.

The output unit 350 of the DFM module 300 may output the DFM analysis information as a result, and the result may be stored in the DFM database 390 . In addition, by being transmitted to the learning information addition module 500 via the user correction module 400, it is possible to go through a flow that is used again as learning information.

As a separate flow, the DFM analysis information output by the output unit 350 is transmitted to the prediction system 10, so that the DFM analysis information can be used to predict the front end scene of the tunnel.

The analysis information obtained from the DFM module 300 may undergo a process of being converted into 'prediction system input scene information' for input to the prediction system 10 . To this end, a separate transformation module (not shown) may be provided, and the transformation processing may be performed in the scene analysis module 12 included in the prediction system 10 . Herein, the description will be based on the conversion processing performed in the scene analysis module 12 .

The scene analysis module 12 re-analyzes the analysis information obtained from the DFM module 300 to define a dark area. In this case, the analysis information obtained from the DFM module 300 may be used for the information on the ground of the Makjangmyeon. Accordingly, there is an advantage that the process of collecting and analyzing the ground information by a professional worker can be omitted in the related art.

The analysis information transmitted from the DFM module 300 to the scene analysis module 12 passes through the scene analysis module 12 as described above, and then the scene processing module 13 of the prediction system 10 described above. And it is configured to predict the dark area of the front scene through the scene prediction module 14 sequentially.

That is, the above-described prediction system 10 is a configuration that utilizes the information on the rock quality of the final scene measured by the depth of the excavation tunnel stored in the actual scene information DB 11, whereas the 'automatic scene analysis system and the prediction system are used. In the 'prediction process of the front scene used,' there is a difference in that the scene analysis information output from the output unit 350 of the DFM module 300 is used instead of the actual scene information DB 11 . Since the scene analysis information output from the output unit 350 of the DFM module 300 is processed information, a lot of noise has been removed compared to the actual measurement information, so the calculation time in the scene analysis module 12 is reduced. There are advantages to doing it.

The scene analysis information output from the output unit 350 of the DFM module 300 is output to the prediction information output unit 20 through the prediction system 10 . The prediction information output unit 20 may be displayed through a display means. The prediction information transmitted to the prediction information output unit 20 may be configured to further go through the prediction information correction unit 30 . This is because the prediction system 10 is configured to predict the front end surface for each preset excavation depth, and by further reflecting the information actually confirmed through excavation, the accuracy and reliability of the analysis information can be improved. The prediction information correction unit 30 is preferably performed by an expert, and can be corrected based on information actually confirmed through excavation, or a portion determined to be an error can be corrected. Through this modification process, the prediction system 10 and the automatic scene analysis system applied to the present invention can be further advanced.

In the above, the present specification has been described with reference to the embodiments shown in the drawings so that those skilled in the art can easily understand and reproduce the present invention, but these are merely exemplary, and those skilled in the art can make various modifications and equivalent other modifications from the embodiments of the present invention. It will be appreciated that embodiments are possible. Therefore, the protection scope of the present invention should be defined by the claims.

10: Prediction system
11: Actual scene information DB
12: Scene analysis module
13: Makjangmyeon processing module
14: scene prediction module
100: learning information generation module
110: learning scene image information input unit
120: analysis information input unit for learning
130: learning information division unit
190: learning information database
200: learning module
210: format conversion unit
220: learning parameter setting unit
230: learning performance unit
240: validation unit
300: DFM module
310: input unit
320: image correction unit
330: reasoning unit
340: joint extraction unit
350: output unit
360: RQD calculator
390: DFM database
400: user modification module
500: learning information addition module

Claims (14)

Through an automatic mak scene analysis system using learning that uses the mak scene image information as an input layer and digital face mapping (DFM) analysis information to the mak scene image information as an output layer, the input mak scene image information is As a method of performing an analysis for -here, the analysis information includes information about the carcinoma, information about the cancer quality, and information about the joint-
(A-1) The learning scene image information input unit 110 loads the learning scene image information prepared in advance for learning, and the learning analysis information input unit 120 loads the learning analysis information corresponding to the closing scene image information, and , setting the learning information by mapping the loaded scene image information for learning and the analysis information for learning (S110); and
(B) by repeatedly performing the step (A-1) for each of the plurality of pre-prepared scenes for learning, constructing an automatic analysis system for scenes (S200); includes,
The learning information set in the (A-1) step (S110) is,
It is generated by detecting the cluster range of the same target in the scene and expressing the target area, and the target includes carcinoma, cancer, and joint,
(T1) Expressing the correction region of the target region using the reference line connecting the specific points included in the target region and the height of each specific point, and using the rate of change of previous correction regions in the order of the excavation depth of the final scene a first manner in which it is generated;
(T2) a second method in which the first and second target areas are expressed in the order of the excavation depth of the makjang surface, and the first and second target areas are generated using a center point distance; or
(T3) a third method in which the target region is divided into multiple equal parts to express a divided region, and is generated using a change rate of a specific point included in the divided region in the order of the excavation depth of the final scene; using any of the
Way.
The method of claim 1,
The first method is
(T-11) a first point P1 of the first height corresponding to the highest height on the boundary line b1 of the target area SZ, and a second point of a second height corresponding to a height lower than the first height ( After confirming P2) and a third point P3 of a third height corresponding to a height lower than the second height, a reference line SL connecting the second point P2 and the third point P3 is formed, step (S11);
(T-12) a first height H1 between the boundary line b1 passing through all of the first to third points P1, P2, and P3 and the reference line SL, and a second height of the target area When the ratio of the height H2 is equal to or greater than the threshold value, by deleting the target range in the direction of the first point P1 with respect to the reference line SL as a boundary, the correction area of the target area is expressed (S12); and
(T-13) Checking the rate of change of the previous correction areas in the order of the excavation depth of the final scene, and predicting the image information and analysis information of the final scene in the front using the rate of change, generating the learning information (S13) ; containing,
Way.
3. The method of claim 2,
The (T-13) step is,
A vertical line passing through the midpoint of each of the final scenes is arranged in the order of the excavation depth, and the high and low points of the correction area are indicated on the vertical lines, wherein the high and low points are the upper boundary line and the lower boundary line of the correction area. It is an intersection point that intersects each of the baselines,
generating a vertical view by connecting the high and low points of each vertical line to track the change in the correction area by image tracking,
Way.
3. The method of claim 2,
In the step (T-12),
The first height is a length between intersections where a vertical line passing through the midpoint of the makjang surface intersects a boundary line and a reference line, respectively,
The second height is the length between intersection points where the vertical line intersects the boundary line of the target area,
Way.
The method of claim 1,
The second method is
(T-21) setting the first and second target regions as homogeneous regions when the distance between the first central point of the first target region and the second central point of the second target region is less than a threshold value (S21); and
(T-22) confirming the rate of change of the previous homogeneous areas in the order of the excavation depth of the last scene, and predicting the target area of the front end face by using the change rate (S22); containing,
Way.
6. The method of claim 5,
The step (T-22) is,
Displaying the high and low points of the first target region on a first vertical line passing through the first central point, and displaying the high and low points of the second target region on a second vertical line passing through the second central point,
In the order of the depth of excavation, the first and second vertical lines are arranged to track the change of the first and second target areas by image tracking, and the first and second vertical lines respectively connect the high and low points of the homogeneous area to the vertical axis creating a diagram,
Way.
The method of claim 1,
The third method is
(T-31) extracting one end point and the other end point of the divided region, respectively (S31); and
(T-32) In the order of excavation depth, the rate of change of one end point and the other end point for each division of the previous target areas of the same homogeneity is confirmed, and the one end point and the other end point for each division area of the target area of the front end scene are predicted and the front Predicting the target area of the scene (S32); containing,
Way.
8. The method of claim 7,
The division area divides the target area in a horizontal direction,
In the step (T-32),
The one end point and the other end point are the highest and lowest points of each division,
The rate of change of each of the one end point and the other end point includes the rate of change of the distance between the one end point and the other end point,
Way.
8. The method of claim 7,
(T-30) Before expressing the divided region by dividing the target region into multiple equal parts,
Detecting the center points of the previous target areas in the order of the excavation depth, and if the distance between the center points of the previous target areas is less than a threshold value, setting the previous target areas to be homogeneous with each other (S30); further comprising,
Way.
10. The method according to any one of claims 1 to 9,
After step (A-1),
(A-2) the learning information dividing unit 130, generating a plurality of divided learning information for each piece of learning information by cropping the learning information as a preset unit (S120);
(A-3) converting the format conversion unit 210 into a learning format by loading the plurality of divided learning information (S130); and
(A-4) performing, by the learning performing unit 230, learning multiple times for each single scene image information using the format-converted plurality of divided learning information (S140); further comprising,
After step (B),
(C) When the scene image information is input to the input unit 310, the inference unit 330 infers the analysis information using the automatic analysis system for the scene scene, and the output unit 350 outputs it (S300) ; further comprising,
Way.
11. The method of claim 10,
After step (B),
(B-1) the validation unit 240, using a plurality of different learning scenes image information to perform the validity verification of the automatic scene analysis system built in the step (B) (S210); further comprising,
N pieces of image information for the learning scene and analysis information corresponding thereto are prepared,
The image information for the learning scenes performed in the step (B) is A (N>A), and the other image information for the curtain scenes for learning performed in the step (B-1) is NA individuals,
Way.
11. The method of claim 10,
The step (C) is,
(C-11) step of inputting the scene image information to the input unit 310 (S311);
(C-12) A step in which a tunnel boundary node is input on the inputted scene image information to the image correction unit 320, the input tunnel boundary node is connected, and only the connected area is extracted, thereby modifying the closing scene image information (S312); and
(C-13) the inference unit 330 infers the analysis information using the modified automatic scene analysis system, and the output unit 350 outputs it (S313); containing,
Way.
11. The method of claim 10,
The step (C) is,
(C-21) step of inputting the scene image information to the input unit 310 (S321);
(C-22) the image correction unit 320 dividing the input scene image information into a plurality of regions using a watershed technique (S322);
(C-23) Inferring, by the inference unit 330, analysis information for each region through the automatic scene analysis system by using each of a plurality of regions of the input scene image information as an input layer (S323) ; and
(C-24) The inference unit 330 infers the digital face-mapped analysis information for the input scene image information by integrating the analysis information for each area, and the output unit 350 outputs it step (S324); containing,
Way.
Using the method according to claim 10, as a method of predicting the front end scene through the prediction system (10),
After step (C),
(C-1) the step of inputting the analysis information of the scene inferred by the inference unit 330 and output from the output unit 350 to the prediction system 10 (S410);
(C-2) generating a target area image by inputting the analysis information of the scene scene to the scene scene analysis module 12 provided in the prediction system 10 (S420);
(C-3) processing, by the scene processing module 13 provided in the prediction system 10, the target region image expressed by the scene analysis module 12 in a preset manner (S430); and
(C-4) Through the curtain scene prediction module 14 provided in the prediction system 10, the front curtain scene is predicted using the image processed by the curtain scene processing module 13, but the first Predicting the front curtain surface using any one of the to third method (S440); further comprising,
A method for predicting the front end scene.

Images