RU2815488C9 - Method of recognizing cracks in scanned image of boring core barrel - Google Patents

Abstract

FIELD: measuring equipment.

SUBSTANCE: invention relates to measurement equipment and relates to a method of identifying cracks in a scanned image of a boring core barrel. Method includes obtaining an image using industrial cameras and converting the imported source image from the RGB format to the HSV colour space, recognizing the characteristics of the cracks and extracting them, image conversion to RGB format and image processing with adjustable brightness, removal of the core identification label from the image, local binarization based on the image unit and morphological processing of the images.

EFFECT: high accuracy and efficiency of detecting cracks on complex backdrops.

7 cl, 7 dwg

Description

Field of technical application

The present invention relates to the field of petroleum exploration and development technologies, in particular, to a method for recognizing cracks in a developed image of a drill core core tube.

Prerequisites for creating the image

In the field of oil and gas exploration and development, fractured oil reservoirs and fractured oil and gas reservoirs account for a significant proportion of proven oil and gas reserves. A rupture in an underground environment without visible displacement under the influence of a stress field is called a crack. As an important pathway for the transportation and accumulation of oil and gas, fractures are widespread in tight sandstones. The total geological resources of tight oil reservoirs in China's main basins are about 11-13.5 billion tons, including highly productive areas that are largely affected by fracture development. On the other hand, shale oil resources have enormous potential, and an important factor affecting their production is the degree of fracture development. Thus, studying the characteristics of fracture development in reservoirs becomes especially important. Fracture recognition can be interpreted using seismic data; Guided by geological theories and laws, the theory of seismic wave propagation and the principles of seismic exploration techniques are used to obtain information about the development characteristics of cracks and faults. Seismic attribute technology is now a common fracture prediction tool that can extract attributes from a seismic data volume or other data volumes derived from a given seismic data volume to provide additional information about the change characteristics of fractures, faults, etc. On the other hand, the characteristics of fracture development in the study area can be summarized through core observations, thin-section identification, conventional logging methods and photo logging. Core observations often calculate core fractures using artificial observation data, and there is a large deviation. Accurate characterization and observation of core fractures in reservoirs is of greater importance to improve the efficiency of oil and gas reservoir development.

Crack recognition in images is very important not only in the field of oil and gas exploration, but is also applicable to varying degrees with relevant knowledge in the fields of tunnel construction, railway transport, bridge construction and medical imaging. In recent years, in the field of crack recognition based on computer vision, the research direction is mainly divided into two parts: first, based on digital image processing, artificial feature recognition is carried out, and using various laws of crack characteristics such as frequencies, edges ,HOG, grayscale, texture and entropy, some feature recognition conditions are set for crack recognition; Traditional image recognition technology mainly identifies the diseased area using the grayscale values of the image, such as Otsu method, edge detection and region growing algorithm, and has a good effect of recognizing diseased images with a simple background and large grayscale difference, but in complex backgrounds, the error rate recognition above. The general problem of traditional image recognition technology is low accuracy, high false alarm rate, inability to recognize diseases at the pixel level, Talmudic processing of images with various complex backgrounds, inability to flexibly recognize, the need to manually extract features, and the direct influence of the preprocessing method on the recognition effect. Secondly, based on deep learning, a convolutional network is created, and this network is used to automatically search for crack characteristics, so that the machine constantly self-regulates according to certain laws to achieve the effect of bringing the input data and output data closer to the label. The principle of the deep learning based method is to achieve the recognition goal by building a network model to train a large number of images. This method has a high recognition speed, but requires a large amount of data for sufficient preparation, and the constructed model is large in size, has low calculation speed and long duration, and the characteristic parameters of cracks have not yet been processed quantitatively.

Subject of the invention

To overcome the shortcomings of digital image processing, the present invention provides a method for detecting cracks by scanning a core tube, which can improve the accuracy and efficiency of extracting cracks in complex backgrounds.

To achieve the goal of accurately extracting fracture characteristics in the case of strong complex background interference, the specific steps of the method for recognizing fractures in a scanned image of a core core tube include:

Step 1: Acquire image using industrial cameras and convert imported original image from RGB format to HSV spatial domain;

step 2: crack characteristics recognition and extraction;

Step 3: Convert the image to RGB format and process the images with adjustable brightness;

step 4: remove core label; the drill core usually has an identification label, which seriously affects the recognition of fractures;

step 5: local binarization based on image block;

step 6: morphological image processing.

Description of the figures

In FIG. 1 shows a method for recognizing cracks in a developed image of a drill core tube.

In FIG. Figure 2 shows the original unwrapped image of the drill core tube.

In FIG. Figure 3 shows the image after processing the HSV color space.

In FIG. Figure 4 shows the HSV color space domain image after label removal.

In FIG. 5 shows a zonal binary image of an image block.

In FIG. Figure 6 shows a binary image of a conventional global threshold.

In FIG. 7 shows an image after processing by the method of the present invention.

Specific method for carrying out the invention

An optimization method for carrying out the present invention is described in more detail below. Although an optimization method for carrying out the present invention is described as below, it should be understood that the present invention can be implemented in various ways and should not be limited to the method of implementation described here. On the contrary, these modes of implementation are provided in order to make the present invention more transparent and complete, as well as to fully convey the scope of the present invention to those skilled in the art.

According to the process of the method for recognizing cracks in a drill core image being developed (FIG. 1) provided by the present invention, the process includes the following specific steps:

In step 1, the full core is captured and a scanned image of the core core tube is obtained using industrial cameras by rotating the core one circle. The resulting unfolded image is based on a neurophysiological perspective, with three different retinal color receptors, that is, red (R), green (G), blue (B) - the three primary colors. In the RGB color space, there are three channels R, G and B, and the range of each color channel is [0, 255]. RGB space can represent a total of 256x256x256 colors. The most commonly used color space in image processing is the RGB model, which is often used for color display and image processing. The middle axis from the origin to the white apex is a grayscale line. In traditional image processing based on RGB image processing of adjustable brightness, hue characteristics such as hue, brightness, saturation are expressed together. When extracting crack characteristics, cracks obviously have unique color tone, brightness and saturation, and applying RGB image processing is equivalent to throwing crack characteristics into color space. Therefore, to better extract the crack characteristics, the RGB image is converted into an HSV color space domain image. H (Hue) or H-channel represents a color tone, with a range of values ranging from 0-360°, and can be understood as an angle, that is, a range of values in a closed ring. S (Saturation) or S-channel represents saturation, with a range of values between [0, 1]; Saturation refers to the degree of freshness of the color, the higher the saturation, the more colorful the color of the image, and the stronger the visual effect, and conversely, the lower the saturation, the weaker the visual effect. V (Value) or V-channel represents intensity, with a range of values between [0, 1], and refers to the degree of brightness of a color, that is, intensity 0 of a color is black, and intensity 1 is white. The present invention converts an RGB image to an HSV image to process an HSV color space domain image. Conversion of an image from RGB color space to HSV color space is carried out according to the following principle:

Where, R′, G′ and B ′ are the ratio corresponding to [0, 1] after transforming the value of each pixel in the three channels, Dmax, Dmin are the maximum and minimum values in R', G' and B', E is the difference between the maximum and minimum values, and H, S and V are the hue, saturation and intensity values, respectively. At this time, the RGB image was converted into an HSV image to subsequently extract the main characteristics of the crack.

Even further, in step 2, in the drill core image being developed, the fractures are mostly white and light, and this color characteristic is used as the main characteristic of the fracture to extract the characteristics. In the original image (Figure 2) captured and imported by the industrial camera, the target crack is white and light in color, but there are complex background, black core surface, red arrow, label, artificial cracks formed by manual manipulation, and other substances, such as dirt, during the coring process; At present, the existing conventional digital image processing technology is not specialized in processing the unwrappable images of the drill core core pipe, but after conventional pre-processing, that is, histogram equalization comparison after limited adaptive histogram equalization, and the Gaussian filtering method, although the fracture characteristics are enhanced, but the red arrow, the artificial cracks in the corresponding original image are also enhanced, and the background is also complex due to the uneven color of the core surface. Using the method of the present invention, after converting an RGB image to an HSV color space image, setting the H, S, and V parameters to appropriate values, blocking and extracting the white and bright crack characteristics in the original image, extracting the HSV colors, and filling in the background that does not correspond to the crack characteristics , in black (Fig. 3), it is noticeable that the cracks can be completely extracted, avoiding the interference of artificial fake cracks, red arrow and complex background color.

Even further, in step 3, post-processing of the unwrapped HSV core tube image after color extraction is performed, and the HSV image is converted to an RGB image, and the RGB image is converted to a grayscale image, in which each point of the image is expressed in black with different saturations , and the number [0, 255] denotes the degree of "gray". An RGB image, that is, a “three-dimensional” image with three channels, has been converted into a “two-dimensional” image with one grayscale channel. And the conversion relationship of RGB value and grayscale value is shown in the following formula:

where Gray is a gray value in the range [0, 255], and R, G and B respectively represent the values in the three RGB channels. According to this formula, the R, G, B values of each pixel point are counted in turn, the grayscale values are calculated, the grayscale value for the corresponding image position is assigned, and after all the pixel points are subjected to the above process once, image processing is completed RGB adjustable brightness.

Even further, in step 4, the drill core is pasted with a label, and there will be a label left on the unfolded image, which seriously affects the recognition of fractures, and due to the similarity between the color characteristics of the label and the fractures, the label fails in HSV color space images, however the label has the correct rectangular in shape and much larger than the width of the crack. Having set a suitable sliding window ω×ω, when sliding the window towards the label, the window is completely filled with a grayscale value of 255, at which point the recognized label is retrieved, and a grayscale image of the core tube of the drill core in which the label has been removed is obtained (Fig. 4).

Even further, in step 5, zonal binarization of the resulting grayscale image is carried out based on the image block. Image binarization is a very important basic image processing method, generally used as a preprocessing technology for many image processing methods; for example, with respect to edge extraction, target recognition, shape processing, image segmentation and optical character recognition, etc., image binarization may be performed first and then post-processing may be performed. In short, binarization refers to finding a threshold in the grayscale values of all pixels in the grayscale image; if the threshold is exceeded, the gray value is assigned 1, and if the value is less than the threshold, the gray value is assigned 0, and so on by analogy with all pixels of the image to obtain a binary image. Current threshold-based binarization methods can be divided into global threshold method and local threshold method. The global thresholding method uses a fixed threshold on the entire image, comparing the grayscale values in the image to a threshold to separate the image into a background or target. This method can work well for images where the target and background pixels can be clearly distinguished, but fails to achieve the goal of distinction when the background is complex or contaminated with other clutter. To overcome the shortcomings of the global threshold method, previous scholars have proposed many local threshold methods, which no longer select a fixed threshold, but compare the pixel value at one point with the pixel value in the surrounding local neighborhood, and adaptively adjust the threshold according to the grayscale of the pixels in local neighborhood, and this pixel point will be binarized in accordance with the adaptive threshold. By comparing different scanned images of the core tube, it is known that in the HSV color space domain, in terms of color characteristics, there is still a disease with the same characteristics as cracks, but many of the interferences of the disease areas are reduced relative to RGB image processing. According to the definition of the binary image in (Fig. 4), the black area with a gray level of 0 is the background, and the white area with a gray level of 1 is the target crack and painful area. The local threshold of the image is binarized. First, the grayscale image is divided into m×n zones, and the values of m and n can be selected according to the size of the original image to obtain suitable values. A single value of the parameter μ, representing the average absolute deviation of gray values in one zone, is introduced to determine the change in gray levels of pixels in a zone. The formula is given below:

where, S is the sum of grayscale values of pixels in one zone, N is the number of pixels in each zone, the gray gradation value of pixel No. i in one zone, and ν is the average value of pixels in one zone, μ is the average absolute deviation of the gray gradation values of pixels in one zone.

Select a suitable value for ϕ , and check each block for the presence of a diseased area using ϕ . μ(m, n ) represents the mean absolute deviation of the zone block (m, n). The mean absolute deviation can represent the sharpness of the grayscale change in a zone; in the presence of cracks, there is a sharp change in grayscale from the background, the average absolute deviation has a greater value, and vice versa, in the presence of a diseased zone or a background zone in the zone block, the change in grayscale is slow, the average absolute deviation has a smaller value. Therefore, when μ(m, n) > ϕ , this indicates that this zone is the target background zone, that is, the zone with the presence of cracks; due to the fact that after extracting the color of the HSV spatial region, the area with the presence of cracks is easily recognized due to the large difference in grayscale changes between the crack and the background. Thus, Otsu binarization is used, which is a representative of global threshold binarization and an adaptive thresholding method proposed by Japanese scientist Nobuyuki Otsu in 1979. The algorithm assumes that according to the threshold, image pixels can be divided into background and target . This optimal threshold is calculated to differentiate the two types of pixels so that the two types of pixels differ as much as possible.

The principle formula of Otsu's algorithm is given as below:

Assume that the threshold for separating target and background is denoted by T, and the gray value of pixels greater than the threshold T is the target and, conversely, the background. In this case, N0, N1 is the number of pixels related to the target and background, N is the total number of pixels in zone no. (m, n), ω0, ω1 - the ratio of the number of target pixels to the total number of pixels and the ratio of the number of background pixels to the total number of pixels, respectively, τ0, τ1 - the average gray value of the target pixels and the average gray value of the background pixels, τ - the average grayscale value of the pixels of the entire zone block, - variance between classes. By entering formula (16) into formula (17), the final simplified formula of formula (18) is obtained.

When μ (m, n)≤ϕ, this means that the zone is in the background zone or background diseased zone, and the local threshold setting method is selected for threshold segmentation; There are many methods for binarizing local thresholds, but here we use the Niblack algorithm, which is a binarization algorithm based on the local characteristics of a grayscale image, with the following principle:

where, (x,y) is the pixel point of the window center w×w, k(x,y) is the sample average value in the vicinity of w×w of this point , l(xy) is the standard deviation in the vicinity of w×w of this point, β - the selection range of the correction factor is [0, 1], g(x,y) is the grayscale value of the pixels of a given point, T(x,y) is the binarization threshold of a given pixel point. The Niblack algorithm can dynamically determine thresholds in each zone to perfectly handle image grayscale unevenness.

Thus, using the local binarization method based on an image with blocks, binarization of the unfolded grayscale image of the core core tube was performed. And after binarization (Fig. 5), the diseased zone is suppressed to a certain extent. When performing direct binarization without extracting the HSV color space (Fig. 6), it can be seen that although the crack characteristics are well represented, the background noise is difficult to remove.

Even further, in step 6, morphological image processing is carried out. In the binary image (Fig. 5) with local zone block threshold obtained in the above step, it can be seen that there is still a lot of white digital noise. Therefore, the binary image of a core tube requires morphological processing, which uses digital morphology as a tool to extract image components from the image that are useful for expressing and describing the shape of zones, such as expansion, corrosion, opening and closing operations, cohesive region processing, etc. d. Corrosion can eliminate boundary points and boundary discrete points. Dilation is the expansion of the brightest parts of the image so that nearby but disconnected cracks can be combined to create one coherent area. The opening operation process is carried out first for the binary image, and the opening operation is actually the operation process for corrosion first and then expansion. Because of this, small objects, as well as individual objects in thin areas, can be eliminated to make large objects smooth and remove fine grain noise without noticeably changing the area of the object. And then a closing operation is performed to connect the places where the crack characteristics are discontinuous and fill the small voids. Finally, the area thresholds of the connected region are cleared, and the area of the crack zone in the image is visually observed, the area threshold is set, the connected region that is smaller than the area threshold is removed, and finally the morphological processing of the binary image is completed.

The method for recognizing cracks in the unfolded image of a drill core pipe has a very good effect of recognizing cracks with noticeable color characteristics. (FIG. 7) shows the final image after using the crack detection method of the present invention. The core of the present invention is that the conversion of an RGB image into an HSV color space domain image is carried out before traditional digital image processing, and by extracting a specific color, that is, the color characteristics of a crack, cracks are accurately recognized and extracted, and then the conversion is carried out to an RGB image to digitize an image. This method adds a local threshold binarization method under the image area blocks in the conventional binarization method, so that the separation of cracks and background is more obvious, and large area interference zones are removed to accurately preserve the crack. Finally, the digital noise in the binary image is suppressed using morphological processing, and finally accurate crack recognition is realized.

Claims (23)

1. A method for recognizing cracks in a developed image of a drill core pipe, characterized in that it includes: Step 1: Acquire an image using industrial cameras and convert the imported source image from RGB format to HSV color space; step 2: crack characteristics recognition and extraction; Step 3: Convert the image to RGB format and process the images with adjustable brightness; step 4: remove the core identification label from the image; step 5: local binarization based on image block; step 6: morphological image processing. 2. A method for recognizing cracks in a developing image of a drill core core pipe according to claim 1, characterized in that the core is fixed and a developing image of the core core pipe is obtained using industrial cameras, rotating the core one revolution. 3. The method for recognizing cracks in a developed image of a drill core pipe according to claim 2, characterized in that in the original image acquired and imported by the industrial camera in step 2, after converting the RGB image into an HSV color space image, the white and bright characteristics of cracks in image, and also fill the background that does not correspond to the characteristics of the cracks with black color. 4. The method for recognizing cracks in a developed image of a drill core pipe according to claim 3, characterized in that in step 3, processing the image of adjustable brightness includes converting the RGB image into an image in grayscale in accordance with the ratio: Gray=0.299×R+0.587×G+0.114×B, where Gray is a gray value in the range [0, 255], and R, G and B respectively represent the values in the three RGB channels; at the same time, the R, G, B values of each pixel are counted in turn, the gray gradation values are calculated, and the gray gradation value is assigned for the corresponding image position. 5. A method for recognizing cracks in a developed image of a drill core core pipe according to claim 4, characterized in that in step 4 a sliding window of size ω×ω is installed; when sliding the window towards the label, the window is completely filled with a grayscale value of 255 and an image is obtained in grayscale the core tube of a drill core in which the label has been removed. 6. A method for recognizing cracks in a developed image of a drill core pipe according to claim 5, characterized in that in step 5, zonal binarization of the resulting image is carried out in grayscale based on image blocks; select a threshold value ϕ, and check each block using the value ϕ and the value μ(m, n), which represents the average absolute deviation of the block of the zone (m, n) and characterizes the sharpness of the change in gray levels in the zone; in this case, the value μ(m,n)>ϕ indicates the presence of cracks, and the value μ(m,n)≤ϕ indicates that the zone is in the background zone or background interference zone; In this case, binarization is used according to the Nobuyuki Otsu method: where N ₀ , N ₁ is the number of pixels related to the target and background, N is the total number of pixels in the zone (m, n), ω0, ω1 is the ratio of the number of target pixels to the total number of pixels and the ratio of the number of background pixels to the total number of pixels respectively, τ0 and τ1 are the average gray gradation value of the target pixels and the average gray gradation value of the background pixels, τ is the average gray gradation value of the pixels of the entire zone block, - variance between classes; To segment thresholds, the Niblack algorithm is used based on local characteristics of the grayscale image: where, (x,y) is the pixel point of the center of the window ω×ω, k(x,y) is the sample average value in the neighborhood w×w of this point, l(x,y) is the standard deviation in the neighborhood w×w of this point , β is the selection range of the correction factor in [0, 1], g (x,y) is the grayscale value of the pixels of a given point, T(x,y) is the binarization threshold of a given pixel point. 7. The method for recognizing cracks in a developed image of a drill core pipe according to claim 6, characterized in that the morphological processing includes suppressing digital noise in the binary image using morphological processing.