CN116630352A

CN116630352A - Rock core measurement method and device based on bidirectional cascade pixel differential network

Info

Publication number: CN116630352A
Application number: CN202310655780.2A
Authority: CN
Inventors: 吴文杰; 金和平; 罗惠恒; 李德龙; 王景晗; 许艳丽; 周超辉; 刘育策; 张晓萌; 姜鹏
Original assignee: China Three Gorges Corp
Current assignee: China Three Gorges Corp
Priority date: 2023-06-02
Filing date: 2023-06-02
Publication date: 2023-08-22

Abstract

The invention relates to the technical field of core measurement, and discloses a core measurement method and device based on a bidirectional cascade pixel differential network, wherein the method comprises the following steps: acquiring a plurality of rock core images, training an edge detection model based on the plurality of rock core images, and generating a bidirectional cascade pixel differential network model; acquiring a core box main body image, inputting the core box main body image into a bidirectional cascade pixel differential network model, and generating an initial edge prediction image; processing the initial edge prediction graph to generate a core rectangular frame; and carrying out core measurement on the core box main body diagram based on the core rectangular frame to generate core characteristics. The method improves the accuracy of the core measurement result and is beneficial to the development of subsequent exploration geological analysis work.

Description

Rock core measurement method and device based on bidirectional cascade pixel differential network

Technical Field

The invention relates to the technical field of core measurement, in particular to a core measurement method and device based on a bidirectional cascade pixel differential network.

Background

Large and medium-sized hydraulic and hydroelectric engineering is mostly positioned in mountain canyons with complex geological conditions, fault structures develop, and the integrity of rock masses in different areas is obviously different. Accurate and objective judgment of geological conditions and rock integrity is an important precondition for design and construction of hydraulic and hydroelectric engineering.

Because field exploration working conditions are complex, the risk is high, and interference of a plurality of environmental factors exists, core measurement is inconvenient in the field, the whole measurement process has the problems of strong subjectivity, long period, high risk degree and the like, and because different people have subjective differences in color identification, color recording may be inaccurate, and the development of subsequent exploration geological analysis work is affected.

Disclosure of Invention

In view of the above, the invention provides a core measurement method and device based on a bidirectional cascade pixel differential network, which are used for solving the problems that the accuracy of core measurement results in the prior art is low and the development of subsequent exploration geological analysis work is affected.

In a first aspect, the present invention provides a core measurement method based on a bidirectional cascade pixel differential network, including:

acquiring a plurality of rock core images, training an edge detection model based on the plurality of rock core images, and generating a bidirectional cascade pixel differential network model;

acquiring a core box main body image, inputting the core box main body image into a bidirectional cascade pixel differential network model, and generating an initial edge prediction image;

processing the initial edge prediction graph to generate a core rectangular frame;

And carrying out core measurement on the core box main body diagram based on the core rectangular frame to generate core characteristics.

According to the core measurement method based on the bidirectional cascade pixel differential network, which is provided by the embodiment, the bidirectional cascade pixel differential network model is used for processing the initial edge prediction graph to generate the core rectangular frame, and further core measurement is performed on the core box main body graph based on the core rectangular frame, so that the accuracy of a core measurement result is improved, the exploration flow is simplified, and the exploration work efficiency is improved.

In an alternative embodiment, training an edge detection model based on a plurality of rock core images generates a bi-directional cascade pixel differential network model, comprising:

acquiring an edge detection data set, training an edge detection model based on the edge detection data set, and determining an optimal network structure; the optimal network structure consists of a pixel differential convolution network and a bidirectional cascade network;

performing edge labeling on the plurality of core images to generate an edge type data set;

performing data enhancement on the edge type data set to generate a core edge data set;

and performing migration learning training on the optimal network structure based on the core edge data set to generate a bidirectional cascade pixel differential network model.

In the above-mentioned alternative embodiment, compared with the previous semantic segmentation and example segmentation methods, training the edge detection model based on a plurality of core images lays a foundation for intelligent core identification and measurement, and improves the accuracy of core measurement results to a certain extent.

In an alternative embodiment, processing the initial edge prediction graph to generate a rectangular frame of the core includes:

performing non-maximum value inhibition processing on the initial edge prediction graph to generate a core box main body edge detection graph;

performing block segmentation processing on the edge detection map of the main body of the core box to generate a block edge detection map;

based on the block edge detection diagram, a core rectangular frame is generated by using a core boundary extraction algorithm.

In the above optional embodiment, the generated rectangular core frame is more accurate by performing non-maximum value inhibition processing, block segmentation processing and core boundary extraction algorithm on the initial edge prediction graph, so as to lay a foundation for subsequent core measurement.

In an alternative embodiment, the core features include:

core quality index value, average core length, core color, core type and layering information of the core box section.

In an alternative embodiment, performing core measurement on a core box main body diagram based on a core rectangular frame to generate core features, including:

Determining a pixel length value based on the rectangular core frame, and respectively calculating a core quality index value and a core average length based on the pixel length value;

performing color gamut correction on the core box main body diagram, and cutting the core box main body diagram after the color gamut correction according to the core rectangular frame to generate rectangular outline information of a core block area;

inputting rectangular outline information of a core block area into a convolutional neural classification network to generate core types;

determining the color of the core through a comparison table database based on the category of the core; the comparison table database comprises matching relations between different core types and different core colors;

layering information is determined based on core category and core color.

In the above alternative embodiment, the corrected main body diagram of the core box is cut to obtain rectangular outline information of the core block area, and the rectangular outline information is sent into the trained convolutional neural classification network for classification and identification according to the block area and the exploration sequence, so that the influence of other environmental factors on the background, such as trees, weeds, bare core boxes and the like, can be eliminated. Meanwhile, factors such as shooting equipment factors, weather, sunlight and the like can be eliminated through color correction, so that a more accurate identification effect is obtained.

In an alternative embodiment, calculating the core quality index value and the core average length based on the pixel length value includes:

obtaining an error coefficient, calculating the length of the core based on the pixel length value and the error coefficient, and calculating the average length of the core based on the length of the core;

and calculating a core quality index value based on the core length.

In the above alternative embodiment, the measure of average length of the core may better represent the core quality level in the case of RQD approximation.

In an alternative embodiment, determining the core color from the look-up table database based on the core category includes:

determining pixel points of the core block based on the rectangular outline information of the core block area, and comparing the pixel points of the core block with a standard color library to generate a predicted color;

determining the color of the core through a comparison table database based on the category of the core;

and comparing the predicted color with the color of the core, and outputting the color of the core when the predicted color is identical to the color of the core.

In the above optional embodiment, according to the core type detection result, an anomaly identification processing algorithm is used, that is, the predicted color is compared with the core color, and the core color is output, so that the core type can be predicted more accurately, and the core layering information can be obtained.

In a second aspect, the present invention provides a core measurement device based on a bidirectional cascade pixel differential network, including:

the training module is used for acquiring a plurality of rock core images, training the edge detection model based on the plurality of rock core images and generating a bidirectional cascade pixel differential network model;

the generation module is used for acquiring a core box main body image, inputting the core box main body image into the bidirectional cascade pixel differential network model and generating an initial edge prediction image;

the processing module is used for processing the initial edge prediction graph to generate a core rectangular frame;

and the measurement module is used for carrying out core measurement on the core box main body diagram based on the core rectangular frame to generate core characteristics.

In a third aspect, the present invention provides a computer device comprising: the core measurement method based on the bidirectional cascade pixel differential network comprises a memory and a processor, wherein the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so that the core measurement method based on the bidirectional cascade pixel differential network in the first aspect or any corresponding implementation mode is executed.

In a fourth aspect, the present invention provides a computer readable storage medium, where computer instructions are stored on the computer readable storage medium, where the computer instructions are configured to cause a computer to perform a core measurement method based on the bidirectional cascaded pixel differential network according to the first aspect or any implementation manner corresponding to the first aspect.

Drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are needed in the description of the embodiments or the prior art will be briefly described, and it is obvious that the drawings in the description below are some embodiments of the present invention, and other drawings can be obtained according to the drawings without inventive effort for a person skilled in the art.

FIG. 1 is a flow chart of a core measurement method based on a bi-directional cascading pixel differential network according to an embodiment of the present invention;

FIG. 2 is a schematic illustration of a core image according to an embodiment of the present invention;

FIG. 3 is a schematic diagram of operator detection in an embodiment of the invention;

FIG. 4 is a schematic diagram of identifying the approximate location of a core using an Anchor box in accordance with an embodiment of the present invention;

FIG. 5 is a schematic diagram of a center pixel differential convolution structure in accordance with an embodiment of the present disclosure;

FIG. 6 is a schematic diagram of a neighborhood pixel differential convolution structure according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a radiation pixel differential convolution structure in accordance with an embodiment of the present disclosure;

FIG. 8 is a flow chart of another method for core measurement based on a bi-directional cascading pixel differential network according to an embodiment of the present invention;

FIG. 9 is a core image artwork in an edge detection dataset according to an embodiment of the present invention;

FIG. 10 is a callout in an edge detection dataset of an embodiment of the present invention;

FIG. 11 is a core image artwork according to an embodiment of the present invention;

FIG. 12 is a core edge map prior to transfer learning in accordance with an embodiment of the present invention;

FIG. 13 is a core edge prediction graph after a fifth training iteration of an embodiment of the present invention;

FIG. 14 is a core edge prediction graph after a tenth training iteration of an embodiment of the present invention;

FIG. 15 is a flow chart of yet another method for core measurement based on a bi-directional cascading pixel differential network in accordance with an embodiment of the present invention;

fig. 16 is a schematic diagram of an on operation using the large rectangular collation image according to the embodiment of the invention;

FIG. 17 is a schematic view of the core boundary extraction effect according to an embodiment of the present invention;

FIG. 18 is a flow chart of yet another method for core measurement based on a bi-directional cascading pixel differential network in accordance with an embodiment of the present invention;

FIG. 19 is a schematic view of a pre-segmentation image of a core box in accordance with an embodiment of the present invention;

FIG. 20 is a schematic illustration of a core block section of an embodiment of the present invention;

FIG. 21 is a block diagram of a core measurement device based on a bi-directional cascading pixel differential network in accordance with an embodiment of the present invention;

Fig. 22 is a schematic diagram of a hardware structure of a computer device according to an embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments of the present invention. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

RQD represents an important rock mass quality evaluation index, and is widely applied to hydraulic engineering. RQD is defined as the ratio of the length of a core segment greater than 10cm (centimeters) to the length of the readback. In geological exploration analysis work, the color difference of the core can reflect different rock characteristics, and the RQD value of the core can directly reflect exploration geological conditions. Because field exploration working conditions are complex, the risk is high, and interference of a plurality of environmental factors exists, core measurement is inconvenient in the field, the whole measurement process has the problems of strong subjectivity, long period, high risk degree and the like, and because different people have subjective differences in color identification, color recording may be inaccurate, and the development of subsequent exploration geological analysis work is affected.

In the field of image edge detection, traditional detection operators such as Sobel (Sobel operator), canny (edge detection operator) and the like all adopt differential gradient information to represent abrupt changes and detail characteristics of an image edge context, but traditional operator models based on manually setting the size of a filter kernel are often limited to shallow layer representation capability, and due to complex core image shooting environment, the surface of a core has many conditions of tiny lines, uneven colors and the like, and accurate edge detection results are difficult to obtain under the complex scene of a core picture by using the traditional detection operators, as shown in fig. 2-3.

The target detection algorithm based on deep learning can only identify the approximate position of the core by using an Anchor box, and can not accurately detect the length of the core; the use of semantic segmentation requires a larger data set to achieve good results while training reasoning times are longer. As shown in fig. 4, semantic segmentation is a computer vision technique that aims at segmenting an image into regions of semantic significance. It can identify to which class each pixel in the image belongs, but cannot accurately identify the contour edge of the object.

In the field of deep learning, there are common edge detection neural networks, such as HED (holistical-Nested Edge Detection, an end-to-end edge detection network), RCF (a precision edge detector using richer convolution features), and the like, and CNN (convolution neural network) used by the same can effectively capture semantic features of an image. However, in the random initialization process of the convolution kernel, the ordinary convolution operation is difficult to focus image gradient information in the training process because of no limitation of gradient coding, so that the training difficulty for edge detection is improved, and the accuracy of edge prediction is affected.

The related art discloses a rapid RQD analysis method, which adopts different image recognition technologies to convert pictures into 8-system images and utilizes image scale software to extract image information. The image identification segmentation is carried out by utilizing digital image processing, the calculation is simple, the efficiency is high, only the characteristics of the gray values of the pixel points are considered, the spatial characteristics are generally not considered, and the core identification rate of the picture with similar background color is not high.

The related art discloses an intelligent method for identifying RQD from a borehole core photo, which is based on Mask-R-CNN (an example segmentation algorithm) deep learning network to identify images, trains single-row cores, and performs statistics on the RQD. The Mask-Head for prediction is introduced (part of a classifier or a full connection layer is removed in a classification network), a segmentation Mask is predicted in a pixel-to-pixel mode, the segmentation precision is improved, but a classification frame and the prediction Mask share an evaluation function, and sometimes a segmentation result is interfered.

Related art discloses a RQD calculation method based on a deep learning model and a core image, and image recognition is performed on the image based on a Unet (deep learning network) and is also performed on a single-row core for recognition statistics. The Unet deep learning network has the advantage of obtaining a better segmentation effect on a small training set.

The related art discloses a core RQD digital statistical method, equipment and a terminal based on an image recognition technology, which are mainly designed on a system level and do not relate to a specific edge detection method.

The related art discloses that a UNet segmentation model is utilized to carry out semantic segmentation on a drill core image, a core region and a background region are obtained, a single row of drill core images are found out, then a core contour is obtained by using a traditional edge detection algorithm, a pixel waveform diagram is manufactured, and a curve fitting method is adopted to fit partial boundary lines to obtain the boundary lines of the core. The method is good for complete core identification, but when more broken stones exist in a core box and a background area is complex, accurate contours are difficult to identify through curve fitting. Meanwhile, the method is low in recognition rate, the recognition points and the boundary line of the core box need to be corrected, the method is only suitable for standard ideal conditions, the requirements on the shooting angle and the recognition effect of the picture of the core box are high, and the applicability to field complex conditions is poor.

The accuracy of the core measurement result is a core requirement of intelligent core measurement, and the related technology is influenced by external factors, so that the accuracy of the core measurement result is low, and further the development of subsequent exploration geological analysis work is influenced.

According to an embodiment of the present invention, there is provided an embodiment of a core measurement method based on a bi-directional cascading pixel differential network, it should be noted that the steps illustrated in the flowchart of the drawings may be performed in a computer system such as a set of computer executable instructions, and that although a logical order is illustrated in the flowchart, in some cases, the steps illustrated or described may be performed in an order different from that herein.

In this embodiment, a core measurement method based on a bidirectional cascade pixel differential network is provided, which may be used in the above mobile terminal, such as a mobile phone, a tablet computer, etc., fig. 1 is a flowchart of a core measurement method based on a bidirectional cascade pixel differential network according to an embodiment of the present invention, as shown in fig. 1, where the flowchart includes the following steps:

step S101, a plurality of rock core images are acquired, an edge detection model is trained based on the plurality of rock core images, and a bidirectional cascade pixel differential network model is generated.

Specifically, the pixel differential convolution has strict reasoning evidence in mathematics, the pixel differential convolution can be converted into a common convolution form through a corresponding calculation formula, two convolution kernels are arranged for image feature extraction during the calculation process of the pixel differential convolution, the differential value of the convolution kernel weight can be converted into the common convolution through calculation during the reasoning, and then the convolution operation is carried out on an input feature map; the method not only can accelerate the speed of the training stage, but also can reduce the required extra operation amount in the reasoning process, namely, the method obtains better network effect by using twice training time, but does not increase the parameter complexity of the network model and the time required by image reasoning in the reasoning process.

Further, as shown in fig. 5, the main implementation manner of the central pixel differential convolution (Central Pixel difference convolution, CPDC) is that during the convolution operation, a convolution kernel central point (such as x1, x2, x3, x4, x5, x6, x7, x8 and x9 in fig. 5) corresponding to each image is selected as a reference point, the neighborhood feature of the reference point is subjected to central difference, that is, other pixel point values (i.e., x 5) in the neighborhood of the reference point are differenced from the reference value (such as x1-x5, x2-x5, x3-x5, x4-x5, x5-x5, x6-x5, x7-x5, x8-x5 and x9-x5 in fig. 5), then the reference point pixel values are set to 0, and then new feature maps obtained by the central pixel differential convolution check are subjected to convolution operation, and finally the result (such as w1, w2, w3, w4, w6, w7, w8 and w9 in fig. 5) is outputted; the conversion formula of the center pixel differential convolution CPDC is as follows:

in the above formula, y1 represents the output result of the center pixel differential convolution, w _i Represents the weight, i represents the number of convolutions, x _i Representing the original range of pixel values (gray scale),representing the new weight matrix.

Further, the neighborhood pixel differential convolution (Angular Pixel difference convolution, APDC) is substantially similar to the central pixel differential convolution, and the convolution kernels are 33, which is different in that the neighborhood pixel differential convolution is to perform two-by-two differential on the neighborhood characteristics of the reference point, that is, the pixels in the neighborhood of the reference point are alternately differenced in the clockwise direction, then the pixel value of the reference point is set to 0, then the convolution operation is performed on the new characteristic diagram obtained by checking the neighborhood pixel differential convolution, and finally the result is output. The specific structure diagram of the APDC is shown in fig. 6, and the difference operation can be performed by converting the differential convolution of the neighboring pixel into two common convolutions through the derivation of the APDC conversion formula, where the differential convolution of the neighboring pixel APDC conversion formula is as follows:

In the above expression, y2 represents the output result of the neighborhood pixel differential convolution.

Further, the difference convolution of the radiation pixels (Radial Pixel difference convolution, RPDC) is slightly different from the two difference convolutions, the APDC and the CPDC mainly perform the difference convolution in the neighborhood of the size of the reference point 33, while the difference convolution of the radiation pixels RPDC has a larger receptive field, the size is 55, the RPDC mainly performs the difference between the outer ring and the inner ring of the difference field in a diagonal form, the specific structure diagram is shown in fig. 7, and the conversion formula of the RPDC is as follows:

in the above equation, y3 represents the output result of the radiation pixel differential convolution.

Further, after model training is finished, the difference value in the common convolution kernel can be calculated first, and then the calculation weight and the image pixel point are subjected to convolution operation, so that the reasoning speed is increased. The differential convolution kernel is trained by using twice the computational power in the training stage, and the speed of the reasoning stage is consistent with that of a common convolution neural network.

Further, in the original PiDiNet network, each Stage layer uses the same truth Label graph to supervise the network. In fact, however, the shallow network Stage layer in the pidanet network can only extract local low-level semantic information because of the receptive field relationship, and the deep network Stage layer can notice large-scale target semantic information. If the same edge Label graph is used for supervising all layers of the network, the network learning training cannot be guided well, all layers cannot learn the scale information suitable for being acquired by the layers well, and generalization capability is poor.

Furthermore, the BSDS500 data set is used as one of the most commonly used data sets in the edge detection field, edges among various objects are included, the edge detection effect on the data set can well evaluate the performance of the model, and the evaluation result has wide acceptance, so that the BSDS500 data set is selected to carry out ablation experimental analysis on the network. The BSDS500 edge detection dataset sums and averages edge truth diagrams annotated by different annotators to obtain an edge probability diagram, wherein the range from 0 to 1,0 is represented as non-edge, and 1 indicates that all annotators have annotations on the pixel.

Further, as different images have different scale characteristics, as parameters and structures of the network model are changed, the scale information extracted by different network layers in the model is different, and the scale information to be extracted by different network layers is basically not feasible to manually select; in order to solve the problems, a Bi-directional cascade module (Bi-Directional Cascade, BDC) is introduced on the basis of a PiDiNet model (pixel differential network model) to better acquire multi-scale information in a network; the mathematical principle of the bidirectional cascade module is as follows: let Y be a Label (Label) image of an edge, which contains not only small-scale edges but also larger-scale edges, so that there is a large scale variation, and in order to make different networks learn to fit the scale extracted by the self receptive field size, the edge Label image can be decomposed into stacks of different scales, expressed as follows:

Y＝∑Y _s (4)

In the above, Y _s And (5) representing Label corresponding to a certain network layer of the deep learning neural network.

Further, let N be _s D is set for the probability diagram of the output characteristic diagram of the s-layer network convolution layer _s The binarization operation of the extraction edge is used, and the supervision mode of the s layer is as follows:

loss＝∑|D _s (N _s )-Y _s | (5)

where loss represents the loss function.

Further, the Label map Y of the edge is used to subtract the output confidence map Y of the other layer _i Wherein i e [1,2, j-1, j+1, ], s]Thereby obtaining Y _j And as a supervision Label of the j-th layer, the calculation formula is as follows:

Y _s ≈∑ _i＝s P _i (6)

in the above, P _i Indicating the confidence error of the i-th layer.

Further, due to Y _s And P _i Is a linear relationship between the two, Y is known from the chain law of derivatives _s For any one P _i The derivatives of (2) are all 1; for arbitrary binarization operation D _i All accept the same gradient, which also means that the network supervision mode also uses each layer in the same Label supervision network, and the middle layer does not automatically learn the scale information suitable for itself; to solve this problem, label map Y may be used _s The method is divided into two complementary supervision, and the specific formula is as follows:

in the above-mentioned method, the step of,representing forward supervision, is compromised>Confidence error indicating layer i in forward supervision,/- >Indicating reverse supervision->Indicating the confidence error of the i-th layer in the reverse supervision.

Further, forward supervision ignores supervision smaller than the S-layer network scale, and reverse supervision ignores supervision larger than the S-layer network scale; the output of the binarization operation of each extracted edge is the supervision of D relative to the parameters of the D layer, so that the network can learn the scale information matched with the feeling of the network layer according to the back propagation of errors.

Further, the details of the network model structure such as the up-sampling mode of the model are improved, and through an ablation experiment, the edge detection ODS (Optimal Dataset Scale, global best, fixed threshold) OIS (Optimal Image Scale, single-image best, best threshold for each image) and the AP (average accuracy) curve are compared, and the results are shown in the following table 1.

Table 1:

further, different edge detection problems usually focus on different scales, and a network with higher average accuracy rate AP can better adapt to different task requirements during transfer learning training, and BDC-PiDiNet (bidirectional cascade pixel differential network) is selected as final transfer learning model data by combining ODS and OIS.

Step S102, a core box main body image is obtained, the core box main body image is input into a bidirectional cascade pixel differential network model, and an initial edge prediction image is generated.

And step S103, processing the initial edge prediction graph to generate a core rectangular frame.

And step S104, carrying out core measurement on the core box main body diagram based on the core rectangular frame to generate core characteristics.

Specifically, core features, including: core quality index value, average core length, core color, core type and layering information of the core box section.

According to the core measurement method based on the bidirectional cascade pixel differential network, the pixel differential convolution network and the bidirectional cascade network are used, an improved lightweight edge detection network BDC-PiDiNet is provided, the BDC-PiDiNet network uses sub-pixel convolution to replace a traditional up-sampling method, a coding part adopts pixel differential convolution to extract original gradient information of an image, a decoding part adopts a bidirectional cascade module to enhance effectiveness of network multi-scale information extraction, the problem that the traditional operator model is often limited to shallow layer characterization capacity and common convolution in a CNN convolution neural network is difficult to focus image gradient information in a training process is solved, a convolution kernel weight differential value is skillfully converted into the common convolution, then convolution operation is carried out on an input feature map, a better network effect is obtained by double training time, but in a reasoning process, parameter complexity of the network model and time required by image reasoning are not increased; the advantages of the traditional operator gradient information and the deep learning convolution operation are combined, the influence of different network structures on the detection effect is discussed through carrying out an ablation experiment, and the optimal BDC-PiDiNet architecture is selected, so that compared with the original PiDiNet network, ODS, OIS and AP values are all improved; in addition, the full-process high-precision rock core classification identification and intelligent measurement method is provided, the intelligent transformation of the exploration industry is facilitated, the problems of complex field exploration working conditions, high risk, high subjectivity, long period, high risk degree and the like caused by interference of a plurality of environmental factors are solved, the exploration process is greatly simplified, and the exploration working efficiency is improved.

In this embodiment, a core measurement method based on a bidirectional cascade pixel differential network is provided, which may be used in the above mobile terminal, such as a mobile phone, a tablet computer, etc., fig. 8 is a flowchart of a core measurement method based on a bidirectional cascade pixel differential network according to an embodiment of the present invention, as shown in fig. 8, where the flowchart includes the following steps:

step S801, a plurality of core images are acquired, and an edge detection model is trained based on the plurality of core images, so that a bidirectional cascade pixel differential network model is generated.

Specifically, the step S801 includes:

step S8011, acquiring an edge detection data set, training an edge detection model based on the edge detection data set, and determining an optimal network structure; the optimal network structure consists of a pixel differential convolution network and a bidirectional cascade network.

Specifically, the edge detection dataset employs a BSDS dataset.

And step S8012, performing edge labeling on the plurality of core images to generate an edge type data set.

In particular, since the BSDS data set includes not only large-scale edges but also small-scale edges, it is necessary to construct some core edge data sets for core boxes and cores to fine tune the network to meet the task requirements.

Further, the original photo is usually a field shot picture of a staff, and because the shooting posture and the geographic position are not fixed, the shape of the core box area in the shot picture is different, in order to solve the above problem, a simple image processing tool is written by using python (computer programming language) codes, and the tool needs to manually select 4 corner points of the core box by using a staff, and the irregular core box is changed into a more regular rectangular box shape through perspective transformation.

Further, after obtaining the core image, line segment (linetrip) marking is required to be carried out on the edge of the core in the core box, and a Labelme tool (which is an image marking tool developed by computer science and artificial intelligence laboratories of the Massachusetts) is used for selecting the contour of the core in a mouse point mode; because the edges of the core area are tidy and most of the core area are straight line segments, in order to improve the marking speed, the edge marking type is selected as a Polygon (Polygon) in the marking process, and the Polygon can automatically generate a closed shape compared with the Linestrip type; because different core boxes have different photo sizes, the edge map is generated by using a manually marked JSON (JavaScript Object Notation, JS object numbered musical notation, which is a lightweight data exchange format) file, and the like, so that the edge thickness of the generated map is ensured to be suitable for the size of an input picture by modifying Labelme2voc.py code bottom layer generation logic, rewriting the Polygon field in the JSON file into Linestrip, modifying an edge generation code, and selecting the edge thickness corresponding to the image pixel size according to the size of the image pixel.

And step S8013, carrying out data enhancement on the edge type data set to generate a core edge data set.

In particular, as shown in fig. 9-10, in order to increase the disturbance rejection of the model, the edge type dataset obtained by labeling is subjected to data enhancement by using an open source tool Imguag, and about 30000 core edge datasets are generated by adopting random combination modes such as scaling, clipping, rotation, flipping, channel RGB (color system) value change and the like.

And step S8014, performing migration learning training on the optimal network structure based on the core edge data set, and generating a bidirectional cascade pixel differential network model.

Specifically, the relevant parameters of BDC-PiDiNet are set as follows: training graphics card V100G, using Adam (an extension to the random gradient descent method) optimizer, global learning rate set to 0.002000, number of iterations 10, weight attenuation value 0.000100, and learning rate divided by 10 after the 5 th and 7 th epoch iterations.

Further, as shown in fig. 11-14, the effects before and after the transfer learning are compared, and the original image is not used in the training set. The core contour is better extracted in the core edge prediction graph generated by the network before the transfer learning, but the noise is more, the surface texture is obvious, the core contour can be better extracted by the BDC-PiDiNe after the transfer learning from the EPOCH5 and EPOCH10 effect graphs, and meanwhile, the core surface texture and other noises are ignored, so that the subsequent core splitting effect is greatly improved, and the network weight parameter trained by the EPOCH10 is selected as the final network parameter of the intelligent detection task.

In the optional implementation manner, about 30000 core edge photo data sets and BDC_Picodiet edge detection lightweight models are manufactured, and the method is different from the prior semantic segmentation and example segmentation methods, so that a new thought and direction are provided for intelligent identification and measurement of the core, and a foundation is laid.

Step S802, a core box main body image is obtained, the core box main body image is input into a bidirectional cascade pixel differential network model, and an initial edge prediction image is generated. Please refer to step S102 in the embodiment shown in fig. 1 in detail, which is not described herein.

Step S803, the initial edge prediction graph is processed to generate a core rectangular frame. Please refer to step S103 in the embodiment shown in fig. 1 in detail, which is not described herein.

Step S804, core measurement is carried out on the core box main body diagram based on the core rectangular frame, and core characteristics are generated. Please refer to step S104 in the embodiment shown in fig. 1 in detail, which is not described herein.

In this embodiment, a core measurement method based on a bidirectional cascade pixel differential network is provided, which may be used for the mobile terminal, such as a mobile phone, a tablet computer, etc., fig. 15 is a flowchart of a core measurement method based on a bidirectional cascade pixel differential network according to an embodiment of the present invention, as shown in fig. 15, where the flowchart includes the following steps:

Step S1501 acquires a plurality of core images, trains an edge detection model based on the plurality of core images, and generates a bidirectional cascade pixel differential network model. Please refer to step S801 in the embodiment shown in fig. 8 in detail, which is not described herein.

Step S1502, a core box main body image is acquired, and the core box main body image is input into a bidirectional cascade pixel differential network model to generate an initial edge prediction image. Please refer to step S802 in the embodiment shown in fig. 8 in detail, which is not described herein.

And step S1503, processing the initial edge prediction graph to generate a core rectangular frame.

Specifically, the step S1501 includes:

in step S15011, non-maximum value suppression processing is performed on the initial edge prediction map, and a core box body edge detection map is generated.

Specifically, different cores generally have different texture information, and besides severe gradient changes at the edges of the cores and the core boxes, the gradient changes of the surfaces of the cores are also large; furthermore, the BDC-PiDiNet network model obtained through transfer learning is used for ignoring the gradient of the surface of the core in the training process, the large-scale information of the edge of the core is mainly learned, the surface texture processing effect on the core image is better, but because the edge data set of the core is purely manually marked, a small error can exist in the marking process, so that a region with more severe gradient change still exists in the edge prediction generation map of the BDC-PiDiNet, the core is directly segmented by using the generated edge prediction map, larger error can be generated, the accuracy of core measurement information is seriously influenced, and therefore, the correlation code of the sharpening processing of the edge image of the core is written by using python language, and the correlation code of the sharpening processing of the edge image of the core is improved by morphological correlation operation, wherein:

Further, after the core box image is processed by the BDC_Picodiet network model, an initial edge prediction graph img is generated, and the edge is thicker, so that the prediction graph is required to be processed by a non-maximum suppression (NMS) operation, and the boundary of the core is thinned.

Further, the shape of the large rectangular convolution kernel is similar to that of the rock core, in the edge detection process, the boundary of the rock core is relatively fuzzy, the edge of the rock core is not closed after being subjected to OTSU (Otsu) binarization treatment, the rock core is intelligently segmented and measured in a closed graph contour detection mode, the influence of the edge of the rock core on the follow-up intelligent measurement algorithm can be generated, and the RQD measurement accuracy is seriously reduced, so that the large rectangular check image is used for carrying out opening operation, the boundary of the rock core is continuous, and as shown in fig. 16, the edge part of the rock core is well closed.

Step S15012, performing block segmentation processing on the core box body edge detection map to generate a block edge detection map.

Specifically, the edge map processed by image morphology not only removes the influence of internal texture of the core, but also sharpens the edge profile of the core, so that the edge of the core is clearer and more continuous, good core profile curve information can be obtained, but also other background environmental influences exist, and the core is required to be processed.

Further, as obvious separation limits are arranged between different block areas of the core box, the block areas are segmented through the separation limits; because the angle problem of shooing, can produce vertical stretching phenomenon when carrying out perspective transformation in rock core case image, highly different between the different rock core case pieces promptly, consequently adopt and carry out rock core case segmentation according to the mode of coordinate partition and probably have great error, can use straight line to detect fusion algorithm, its relevant code is:

further, assuming that the image size is w×h, the detection result is screened by setting the length of a straight line, the inclination angle, and the like using an LSD straight line detection algorithm (Line Segment Detector, a line segment detection algorithm); after detecting the straight line existing in the image, the expansion operation is needed to be used for detecting the straight line, and the operation can be used for communicating the shorter white straight line area so as to obtain the center point of the communicated area and highly group the center point by the coordinates of the center point; obtaining a core box block edge detection line DetectLines by performing line fitting on the coordinates of the central points; since the edge of the core in the original image is also a straight line, many edges obtained by the straight line detection mode are core edges, so that the edges need to be processed through priori knowledge.

Further, firstly dividing the core box image into 5 blocks according to priori knowledge, wherein each block has the height of H/5, H represents the height of the core box image, an equal cutting block edge line RealLines is fitted according to the block height, the RealLines is used for traversing the DetectLines, the difference value between the straight line and the predicted straight line is calculated and divided into an upper direction and a lower direction, the height difference between the upper direction and the lower direction is calculated, if the RealLine has two DetectLines on the upper side and the lower side at the same time and the distance is close, the RealLine is possibly positioned between the two core edges, and the RealLine is unchanged if the position is correct; if the two detectlines are far apart, selecting the nearest DetectLine to replace the RealLine; and the partition number of the core box can be guaranteed to be 5, and meanwhile, the influence of the false-detected core straight line on the blocking of the core box is well considered.

Step S15013, based on the block edge detection map, a core rectangular frame is generated by using a core boundary extraction algorithm.

Specifically, after obtaining the block edge detection diagram, an OpenCV (cross-platform computer vision and machine learning software library) tool can be used for obtaining the outline of a single core in the core box diagram by using the core box edge detection diagram, and the block area of the single core can be obtained by using the core box partition diagram, so that the intelligent calculation and display of the RQD are facilitated; the core boundary extraction algorithm code is as follows:

Further, the core edge profile map contains all core profiles and also contains some noise profiles, and the surface of the core still has some texture edges, so that the detection result is affected; in addition, because errors may exist in the process of manually selecting the corner points of the core box, the edge of the junction between the outer area of the core box and the core box is also detected and segmented, and the detected non-core contours are required to be filtered; setting related filtering rules by using priori knowledge of the core box, and filtering the detected closed edges; the specific filtering rules are as follows: and (1) the core profile height-width ratio S is smaller than 3. The main detection object is a core with the height being more than 10cm, and the width of the core is usually more than the height of the core; (2) core height rock_h is less than Block height block_h: the rock core is stored in the rock core box block area, so the height is inevitably smaller than the height of the block area, and the contour height is larger than the height of the block area, which is usually the noise area at the left and right edges of the rock core box diagram; (3) The center of the profile is greater than a set threshold value from the left and right boundaries of the core box diagram, the center position of the core is usually at a relatively center position inside the core box, and the center position is usually a noise area relatively close to the boundary of the core box diagram; (4) When classifying the rock cores according to the block areas, filtering boxes (block areas) with the central height of the rock core profile being relatively close to the boundary of the block areas, wherein the boxes are set to be 1/5 of the height of the rock cores, and the boxes are used for filtering the profile formed by the rock core box and the areas outside the rock cores; (5) Areas of smaller area, typically core surface noise contours and some broken rock, are filtered out, setting the threshold size to 1/50 of the image size.

Further, as shown in fig. 17, the core boundary of the final contour segmentation result and the edge detection is shown, the core boundary extraction algorithm preferably extracts block region boundary information corresponding to the core, including center coordinates, width, height, rotation angle and the like, and further determines a core rectangular frame based on the block region boundary information, so that follow-up core intelligent measurement is facilitated.

In the alternative embodiment, through setting the interval threshold value, the straight line detection fusion algorithm is improved, the problem that the detection straight line of the original straight line detection fusion algorithm is smaller than 4 when the qualified straight line is smaller than 5, the case body segmentation area is smaller than 4, the automation of the core case segmentation measurement can be truly realized, the contour recognition extraction rule is set by applying priori knowledge and characteristic engineering, the non-core closed contour line is screened and noise is reduced, the follow-up calculation of the core segment RQD is facilitated, and the extremely effective effect is achieved.

Step S1504, core measurement is carried out on the core box main body diagram based on the core rectangular frame, and core characteristics are generated. Please refer to step S804 in the embodiment shown in fig. 8, which is not described herein.

In this embodiment, a core measurement method based on a bidirectional cascade pixel differential network is provided, which may be used for the mobile terminal, such as a mobile phone, a tablet computer, etc., fig. 18 is a flowchart of a core measurement method based on a bidirectional cascade pixel differential network according to an embodiment of the present invention, as shown in fig. 18, where the flowchart includes the following steps:

Step S1801, acquiring a plurality of core images, training an edge detection model based on the plurality of core images, and generating a bidirectional cascade pixel differential network model. Please refer to step S1501 in the embodiment shown in fig. 15 in detail, which is not described herein.

Step S1802, a core box main body image is acquired, and the core box main body image is input into a bidirectional cascade pixel differential network model to generate an initial edge prediction image. Please refer to step S1502 in the embodiment shown in fig. 15 in detail, which is not described herein.

Step S1803, processing the initial edge prediction map to generate a rectangular core frame. Please refer to step S1502 in the embodiment shown in fig. 15 in detail, which is not described herein.

Step S1804, performing core measurement on the main body diagram of the core box based on the rectangular core frame, and generating core features.

Specifically, the step S1804 includes:

in step S18041, a pixel length value is determined based on the rectangular core frame, and a core quality index value and a core average length are respectively calculated based on the pixel length value.

In some alternative embodiments, step S18041 includes:

and a step a1, obtaining an error coefficient, calculating the length of the core based on the pixel length value and the error coefficient, and calculating the average length of the core based on the length of the core.

Specifically, after obtaining a rectangular frame of the core boundary (i.e. a rectangular frame of the core), we can perform calculation of the core length and measurement of the RQD by measuring the pixel length value of the rectangular frame; because the length of the core box is usually 1 meter, and the length of the core box cut out by the original graph is about 1 meter, the equal proportion measurement and calculation can be carried out according to the pixel value of the rectangular outline of the core and the length of the core box, and the formula is as follows:

in the above formula, rock_l represents the core length, pixel _{box_l} The length of the rectangular frame of the core is the length of the Pixel _{blocx_l} θ represents the error coefficient for the length of the core block region.

Further, through multiple measurements, the pixel width of the cut core box image is about 1.01 times of the actual width, so that 1.01 is usually taken, and the final measurement result is converted into cm (centimeter) units, so that the display is convenient.

Further, since a single RQD value does not exhibit well the core quality, for example, the RQD calculation is 90% for both 9 cores 10cm long and 1 core 90cm long, but the crushing is significantly better for a single core 90cm long, the average core length is calculated using the following calculation formula:

in the above formula, avg_l is used to represent the average length of the core block with a length greater than 10cm, and the greater the RQD, the better the core breaking degree.

And a step a2, calculating a core quality index value based on the core length.

Specifically, the calculation formula of the RQD (core quality index value) is as follows:

wherein Sigma rock_l #, is equal to _L>10cm ) The sum of the lengths of the cores in each block, namely the sum of the lengths of the core boxes, is usually 1 meter.

In this alternative embodiment, a measure of the average length of the core is provided, which may better demonstrate the core quality level in the case of RQD approximation.

Step S18042, performing color gamut correction on the core box main body diagram, and cutting the core box main body diagram after the color gamut correction according to the core rectangular frame to generate rectangular contour information of the core block area.

And step S18043, inputting rectangular outline information of the core block area into a convolutional neural classification network to generate a core type.

Specifically, as shown in fig. 19-20, by cutting the identified core contour to obtain the color-corrected core block diagram, according to the block area and the exploration sequence, eliminating the influence of other environmental factors on the background, such as trees, weeds, bare core boxes and the like, and further respectively sending the core block contour diagram and the core box diagram into the trained convolutional neural classification network VGG16 through sliding windows for classification and identification, better lithology identification effect can be obtained, and simultaneously, factors of shooting equipment, weather, sunlight and the like can be eliminated through color correction, so that better core identification effect can be obtained.

Further, according to the core type detection result, automatically identifying lithology classification abnormal sections by using an abnormal identification processing algorithm; the rock core block identification only gives a lithology result, and can not learn and acquire the relations among different rock core blocks well, and the rock core box sections are required to be identified in a sliding window mode in a blocking mode; in general, in the lithology recognition result of the sliding window mode, the problem that the intercepted lithology classification is inconsistent due to different window sizes exists in the same rock interval, the lithology mutation condition occurs, and the lithology recognition error result needs to be processed aiming at the condition; the identifying exception handling algorithm is provided, lithology classifying exception segments are automatically identified, windows of 8cm, 12cm and 16cm are taken respectively by taking the middle point of a sliding window as the center, lithology classifying VGG16 networks are sent to be identified by taking 4cm as the step length, lithology types with the largest current result are taken as the final result, and meanwhile, the results are summarized, so that rock core type change information with different heights can be given.

Step S18044, determining the color of the core through a comparison table database based on the core type; the comparison table database comprises matching relations between different core types and different core colors.

In some alternative embodiments, step S18044 includes:

and b1, determining pixel points of the core block based on the rectangular outline information of the core block area, and comparing the pixel points of the core block with a standard color library to generate a predicted color.

Specifically, the core picture can influence the identification of the color of the core due to factors such as shooting angles, weather, illumination and the like, and then the color of the whole image is calibrated by identifying the three primary colors such as the standard color chart red, green and blue in the exploration condition plate in the core field shooting picture, intelligently identifying the color chart and then restoring the color chart into the standard RGB pixel value.

Further, cutting the core box main body diagram after color gamut correction according to the core rectangular frame to generate core block rectangular outline information, traversing each pixel point of the core block according to the core block rectangular outline information, comparing the pixel point with an international standard color library to obtain RGB pixel values, selecting the standard color with the minimum mean square error as a final predicted color, and carrying out affine transformation on the core outline based on the principle so as to facilitate all traversal of the image pixel points, wherein the algorithm flow is as follows:

and b2, determining the color of the core through a comparison table database based on the core type.

And b3, comparing the predicted color with the color of the core, and outputting the color of the core when the predicted color is the same as the color of the core.

Specifically, by the sliding window mode lithology recognition method, a lithology distribution result (namely a rock core classification result) can be obtained, lithology at different stages is judged, corresponding color relations are judged, possible inaccurate places are judged, and further verification is carried out; comparing the lithology recognition result with the data in the core color comparison table database to see if there is an inconsistent or unreasonable place and generating a corresponding cross-validated core classification result, e.g., if a core block is recognized as sandstone, but its color is very different from the color of sandstone in the database, then an anomaly recognition situation may be possible; cross-verifying the core classification result, namely correcting or reclassifying the abnormal recognition condition according to the data in the core color comparison table database to obtain more reliable core type and layering information; for example, if a core block is identified as sandstone, but its color is very different from the color of sandstone in the database, it is reclassified as another type of core, or its classification probability is adjusted.

For example, the core block 5 is identified as a yellowish sandy shale, but its color is not consistent with the color of sandy shale in the database, and thus may be an anomaly identification situation. According to the color information of the core of other types in the database, the core can be reclassifying to the types with higher possibility such as light yellow carbonates or light yellow clay, and the like, and the corresponding classification probability is given; the cross-validation results may be light yellow carbonates (0.6), light yellow clays (0.3), light yellow sandy shale (0.1), etc.

In the alternative embodiment, an edge detection core contour recognition algorithm is provided, so that the core contour can be accurately recognized, and the accuracy of automatic core measurement is improved.

In step S18045, layering information is determined based on the core type and the core color.

Specifically, judging whether an interface with lithology change exists or not by comparing lithology types and colors of different core blocks, and taking the interface as a layering boundary (namely layering information) if the interface exists; for example, if the lithology type and color of core block 1 and core block 2 are both gray-white shale and the lithology type and color of core block 3 are reddish-brown shale, then a layered boundary may be considered to exist between core block 2 and core block 3. By the method, the whole core box section can be divided into a plurality of different layers, and information such as thickness, position, lithology type and the like of each layer is given.

Further, the system will output a detailed survey report containing detailed results of core classification identification and RQD measurements, providing valuable references and guidance to the survey workers.

In the alternative embodiment, according to the core type detection result, the abnormal identification processing algorithm is used for automatically identifying lithology classification abnormal sections, and the multi-dimensional data are combined with core color information and position information to cross-verify the core classification result, so that the core type can be predicted more accurately, and core layering information can be obtained.

The following describes, by way of a specific embodiment, a procedure for core measurement based on a bi-directional cascading pixel differential network.

Example 1:

constructing an edge detection network model by combining a bidirectional cascade network with pixel differential convolution, training by using a common edge detection effect evaluation data set, modifying and evaluating the network model, and selecting an optimal network structure;

performing edge labeling on the original data through the segmented core box main body diagram, establishing a core edge image dataset, and performing data augmentation;

performing migration learning training on the optimal network structure through the rock core image data set, and setting and selecting an optimal iterative network model;

Morphological processing is carried out on the main body diagram of the incoming core box, and a non-maximum suppression algorithm is applied to ensure that the contour of the edge of the core contour is clear and closed;

performing intelligent block segmentation processing on the edge detection diagram of the core box main body through a core box block segmentation algorithm to obtain block position coordinate information;

calculating a core box segment RQD value and a core average length according to the block area information through a core intelligent measurement algorithm, so as to realize RQD intelligent measurement;

performing color gamut calibration on the main body diagram of the core box through a color calibration algorithm, cutting the picture according to the contour of the core, and preliminarily estimating the color of the core through a core color measurement algorithm;

and acquiring core type and layering information by using an image classification deep learning network model, fusing multidimensional data and combining core colors and a core profile map, and giving an exploration report.

In the embodiment, the rock is segmented and measured based on the rock core edge prediction graph, so that the rock core characteristics can be obtained rapidly and accurately, the automation of rock core measurement under the field condition can be realized, the applicability is wider, the task requirement of intelligent rock core measurement is better met, the measurement flow is simplified, the exploration efficiency is improved, the method is superior to the traditional manual detection method, and the RQD calculation accuracy is superior to that of the related technology; and the problems that other environmental factors such as trees, weeds, exposed core boxes and the like are influenced on the background are solved. Meanwhile, as the rock core picture shooting equipment is different, imaging is different, factors such as weather and sunlight also influence the identification accuracy of the rock core color, the picture is required to be color corrected, and a color corrected rock core block diagram obtained by cutting the identified rock core contour is sent into a convolutional neural classification network, so that the influence of shooting background environment can be removed, and better and more accurate identification and measurement effects can be obtained.

In this embodiment, a core measurement device based on a bidirectional cascaded pixel differential network is further provided, and the core measurement device is used to implement the foregoing embodiments and preferred embodiments, and is not described herein again. As used below, the term "module" may be a combination of software and/or hardware that implements a predetermined function. While the means described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware, is also possible and contemplated.

The present embodiment provides a core measurement device based on a bidirectional cascade pixel differential network, as shown in fig. 21, including:

the training module 211 is configured to acquire a plurality of core images, train the edge detection model based on the plurality of core images, and generate a bidirectional cascade pixel differential network model;

the generating module 212 is configured to obtain a core box main body image, input the core box main body image into a bidirectional cascade pixel differential network model, and generate an initial edge prediction image;

the processing module 213 is configured to process the initial edge prediction graph to generate a rectangular core frame;

and the measurement module 214 is used for carrying out core measurement on the core box main body diagram based on the core rectangular frame to generate core characteristics.

In some alternative embodiments, training module 211 includes:

the training unit is used for acquiring an edge detection data set, training an edge detection model based on the edge detection data set and determining an optimal network structure; the optimal network structure consists of a pixel differential convolution network and a bidirectional cascade network; the marking unit is used for marking edges of the plurality of core images and generating an edge type data set; the data enhancement unit is used for carrying out data enhancement on the edge type data set to generate a core edge data set; and the training unit is used for performing migration learning training on the optimal network structure based on the core edge data set, and generating a bidirectional cascade pixel differential network model.

In some alternative embodiments, the processing module 213 includes: the non-maximum value suppression processing unit is used for performing non-maximum value suppression processing on the initial edge prediction graph to generate a core box main body edge detection graph; the block segmentation processing unit is used for carrying out block segmentation processing on the edge detection diagram of the main body of the core box to generate a block edge detection diagram; and the core boundary extraction unit is used for generating a core rectangular frame by utilizing a core boundary extraction algorithm based on the block edge detection diagram.

In some alternative embodiments, the measurement module 214 includes: the calculating unit is used for determining a pixel length value based on the rectangular core frame and respectively calculating a core quality index value and a core average length based on the pixel length value; the correcting unit is used for carrying out color gamut correction on the core box main body diagram, cutting the core box main body diagram after the color gamut correction according to the core rectangular frame, and generating rectangular contour information of the core block area; the generating unit is used for inputting rectangular outline information of the core block area into the convolutional neural classification network to generate core types; the comparison unit is used for determining the color of the core through a comparison table database based on the category of the core; the comparison table database comprises matching relations between different core types and different core colors; and the determining unit is used for determining the layering information based on the core type and the core color.

In some alternative embodiments, the computing unit includes: the first calculating subunit is configured to obtain an error coefficient, calculate a core length based on the pixel length value and the error coefficient, and calculate a core average length based on the core length. And the second calculating subunit is used for calculating the core quality index value based on the core length.

In some alternative embodiments, the control unit comprises: the comparison subunit is used for determining the pixel points of the core block based on the rectangular outline information of the core block area, and comparing the pixel points of the core block with a standard color library to generate a predicted color; the determining subunit is used for determining the color of the core through the comparison table database based on the category of the core; and the comparison subunit is used for comparing the predicted color with the color of the core, and outputting the color of the core when the predicted color is the same as the color of the core.

Further functional descriptions of the above respective modules and units are the same as those of the above corresponding embodiments, and are not repeated here.

A core measurement device based on a bi-directional cascading pixel differential network in this embodiment is presented as a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit ) circuit, a processor and a memory that execute one or more software or firmware programs, and/or other devices that can provide the above functions.

The embodiment of the invention also provides computer equipment, which is provided with the rock core measuring device based on the bidirectional cascade pixel differential network shown in the figure 21.

Referring to fig. 22, fig. 22 is a schematic structural diagram of a computer device according to an alternative embodiment of the present invention, as shown in fig. 22, the computer device includes: one or more processors 10, memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are communicatively coupled to each other using different buses and may be mounted on a common motherboard or in other manners as desired. The processor may process instructions executing within the computer device, including instructions stored in or on memory to display graphical information of the GUI on an external input/output device, such as a display device coupled to the interface. In some alternative embodiments, multiple processors and/or multiple buses may be used, if desired, along with multiple memories and multiple memories. Also, multiple computer devices may be connected, each providing a portion of the necessary operations (e.g., as a server array, a set of blade servers, or a multiprocessor system). One processor 10 is illustrated in fig. 22.

The processor 10 may be a central processor, a network processor, or a combination thereof. The processor 10 may further include a hardware chip, among others. The hardware chip may be an application specific integrated circuit, a programmable logic device, or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general-purpose array logic, or any combination thereof.

Wherein the memory 20 stores instructions executable by the at least one processor 10 to cause the at least one processor 10 to perform a method for implementing the embodiments described above.

The memory 20 may include a storage program area that may store an operating system, at least one application program required for functions, and a storage data area; the storage data area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include high-speed random access memory, and may also include non-transitory memory, such as at least one magnetic disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, memory 20 may optionally include memory located remotely from processor 10, which may be connected to the computer device via a network. Examples of such networks include, but are not limited to, the internet, intranets, local area networks, mobile communication networks, and combinations thereof.

Memory 20 may include volatile memory, such as random access memory; the memory may also include non-volatile memory, such as flash memory, hard disk, or solid state disk; the memory 20 may also comprise a combination of the above types of memories.

The computer device further comprises input means 30 and output means 40. The processor 10, the memory 20, the input device 30 and the output device 20 may be connected by a bus or otherwise, in the figure X by way of example.

The input device 30 may receive input numeric or character information and generate key signal inputs related to user settings and function control of the computer apparatus, such as a touch screen, a keypad, a mouse, a trackpad, a touchpad, a pointer stick, one or more mouse buttons, a trackball, a joystick, and the like. The output means 40 may include a display device, auxiliary lighting means (e.g., LEDs), tactile feedback means (e.g., vibration motors), and the like. Such display devices include, but are not limited to, liquid crystal displays, light emitting diodes, displays and plasma displays. In some alternative implementations, the display device may be a touch screen.

The embodiments of the present invention also provide a computer readable storage medium, and the method according to the embodiments of the present invention described above may be implemented in hardware, firmware, or as a computer code which may be recorded on a storage medium, or as original stored in a remote storage medium or a non-transitory machine readable storage medium downloaded through a network and to be stored in a local storage medium, so that the method described herein may be stored on such software process on a storage medium using a general purpose computer, a special purpose processor, or programmable or special purpose hardware. The storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, a solid state disk or the like; further, the storage medium may also comprise a combination of memories of the kind described above. It will be appreciated that a computer, processor, microprocessor controller or programmable hardware includes a storage element that can store or receive software or computer code that, when accessed and executed by the computer, processor or hardware, implements the methods illustrated by the above embodiments.

Although embodiments of the present invention have been described in connection with the accompanying drawings, various modifications and variations may be made by those skilled in the art without departing from the spirit and scope of the invention, and such modifications and variations fall within the scope of the invention as defined by the appended claims.

Claims

1. A core measurement method based on a bidirectional cascade pixel differential network, the method comprising:

acquiring a core box main body image, inputting the core box main body image into the bidirectional cascade pixel differential network model, and generating an initial edge prediction image;

2. The method of claim 1, wherein the training an edge detection model based on the plurality of rock core images generates a bi-directional cascading pixel differential network model comprising:

and performing migration learning training on the optimal network structure based on the core edge data set to generate the bidirectional cascade pixel differential network model.

3. The method of claim 1, wherein processing the initial edge prediction graph to generate a core rectangular box comprises:

performing block segmentation processing on the edge detection map of the core box main body to generate a block edge detection map;

and generating the rectangular core frame by using a core boundary extraction algorithm based on the block edge detection diagram.

4. The method of claim 1, wherein the core feature comprises:

5. The method of claim 4, wherein the performing core measurements on the core box body map based on the core rectangular box to generate core features comprises:

Determining a pixel length value based on the core rectangular frame, and respectively calculating the core quality index value and the core average length based on the pixel length value;

performing color gamut correction on the core box main body diagram, and cutting the core box main body diagram subjected to the color gamut correction according to the core rectangular frame to generate rectangular contour information of a core block area;

inputting the rectangular outline information of the core block area into a convolutional neural classification network to generate the core type;

determining the color of the core through a comparison table database based on the core type; the comparison table database comprises matching relations between different core types and different core colors;

and determining the layering information based on the core category and the core color.

6. The method of claim 5, wherein the calculating the core quality index value and the core average length based on the pixel length values, respectively, comprises:

obtaining an error coefficient, calculating a core length based on the pixel length value and the error coefficient, and calculating the average core length based on the core length;

and calculating the core quality index value based on the core length.

7. The method of claim 5, wherein determining core color from a look-up table database based on the core category comprises:

determining the color of the core through a comparison table database based on the core type;

and comparing the predicted color with the core color, and outputting the core color when the predicted color is identical to the core color.

8. A core measurement device based on a bidirectional cascade pixel differential network, the device comprising:

9. A computer device, comprising:

a memory and a processor, the memory and the processor are in communication connection, the memory stores computer instructions, and the processor executes the computer instructions, so as to execute a core measurement method based on a bidirectional cascade pixel differential network as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, wherein the computer-readable storage medium has stored thereon computer instructions for causing a computer to perform a core measurement method based on a bi-directional cascading pixel differential network as set forth in any one of claims 1 to 7.