Method and system for automatic identification of logging rock cuttings

Abstract

Embodiments of the present invention relate to the technical field of logging rock cuttings detection. Provided are a method and system for automatic identification of logging rock cuttings. The method comprises: collecting a rock cuttings sample, and performing sample preparation on the rock cuttings sample to obtain rock cuttings under detection; separately collecting a white-light dry image and a fluorescence image of the rock cuttings under detection, and performing image registration on the fluorescence image on the basis of the white-light dry image; on the basis of the white-light dry image, performing rock cuttings instance segmentation result identification to obtain a first identification result; on the basis of the registered fluorescence image, performing rock cuttings fluorescence imaging result identification to obtain a second identification result; and performing outlier removal on the first identification result and the second identification result, and outputting a detection result. The solution of the present invention solves the problems of low detection efficiency and low detection accuracy of existing logging rock cuttings detection schemes.

Description

Automatic identification method and system for logging cuttings

[0001] Cross-reference to related applications

[0002] This application claims the benefit of Chinese Patent Application No. 202411185666.9, filed on August 27, 2024, the contents of which are incorporated herein by reference. Technical Field

[0003] This invention relates to the field of drilling cuttings detection technology, specifically to an automatic method and system for identifying drilling cuttings. Background Technology

[0004] During oil drilling, drill bits forcefully impact underground rocks, breaking them into cuttings. These cuttings are carried to the surface with the circulating drilling fluid and then filtered through a drilling fluid vibrating screen to separate the cuttings samples needed for research and analysis. These filtered cuttings samples are processed in two parts: one part is collected and cleaned by geological extraction workers, while the other part is discharged into a wastewater pond as waste. The cuttings samples processed by geological extraction workers are used for recording and analyzing cuttings logging data. For a long time, cuttings collection was mainly done manually. Workers needed to circulate drilling mud carrying cuttings to the surface, then manually collect and clean the cuttings to ensure that the drilling mud on the surface of the cuttings was washed away. The collected cuttings were then dried for analysis. However, this manual operation was extremely labor-intensive, and the quality stability of the cuttings was poor, easily leading to problems such as missed or incorrect collection during the cuttings collection process, thus affecting the accuracy and completeness of geological data.

[0005] Currently, image recognition of rock cuttings still relies on traditional methods. At the drilling site, logging personnel visually assess the rock cuttings using white light and fluorescence, then perform information identification, processing, and reporting. This entire process continues throughout the drilling operation. With the continuous development of drilling technology, especially the widespread use of PDC drill bits, drilling speeds have increased significantly, resulting in finer rock cuttings. This places higher demands on the skills and physical strength of logging personnel, making manual logging operations more challenging and demanding. Due to the high-intensity continuous work, the manual assessment and identification of rock cuttings using white light and fluorescence is easily affected, thus impacting the accuracy of the final logging results. Therefore, it can be inferred that problems may arise in the continuity and accuracy of sample collection, as well as the reliability of particle size identification and characterization. To address the low detection efficiency and low accuracy of existing logging rock cuttings detection schemes, a new logging rock cuttings detection scheme is needed. Summary of the Invention

[0006] The purpose of this invention is to provide an automatic logging cuttings identification method and system to at least solve the problems of low detection efficiency and low detection accuracy in existing logging cuttings detection schemes.

[0007] To achieve the above objectives, the first aspect of the present invention provides an automatic identification method for logging cuttings, the method comprising: collecting cuttings samples and performing sample preparation processing on the cuttings samples to obtain detection cuttings; acquiring white light dry image and fluorescence image of the detection cuttings respectively, and performing image registration on the fluorescence image based on the white light dry image; performing cuttings instance segmentation result identification based on the white light dry image to obtain a first identification result; performing cuttings fluorescence imaging result identification based on the registered fluorescence image to obtain a second identification result; performing anomaly removal processing on the first identification result and the second identification result, and outputting the detection result.

[0008] Optionally, the step of performing sample preparation processing on the rock cuttings sample to obtain test rock cuttings includes: performing a cleaning process on the rock cuttings sample, and performing a leveling process on the cleaned rock cuttings sample; performing a drying process on the leveled rock cuttings sample, and using the dried rock cuttings sample as the test rock cuttings.

[0009] Optionally, the step of performing image registration on the fluorescence image based on the white light interference image includes: calculating single-channel grayscale images of the white light interference image and the fluorescence image respectively to obtain a grayscale white light image and a grayscale fluorescence image respectively; cropping the grayscale fluorescence image by grayscale values, retaining only pixels with grayscale values ​​at the top of a preset percentage in the grayscale fluorescence image; calculating the color distribution histograms of the grayscale white light image and the grayscale fluorescence image respectively to obtain the color distribution histograms of the grayscale white light image and the grayscale fluorescence image respectively; performing color histogram matching on the grayscale fluorescence image after grayscale value cropping, using the grayscale white light image as a reference, based on the color distribution histograms of the grayscale white light image and the grayscale fluorescence image; performing SIFT feature extraction on the grayscale white light image and the grayscale fluorescence image, and performing fluorescence image registration based on the extracted SIFT features, mapping each pixel in the fluorescence image to the same coordinate system as the grayscale white light image to obtain the registered fluorescence image.

[0010] Optionally, the step of performing SIFT feature extraction on the grayscale white light image and the grayscale fluorescence image includes: calculating key points in the image based on the Harris corner detection algorithm in the grayscale white light image and the grayscale fluorescence image respectively; calculating the edge orientation and intensity of the neighboring window centered on each key point, and calculating the SIFT feature descriptor for each key point; the mapping relationship between each pixel in the fluorescence image and the same coordinate system as the grayscale white light image is as follows:

[0011] wherein, are the coordinates of the fluorescence image; [x, y] are the coordinates of the white light image; H′ is the inverse transformation matrix of the preset offset parameter.

[0012] Optionally, performing recognition of the cuttings instance segmentation result based on the white light dry illumination image to obtain a first recognition result includes: sequentially performing superpixel extraction, superpixel color similarity merging, and superpixel morphological similarity on the white light dry illumination image to obtain the white light dry illumination image after instance segmentation; performing analysis of each first recognition target based on the white light dry illumination image after instance segmentation respectively to obtain the analysis results of each first recognition target as the first recognition result; wherein, the first recognition result includes: the white light color recognition result of the cuttings, the morphological distribution result of the cuttings, the size distribution result of the cuttings, and the naming result of the cuttings.

[0013] Optionally, the execution rule for performing superpixel extraction on the white light dry illumination image is: performing superpixel segmentation on the white light dry illumination image based on the graph-based image segmentation algorithm to obtain multiple superpixels as the superpixel extraction result; the execution rule for performing superpixel color similarity merging on the white light dry illumination image is: calculating the color similarity of each adjacent superpixel respectively to obtain a color similarity set; selecting the group of color similarities with the largest preset number of color similarities in the color similarity set, and merging the adjacent superpixels corresponding to the selected group of color similarities to obtain the superpixel color similarity merging result; wherein, the calculation rule for the color similarity of each adjacent superpixel is:

[0014] (i < j and S i and S j are adjacent)

[0015] wherein, S i and S j are two adjacent superpixels; ColorSim i,j is the color similarity between superpixel S i and superpixel S j ; h i is the vector obtained by splicing the color histograms of superpixel S i ; h j is the vector obtained by splicing the color histograms of superpixel S j

[0016] Optionally, the execution rule for performing superpixel morphological similarity merging on the white light dry illumination image is: calculating the boundaries of each superpixel respectively, and performing morphological similarity between each adjacent superpixel based on the boundaries of each superpixel, and the calculation rule is:

[0017] i < j and Si and S j Adjacency

[0018] Among them, hapeSim i,j For Superpixel S i With SuperPixel S j Morphological similarity between them; S i，ymin ,S i，ymax ,S i，xmin ,S i，xmax Superpixel S i Four boundaries facing each other; S j，ymin ,S j，ymax ,S j，xmin ,S j，xmax Don't be superpixel S j Four boundaries facing each other; select the morphological similarity group with the highest morphological similarity from the preset number of morphological similarity groups in the morphological similarity set, and merge the adjacent superpixels corresponding to the morphological similarity groups obtained by the filtering to obtain the superpixel morphological similarity merging result.

[0019] Optionally, the recognition rule for the white light color recognition result of the rock debris is as follows: extract the RGB values ​​of each pixel in the white light dry image after instance segmentation, and perform 3-sigma anomaly removal processing on the distribution of each color channel of R, G, and B respectively; for the remaining pixels after anomaly removal processing, search for the closest color based on the preset white light color table, and use it as the color recognition result of the corresponding pixel point; based on the color recognition results of all remaining pixels after anomaly removal processing, obtain the white light color recognition result of the rock debris.

[0020] Optionally, the identification rules for the rock debris morphology distribution results and the rock debris size distribution results are as follows: calculate the eccentricity of the minimum circumscribed ellipse of each rock debris instance in the white light dry image after instance segmentation, and perform the corresponding rock debris instance morphology distribution result description based on the calculated eccentricity; wherein, the smaller the eccentricity, the closer the rock debris is to a circle, and the larger the eccentricity, the finer and longer the rock debris; calculate the proportion of image pixels occupied by each rock debris instance in the white light dry image after instance segmentation, and perform the rock debris size distribution result description for each rock debris instance based on the ratio between the pre-camera viewports of the proportion of image pixels occupied.

[0021] Optionally, the identification rule for the rock debris naming result is as follows: Based on the linear spectral clustering superpixel segmentation algorithm, superpixel segmentation is performed on each rock debris instance in the white light dry image after instance segmentation. Regions with sizes larger than a preset size threshold and eccentricities greater than a preset eccentricity threshold are filtered out based on the segmentation results, and the remaining superpixels are used as identification samples. The particle size of the remaining superpixels is calculated, and the particle size distribution rule and average particle size are statistically analyzed based on the particle size. Based on the particle size distribution rule and the average particle size, corresponding rock debris features are matched in a pre-constructed rock debris knowledge graph database, and the name of the rock debris with the highest matching degree is used as the current rock debris naming result.

[0022] Optionally, the second recognition result includes: rock fragment fluorescence color recognition result and rock fragment fluorescence intensity recognition result; the recognition rule for the rock fragment fluorescence color recognition result is as follows: extract the RGB values ​​of each pixel in the registered fluorescence image, and perform 3-sigma anomaly removal processing on the distribution of each color channel of R, G, and B respectively; for the remaining pixels after anomaly removal processing, search for the closest color based on a preset fluorescence color table, and use it as the color recognition result of the corresponding pixel point; based on the color recognition results of all remaining pixels after anomaly removal processing, obtain the fluorescence color recognition result.

[0023] Optionally, the identification rule for the rock cutting fluorescence intensity identification result is as follows: identify the region containing fluorescence in the registered fluorescence image, and calculate the fluorescence intensity based on the RGB values ​​of all pixels to obtain a single-channel grayscale image; wherein, the grayscale value of the single-channel grayscale image is the fluorescence intensity; based on the fluorescence color identification result, filter out regions that are not the current fluorescence color, and statistically analyze the fluorescence intensity distribution histogram of the remaining regions as the rock cutting fluorescence intensity identification result.

[0024] Optionally, the step of performing anomaly removal processing on the first identification result and the second identification result and outputting the detection result includes: obtaining feature sets of the first identification result and the second identification result respectively, obtaining a first identification result feature set and a second identification result feature set; and constructing normal distribution models of the first identification result feature set and the second identification result feature set respectively, expressed as:

[0025] Where, x i ∈X,X={x1,…x N} represents the first or second recognition result feature set; N represents the number of features in the first or second recognition result feature set; μ represents the normal distribution value corresponding to the i-th first or second recognition result feature; features that deviate from the preset proportion range of the normal distribution center are filtered as outliers to obtain the detection result after anomaly removal.

[0026] A second aspect of the present invention provides an automatic logging cuttings identification system, the system comprising: a collection unit for collecting cuttings samples and performing sample preparation processing on the cuttings samples to obtain detection cuttings; a registration unit for collecting white light dry image and fluorescence image of the detection cuttings respectively, and performing image registration on the fluorescence image based on the white light dry image; an identification unit for: performing cuttings instance segmentation result identification based on the white light dry image to obtain a first identification result; performing cuttings fluorescence imaging result identification based on the registered fluorescence image to obtain a second identification result; and an anomaly removal unit for performing anomaly removal processing on the first identification result and the second identification result, and outputting detection results.

[0027] Optionally, the acquisition unit includes: a rock cuttings washing module for collecting rock cuttings samples into a sampler and washing the rock cuttings samples; a rock cuttings leveling module for leveling the washed rock cuttings samples; a rock cuttings drying module for drying the leveled rock cuttings samples; and a rock cuttings conveying module for connecting the rock cuttings washing module, the rock cuttings leveling module, and the rock cuttings drying module to realize the transfer of rock cuttings samples.

[0028] On the other hand, the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described automatic logging cuttings identification method.

[0029] Through the above technical solution, the present invention collects rock cuttings samples and performs sample preparation processing on the samples to obtain detectable rock cuttings, realizing automated sample preparation without manual preparation and improving sample preparation efficiency. It also ensures overall detection efficiency. Furthermore, the present invention simultaneously acquires white light dry illumination images and fluorescence images, enabling multi-parameter detection of rock cuttings and ensuring detection accuracy.

[0030] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0031] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings:

[0032] Figure 1 is a flowchart of the steps of an automatic logging cuttings identification method provided in one embodiment of the present invention;

[0033] Figure 2 is a flowchart of the identification process of the first identification result provided in one embodiment of the present invention;

[0034] Figure 3 is a system structure diagram of an automatic logging cuttings identification system provided in one embodiment of the present invention;

[0035] Figure 4 is a schematic diagram of the acquisition unit provided in one embodiment of the present invention. Detailed Implementation

[0036] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0037] Figure 1 is a flowchart of an automatic logging cuttings identification method according to an embodiment of the present invention. As shown in Figure 1, an embodiment of the present invention provides an automatic logging cuttings identification method, the method comprising:

[0038] Step S10: Collect rock cuttings samples and perform sample preparation processing on the rock cuttings samples to obtain test rock cuttings.

[0039] Specifically, the sample preparation process for the rock cuttings sample to obtain the test rock cuttings includes: cleaning the rock cuttings sample and leveling the cleaned rock cuttings sample; drying the leveled rock cuttings sample and using the dried rock cuttings sample as the test rock cuttings.

[0040] In this embodiment of the invention, the processing and analysis of rock cuttings samples are crucial in the field of oil drilling, as rock cuttings carry information about underground rocks and can provide key data for geological exploration and oilfield development. To improve the efficiency and accuracy of rock cuttings sample processing, modern technology has begun to play a significant role in the process. A key technology is the cleaning of rock cuttings samples. Traditional manual cleaning methods suffer from high labor intensity, low efficiency, and inconsistent quality. Introducing automated cleaning equipment can greatly improve cleaning efficiency and quality stability. These devices can thoroughly and precisely clean rock cuttings samples according to preset cleaning programs, ensuring that mud and impurities on the surface of the rock cuttings are effectively removed, thereby reducing human error and improving data accuracy.

[0041] Furthermore, after cleaning, leveling the rock cuttings samples is a crucial step. Leveling makes the surface of the rock cuttings samples smoother, which is beneficial for subsequent observation and analysis. Modern rock cuttings processing equipment can achieve automatic leveling, ensuring the smoothness and consistency of the rock cuttings sample surface, thereby improving the accuracy and reliability of subsequent analysis.

[0042] Furthermore, the cleaned and leveled rock cuttings samples undergo drying. Drying removes moisture from the rock cuttings, making them drier and more stable. Traditional drying methods may suffer from uneven temperature and inaccurate time, affecting the quality of the rock cuttings samples. Modern drying equipment, however, can precisely control temperature and time, ensuring that the rock cuttings samples are processed under optimal drying conditions, thereby guaranteeing their quality and stability.

[0043] Step S20: Acquire white light dry image and fluorescence image of the detected rock debris respectively, and perform image registration on the fluorescence image based on the white light dry image.

[0044] Specifically, although the white light and fluorescence images captured by the device originate from the same lens, slight misalignment of the lens or tray may cause misalignment between the two images. Therefore, image registration is required to align the physical coordinates of the fluorescence image to the physical coordinates of the white light image. The output of this step is the registered rock debris fluorescence image. Calculate the single-channel grayscale images of the white light interference image and the fluorescence image separately to obtain grayscale white light image and grayscale fluorescence image respectively; crop the grayscale fluorescence image by retaining only pixels with grayscale values ​​at the top of a preset percentage; calculate the color distribution histograms of the grayscale white light image and the grayscale fluorescence image respectively; based on the color distribution histograms of the grayscale white light image and the grayscale fluorescence image, and using the grayscale white light image as a reference, perform color histogram matching on the grayscale fluorescence image after grayscale value cropping; perform SIFT feature extraction on the grayscale white light image and the grayscale fluorescence image, and perform fluorescence image registration based on the extracted SIFT features, mapping each pixel in the fluorescence image to the same coordinate system as the grayscale white light image to obtain the registered fluorescence image.

[0045] In one possible implementation, white light images, captured under sufficient lighting conditions, generally have high contrast and contain relatively clear details. In contrast, fluorescence images, captured under low light conditions, have very low contrast and poor detail. Therefore, by using the white light image as a reference and performing image color histogram matching on the fluorescence image, features similar to those in the white light image can be recovered from the fluorescence image. Single-channel grayscale images of both the white light and fluorescence images are calculated. Both white light and fluorescence images can generally be processed into RGB three-channel, 256-color images. The grayscale value for each pixel is calculated using an empirical formula based on photometric intensity. The formula is as follows:

[0046] Where R, G, and B are integers from 0 to 255, and the calculated gray value I is the weighted sum rounded down.

[0047] Furthermore, the grayscale fluorescence image is cropped. The pixels with the highest grayscale values ​​(m%) in the grayscale fluorescence image are selected, and their minimum grayscale value is denoted as τ. The grayscale values ​​of all pixels in the grayscale fluorescence image with grayscale values ​​greater than τ are then set to τ. The purpose of this step is to reduce the interference of strong fluorescence on the image contrast balance. In practice, m can be set according to experiments. The recommended value for this invention is m% = 0.1%.

[0048] Further, calculate the color distribution histograms for grayscale white light and grayscale fluorescence images separately. This step involves statistically analyzing the distribution of each pixel's grayscale value within the range of 0 to 255, and denoting the color distribution histograms for grayscale white light and grayscale fluorescence as H. bright and H flu These are 256-dimensional vectors. Then calculate H. bright and H flu The cumulative density distribution is C bright and C flu , which are also 256-dimensional vectors respectively.

[0049] Furthermore, using the grayscale white light image as a reference, color histogram matching is performed on the grayscale fluorescence image after grayscale value cropping. First, for C... bright and C flu Interpolation is performed, treating it as an invertible mapping from the real number interval [0, 256) to the interval [0, 1]. Then, the mapping is performed on each pixel in the grayscale fluorescence image:

[0050] Where I is the original gray value of the pixel in the gray-scale fluorescence image, and I′ is the mapped gray value.

[0051] Preferably, the step of performing SIFT feature extraction on the grayscale white light image and the grayscale fluorescence image includes: calculating key points in the image based on the Harris corner detection algorithm in the grayscale white light image and the grayscale fluorescence image respectively; calculating the edge orientation and intensity of the neighboring window centered on each key point, and calculating the SIFT feature descriptor for each key point; the mapping relationship between each pixel in the fluorescence image and the same coordinate system as the grayscale white light image is as follows:

[0052] in, [x, y] represents the coordinates of the fluorescence image; [x, y] represents the coordinates of the white light image; H′ is the inverse transformation matrix with preset offset parameters.

[0053] In one possible implementation, the image coordinate offset model used in this invention is a non-reflective similarity transformation, which considers coordinate offsets caused by lens rotation, scale, and translation. Its transformation matrix (i.e., the parameters to be estimated) is:

[0054] That is, from the coordinates [x,y] of the white light image to the coordinates of the fluorescence image. The mapping method is as follows:

[0055] Its inverse transformation matrix is:

[0056] From the coordinates of the fluorescence image The mapping method to the coordinates [x, y] of the white light image is as follows:

[0057] The SIFT feature extraction and matching steps generated a set of paired keypoints from the white light and fluorescence images:

[0058] By minimizing its minimum mean square error, its optimal coordinate offset parameters can be estimated.

[0059] in, and [x i ,y i They are respectively and The coordinates.

[0060] After estimating the left offset parameters, each pixel in the fluorescence image is mapped to a new coordinate system using H′. Then, the image is filled using methods such as bilinear interpolation to obtain the registered fluorescence image.

[0061] Step S30: Based on the white light dry image, perform rock debris instance segmentation result recognition to obtain the first recognition result.

[0062] Specifically, superpixel extraction, superpixel color similarity merging, and superpixel morphological similarity are sequentially performed on the white light dry image to obtain a white light dry image after instance segmentation; based on the white light dry image after instance segmentation, each first identification target analysis is performed to obtain the analysis results of each first identification target, which are used as the first identification results; wherein, the first identification results include: rock debris white light color identification results, rock debris morphological distribution results, rock debris size distribution results, and rock debris naming results.

[0063] Preferably, to analyze the size, shape, and color of each rock fragment, it is necessary to identify the pixel range of each particle in the image, i.e., to perform instance segmentation. The output of this step is the image segmentation result for each rock fragment in the image. This technique mainly includes three steps: superpixel extraction from white light-illuminated rock fragment images, superpixel color similarity merging, and superpixel morphological similarity merging. Based on real data sampling and estimation, after using the present invention to perform instance segmentation on white light-illuminated rock fragment images, the accuracy (the proportion of correct segmentation results) of rock fragment instance segmentation is 98.4%, and the recall (the proportion of real rock fragments that are correctly segmented) is 99.1%. The specific implementation process of the technique is as follows:

[0064] Superpixel extraction: First, the image is segmented into superpixels using an efficient graph-based image segmentation method, including the following steps:

[0065] 1) Calculate the dissimilarity of each pixel in the image to its 8 neighboring pixels. The dissimilarity is calculated using the Euclidean distance of the RGB three-channel pixel values. (The last sentence appears to be incomplete and requires further context.) i With a neighbor v j Let the connection of be denoted as an edge e in graph G, and let the weight of the edge be the dissimilarity. Let v be the distance between each edge. i Assign a flag, initially Id(v) i ).

[0066] 2) Sort all edges of all graphs G in ascending order of dissimilarity to obtain E = {e1, e2, ..., e}. N}

[0067] 3) Find the edge e with the smallest weight in E. <v i ,v j > If the merge condition Id(v) is met i )≠Id(v j And e has a dissimilarity greater than Id(v) i ), Id(v j If there is no internal dissimilarity, then proceed to step 4).

[0068] 4) Set Id(v) j ) is assigned the value Id(v i Update the dissimilarity threshold for this class. (e) i Remove from E.

[0069] 5) If the number of elements in E is greater than or equal to the preset number of instances, proceed to step 3); otherwise, end.

[0070] The result after that is a set composed of multiple pixels, which is the image segmentation result of this method and is called a superpixel.

[0071] Furthermore, for superpixel color similarity merging: As can be seen from the result of superpixel extraction, after instance segmentation of the white-light dry illumination rock debris image, there is a serious over-segmentation phenomenon, that is, the same rock debris instance is segmented into several superpixels. Therefore, the solution of this invention uses a superpixel merging method based on color similarity to post-process the segmentation result of the efficient graph-based image segmentation method and merge the superpixels belonging to the same rock debris. It includes the following steps:

[0072] 1) For each superpixel calculated in the superpixel extraction step, calculate the color distribution feature. The color distribution feature is to statistically calculate the RGB three-channel colors of all pixels in the area covered by the superpixel, obtain the distribution histogram and normalize it:

[0073] Among them, refers to the number of pixels with color c (one of R, G, B) and non-pixel value k (0 ≤ k ≤ 255) in the superpixel. The final distribution histogram is a vector obtained by splicing the color histograms of the three channels

[0074] 2) Calculate the color similarity for each pair of adjacent superpixels S i and S j respectively:

[0075] (i < j and S i is adjacent to S j )

[0076] Suppose that a total of M1 superpixels are extracted in the superpixel extraction step, and a total of P1 pairs of adjacent superpixels calculate the color similarity. According to a preset threshold τ color (0 ≤ τ color < 1), select the largest groups from the P1 groups of color similarities. According to the selected combinations of superpixels with high similarities, perform superpixel merging respectively. Superpixel merging is to take the union of S i and S j and add it to the superpixel segmentation result, and remove S i and S j . Suppose that after superpixel merging, there are M2 (M2 ≤ M1) remaining superpixels.

[0077] Furthermore, superpixel morphological similarity merging: Although there are certain color differences between some adjacent superpixels, they are morphologically inclusive of each other and essentially belong to different parts of the same rock debris. Therefore, the solution of the present invention uses a superpixel merging method based on morphological similarity to merge superpixels with different colors that belong to the same rock debris. The method includes the following steps:

[0078] 1) Based on each calculated superpixel S, calculate the upper, lower, left, and right boundaries S ymin ,S ymax ,S xmin ,S xmax .

[0079] 2) For each pair of adjacent superpixels S i and S j calculate their morphological similarity:

[0080] i < j and S i and S j are adjacent

[0081] Similar to the superpixel color similarity merging, assuming that there are M2 remaining superpixels after the superpixel color similarity merging, there are P2 pairs of adjacent superpixels. According to a preset threshold τ shape (0 ≤ τ shape < 1), select the combinations in the P2 groups with a morphological similarity greater than τ shape and merge these superpixels. Assume that there are M3 remaining superpixels after the merging.

[0082] The first recognition result includes: the recognition result of the white light color of the rock debris, the morphological distribution result of the rock debris, the size distribution result of the rock debris, and the naming result of the rock debris. As shown in Figure 2, it includes the following steps:

[0083] Step S301: Perform the recognition of the white light color of the rock debris.

[0084] Specifically, extract the RGB values of each pixel in the white light dry illumination image after instance segmentation, and perform 3-sigma anomaly removal processing on the distribution of each color channel of R, G, and B respectively; for the remaining pixels after the anomaly removal processing, search for the closest color based on a preset white light color table as the color recognition result of the corresponding pixel point, and obtain the recognition result of the white light color of the rock debris based on the color recognition results of all the remaining pixels after the anomaly removal processing.

[0085] In one possible implementation, based on the rock debris instance segmentation result and the current color table, the main color of each rock debris is extracted and matched against the color table. The final output is the overall color recognition result. After obtaining the rock debris instance segmentation result from the white light interfering image rock debris instance segmentation step, color filtering is performed on each rock debris instance, and the corresponding color name is matched against the color table. The RGB values ​​of each pixel in the rock debris instance (white light or fluorescence) are extracted, and 3-sigma anomaly removal processing is performed on the distribution of each color channel (R, G, B). That is, the mean μ and standard deviation σ of the sample distribution of each color channel are calculated respectively. If the pixel value of a certain pixel in this channel is not within the interval [μ-3σ, μ+3σ], then this pixel is removed from the sample. For the pixels remaining after color filtering, the closest color is searched in the color table. The RGB values ​​are treated as vectors, and the distance between RGB values ​​is measured by the vector L1 distance. For the RGB value of the i-th pixel (R... i G i B i ) and a color table of length L {(N j ,(R j G j B j ))|1≤j≤L}(where N j (The color name is N), the color name of this pixel. k for:

[0086] After actual identification and processing, the RBG values ​​of each pixel in the white rock debris example shown in the image below were extracted. After 3-sigma anomaly removal, 86% of the pixel area was retained. After color table matching, the color name of each retained pixel can be determined, of which 80% is gray, 5% is black, and 1% is light gray.

[0087] Preferably, the present invention further includes updating the color table. If the RGB values ​​of the colors extracted from the current image do not exist in the current color table, the user can name them and update them to the corresponding color table. The updated color table will take effect during the color matching process of the current and subsequent images. The user can choose to update the color table based on the color matching results. The user can select several pixels and their RGB values ​​(for example, a pixel extracted from the image with RGB values ​​of (73, 63, 70)), and enter the corresponding color name (for example, "dark gray"), which will then be added to the color table, and color matching will be performed again, and this updated color table will be applied to other subsequent images.

[0088] Step S302: Perform rock debris morphology distribution identification.

[0089] Specifically, the eccentricity of the minimum bounding ellipse containing each rock fragment instance in the segmented white-light dry image is calculated. Based on the calculated eccentricity, the corresponding morphological distribution of the rock fragment instance is described. The smaller the eccentricity, the closer the rock fragment is to a circle; the larger the eccentricity, the finer and longer the rock fragment. The calculation rules are as follows:

[0090] Step S303: Perform rock cuttings size distribution identification.

[0091] Specifically, the proportion of image pixels occupied by each rock fragment instance in the segmented white light image is calculated. Based on the ratio of this proportion to the camera's field of view, the size distribution of each rock fragment instance is described. Since the camera's field of view is known and fixed, the cross-sectional area of ​​the rock fragment is estimated by dividing the proportion of pixels occupied by each rock fragment instance by the camera's field of view, thereby estimating its size. For example, if the proportion of image pixels occupied by the rock fragment is 0.1%, and the camera's field of view is 30mm × 40mm, then the estimated cross-sectional area of ​​the rock fragment is 1.2mm. 2 .

[0092] Step S304: Perform rock cuttings naming.

[0093] Specifically, based on the linear spectral clustering superpixel segmentation algorithm, superpixel segmentation is performed on each rock debris instance in the white light dry image after instance segmentation. Based on the segmentation results, regions with sizes larger than a preset size threshold and eccentricities larger than a preset eccentricity threshold are filtered out, and the remaining superpixels are used as recognition samples. The particle size of the remaining superpixels is calculated, and the particle size distribution rules and the average particle size are statistically analyzed based on the particle size. Based on the particle size distribution rules and the average particle size, corresponding rock debris features are matched in a pre-constructed rock debris knowledge graph database, and the name of the rock debris with the highest matching degree is used as the current rock debris naming result.

[0094] In this embodiment of the invention, the naming of rock fragments as coarse sand, fine sand, medium sand, silt, etc. mainly depends on the distribution range of particle size in the rock fragments. Therefore, it is necessary to further divide the individual rock fragments into particle instances and count the particle size distribution range, and name them according to empirical rules.

[0095] Rock cuttings contain a large number of fine particles, which are generally similar in size; however, due to the limitations of camera resolution, the edges of some particles are not clearly defined. Therefore, it is only necessary to identify a portion of particles with distinct boundaries to estimate the overall particle size distribution.

[0096] This technique first uses a linear spectral clustering superpixel segmentation algorithm to segment individual rock debris images into superpixels. Then, it filters out regions with large sizes or high eccentricity. Finally, it statistically analyzes the remaining superpixel sizes as samples for grain size distribution. The steps include:

[0097] 1) Represent each pixel p in the image as a vector. <l p ,α p ,β p ,x p ,y p > where l, α, and β are the pixel values ​​of the pixel in the LAB color space, and x and y are the horizontal and vertical coordinates of the pixel.

[0098] 2) Randomly sample K pixels M = {m} from the image. k = <l k ,α k ,β k ,x k ,y k >|k∈{1,…,K}}. In this scheme, it is recommended to set this to 10% of the total number of pixels.

[0099] 3) For each pixel p, set its label L(p) to the center with the nearest Euclidean distance:

[0100] (d) For each center m k For ∈M, update to the mean of the corresponding label pixels:

[0101] Where C k ={p|L(p)=k}

[0102] 4) If the cluster center has changed compared to the previous round, repeat step 3); otherwise, treat pixels with the same label as belonging to the same superpixel, output K superpixels and end.

[0103] After segmentation, the size and eccentricity of each superpixel can be calculated using the same method as in the shape / size recognition step. Since rock fragments are essentially the smallest components of rock debris and are generally well-rounded, the largest 70% of superpixels and superpixels with an eccentricity greater than 0.8 can be filtered out. Finally, the particle size distribution is calculated based on the camera's field of view, and the average particle size is calculated. The rock fragments are then named by referring to the corresponding national or industry standards.

[0104] Step S40: Based on the registered fluorescence image, perform rock debris fluorescence imaging result recognition to obtain a second recognition result.

[0105] Specifically, the second identification result includes: rock fragment fluorescence color identification result and rock fragment fluorescence intensity identification result; the identification rule for the rock fragment fluorescence color identification result is as follows: extract the RGB values ​​of each pixel in the registered fluorescence image, and perform 3-sigma anomaly removal processing on the distribution of each color channel of R, G, and B respectively; for the remaining pixels after anomaly removal processing, search for the closest color based on the preset fluorescence color table, and use it as the color identification result of the corresponding pixel point; based on the color identification results of all remaining pixels after anomaly removal processing, obtain the fluorescence color identification result.

[0106] Preferably, the identification rule for the rock cutting fluorescence intensity identification result is as follows: identify the region containing fluorescence in the registered fluorescence image, and calculate the fluorescence intensity based on the RGB values ​​of all pixels to obtain a single-channel grayscale image; wherein, the grayscale value of the single-channel grayscale image is the fluorescence intensity; based on the fluorescence color identification result, filter out regions that are not the current fluorescence color, and statistically analyze the fluorescence intensity distribution histogram of the remaining regions as the rock cutting fluorescence intensity identification result.

[0107] In this embodiment of the invention, the fluorescence intensity of the rock debris fluorescence image is calculated based on the RGB values ​​of all pixels. The calculation method refers to the luminance (I) calculation formula in the CCIR 6011 international standard: I = 0.299*R + 0.587*G + 0.114*B

[0108] This results in a single-channel grayscale image, where the grayscale value is the fluorescence intensity.

[0109] Then, based on the current fluorescence color of the device and the fluorescence color recognition results of the rock fragments, areas that are not the current fluorescence color are filtered out. For example, if the current fluorescence color is yellow, all non-yellow pixels are excluded from the analysis range.

[0110] Step S50: Perform anomaly removal processing on the first identification result and the second identification result, and output the detection result.

[0111] Specifically, feature sets of the first recognition result and the second recognition result are obtained respectively, resulting in a first recognition result feature set and a second recognition result feature set; normal distribution models of the first recognition result feature set and the second recognition result feature set are constructed respectively, as follows:

[0112] Where, x i ∈X,X={x1,…x N} represents the first or second recognition result feature set; N represents the number of features in the first or second recognition result feature set; μ represents the normal distribution value corresponding to the i-th first or second recognition result feature; features that deviate from the preset proportion range of the normal distribution center are filtered as outliers to obtain the detection result after anomaly removal.

[0113] In one possible implementation, assume that a sample of some feature (shape, size, or fluorescence intensity) of rock debris instances in the image is X = {x1, ... x2}. N Using a normal distribution as a model, its distribution parameters are estimated as follows:

[0114] Samples deviating from the center of the normal distribution are considered outliers; therefore, only samples within the 5%–95% distribution range are retained: X 去异常后 ={x|x∈X and Q} N(μ,σ) (0.05)≤x≤Q N(μ,σ) (0.95)

[0115] Among them, Q N(μ,σ) Let be the quantile function of the normal distribution N(μ,σ). Note that 5% to 95% refers to the normal range of the sample distribution estimated by the normal distribution, not the actual 5% to 95% of the sample. Therefore, in practice, samples exceeding or falling below 10% may be considered outliers and removed.

[0116] Figure 3 is a system structure diagram of an automatic logging cuttings identification system provided in one embodiment of the present invention. As shown in Figure 3, the present invention provides an automatic logging cuttings identification system, the system comprising: a collection unit for collecting cuttings samples and performing sample preparation processing on the cuttings samples to obtain detection cuttings; a registration unit for collecting white light dry image and fluorescence image of the detection cuttings respectively, and performing image registration on the fluorescence image based on the white light dry image; an identification unit for: performing cuttings instance segmentation result identification based on the white light dry image to obtain a first identification result; performing cuttings fluorescence imaging result identification based on the registered fluorescence image to obtain a second identification result; and an anomaly removal unit for performing anomaly removal processing on the first identification result and the second identification result, and outputting the detection result.

[0117] Preferably, as shown in Figure 4, the acquisition unit includes: a rock cuttings washing module for collecting rock cuttings samples into a sampler and washing the rock cuttings samples; a rock cuttings leveling module for leveling the washed rock cuttings samples; a rock cuttings drying module for drying the leveled rock cuttings samples; and a rock cuttings conveying module for connecting the rock cuttings washing module, the rock cuttings leveling module, and the rock cuttings drying module to realize the transfer of rock cuttings samples.

[0118] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the above-described automatic logging cuttings identification method.

[0119] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware. This program is stored in a storage medium and includes several instructions to cause a microcontroller, chip, or processor to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0120] The optional embodiments of the present invention have been described in detail above with reference to the accompanying drawings. However, the embodiments of the present invention are not limited to the specific details described above. Within the scope of the technical concept of the embodiments of the present invention, various simple modifications can be made to the technical solutions of the embodiments of the present invention, and these simple modifications all fall within the protection scope of the embodiments of the present invention. It should also be noted that the various specific technical features described in the above specific embodiments can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the embodiments of the present invention will not further describe the various possible combinations.

[0121] Furthermore, various different embodiments of the present invention can be combined in any way, as long as they do not violate the spirit of the embodiments of the present invention, they should also be regarded as the content disclosed by the embodiments of the present invention.

Claims

1. A method for automatic identification of logging cuttings, characterized in that, The method includes: Rock cuttings samples were collected, and sample preparation was performed on the rock cuttings samples to obtain test rock cuttings; White light dry image and fluorescence image of the detected rock cuttings are acquired respectively, and image registration is performed on the fluorescence image based on the white light dry image; Based on the white light dry image, rock cutting instance segmentation result recognition is performed to obtain a first recognition result; Based on the registered fluorescence image, rock cutting fluorescence imaging result recognition is performed to obtain a second recognition result; Anomaly removal processing is performed on the first identification result and the second identification result, and the detection result is output.

2. The method according to claim 1, characterized in that, The sample preparation process for the rock cuttings sample to obtain the rock cuttings for testing includes: The rock cuttings samples were cleaned and then leveled. The leveled rock cuttings samples were dried, and the dried rock cuttings samples were used as test samples.

3. The method according to claim 1, characterized in that, The step of performing image registration on the fluorescence image based on the white light interference image includes: Calculate the single-channel grayscale images of the white light interference image and the fluorescence image respectively, and obtain the grayscale white light image and the grayscale fluorescence image respectively; The grayscale fluorescence image is cropped to retain only the pixels whose grayscale value is the first preset percentage in the image. Calculate the color distribution histograms of the grayscale white light image and the grayscale fluorescence image respectively, and obtain the color distribution histograms of the grayscale white light image and the grayscale fluorescence image respectively; Based on the color distribution histograms of grayscale white light images and grayscale fluorescence images, and using the grayscale white light image as a reference, color histogram matching is performed on the grayscale fluorescence image after grayscale value cropping. SIFT feature extraction is performed on grayscale white light image and grayscale fluorescence image, and fluorescence image registration is performed based on the extracted SIFT features. Each pixel in the fluorescence image is mapped to the same coordinate system as the grayscale white light image to obtain the registered fluorescence image.

4. The method according to claim 3, characterized in that, The SIFT feature extraction of grayscale white light images and grayscale fluorescence images includes: In the grayscale white light image and the grayscale fluorescence image, key points in the image are calculated based on the Harris corner detection algorithm, respectively; For each keypoint, calculate the edge orientation and intensity within a neighboring window centered on it, and compute a SIFT feature descriptor for each keypoint. The mapping relationship between each pixel in the fluorescence image and the same coordinate system as the grayscale white light image is as follows: in, The coordinates of the fluorescence image; [x,y] are the coordinates of the white light image; H′ is the inverse transformation matrix with preset offset parameters.

5. The method according to claim 4, characterized in that, The step of performing rock debris instance segmentation result recognition based on the white light dry image to obtain a first recognition result includes: Superpixel extraction, superpixel color similarity merging, and superpixel morphological similarity are sequentially performed on the white light irradiation image to obtain the instance-segmented white light irradiation image; Based on the instance-segmented white light interfering images, the analysis of each first identification target is performed to obtain the analysis results of each first identification target, which are used as the first identification results; among them, The first identification result includes: Results of white light color identification of rock cuttings, distribution of rock cutting morphology, distribution of rock cutting size, and identification of rock cuttings.

6. The method according to claim 5, characterized in that, The execution rules for superpixel extraction of the white light interpolated image are as follows: The graph-based image segmentation algorithm performs superpixel segmentation on the white light interference image to obtain multiple superpixels, which are used as the superpixel extraction results. The execution rule for superpixel color similarity merging of the white light interference image is as follows: A color similarity set is obtained by calculating the color similarity of each adjacent superpixel separately; In the color similarity set, select the color similarity group with the highest color similarity among a preset number of color similarity groups. Then, merge the adjacent superpixels corresponding to the color similarity groups obtained through the filtering process to obtain the superpixel color similarity merging result; wherein... The rule for calculating the color similarity between adjacent superpixels is as follows: (i < j and S i is adjacent to S j ) Among them, S i With S j For two adjacent superpixels; ColorSim i,j For SuperPixel S i With SuperPixel S j Color similarity between them; h i For SuperPixel S i Vectors obtained by concatenating color histograms; h j For SuperPixel S j The vector obtained by concatenating color histograms.

7. The method according to claim 6, characterized in that, The execution rule for superpixel morphological similarity merging of the white light interpolated image is as follows: Calculate the boundary of each superpixel separately, and perform morphological similarity calculation between adjacent superpixels based on the boundary of each superpixel. The calculation rules are as follows: i < j and S i and S j adjacent Among them, hapeSim i,j For SuperPixel S i With SuperPixel S j Morphological similarity between them; S i，ymin ,S i，ymax ,S i，xmin ,S i，xmax Superpixel S i Four boundaries facing each other; S j，ymin ,S j，ymax ,S j，xmin ,S j，xmax Don't be superpixel S j Four boundaries facing each other; In the set of morphological similarity, select the morphological similarity group with the highest morphological similarity among the preset number of morphological similarity groups, and merge the adjacent superpixels corresponding to the morphological similarity groups obtained by the filtering to obtain the superpixel morphological similarity merging result.

8. The method according to claim 5, characterized in that, The recognition rules for the white light color recognition results of the rock cuttings are as follows: Extract the RGB values ​​of each pixel in the white light interpolated image after instance segmentation, and perform 3-sigma anomaly removal processing on the distribution of each color channel of R, G, and B respectively; For the remaining pixels after anomaly removal, the closest color is searched based on the preset white light color table and used as the color recognition result for the corresponding pixel. Based on the color recognition results of all remaining pixels after anomaly removal, the white light color recognition result of the rock debris is obtained.

9. The method according to claim 5, characterized in that, The identification rules for the rock fragment morphology distribution results and the rock fragment size distribution results are as follows: The eccentricity of the minimum bounding ellipse containing each rock debris instance in the segmented white-light dry image is calculated, and the morphological distribution of the corresponding rock debris instance is described based on the calculated eccentricity; whereby... The smaller the eccentricity, the closer the rock fragments are to round shapes; the larger the eccentricity, the finer and longer the rock fragments are. The proportion of image pixels occupied by each rock debris instance in the white light dry image after instance segmentation is calculated, and the rock debris size distribution of each rock debris instance is described based on the ratio between the pre-camera view areas of the proportion of image pixels occupied.

10. The method according to claim 5, characterized in that, The identification rule for the rock debris naming results is as follows: based on the linear spectrum clustering superpixel segmentation algorithm, superpixel segmentation is performed on each rock debris instance in the white light dry image after instance segmentation. Based on the segmentation results, regions with a size greater than a preset size threshold and an eccentricity greater than a preset eccentricity threshold are filtered out, and the remaining superpixels are used as identification samples. Calculate the particle size of the remaining superpixels, and statistically analyze the particle size distribution rules and the average particle size based on the particle size. Based on the particle size distribution rules and the average particle size, corresponding rock debris features are matched in a pre-constructed rock debris knowledge graph database, and the name of the rock debris with the highest matching degree is taken as the current rock debris naming result.

11. The method according to claim 4, characterized in that, The second identification result includes: Results of fluorescence color identification and fluorescence intensity identification of rock cuttings; The recognition rules for the fluorescence color recognition results of the rock fragments are as follows: The RGB values ​​of each pixel in the registered fluorescence image are extracted, and the distribution of each color channel (R, G, B) is subjected to 3-sigma anomaly removal processing. For the remaining pixels after anomaly removal, the closest color is searched based on a preset fluorescence color table and used as the color recognition result for the corresponding pixel. Based on the color recognition results of all remaining pixels after anomaly removal, the fluorescence color recognition result is obtained.

12. The method according to claim 11, characterized in that, The identification rules for the rock cutting fluorescence intensity identification results are as follows: In the registered fluorescence image, regions containing fluorescence are identified, and the fluorescence intensity is calculated based on the RGB values ​​of all pixels to obtain a single-channel grayscale image; where, The grayscale value of the single-channel grayscale image is the fluorescence intensity; Based on the fluorescence color recognition results, regions that are not the current fluorescence color are filtered out, and the fluorescence intensity distribution histogram of the remaining regions is statistically analyzed as the result of rock debris fluorescence intensity recognition.

13. The method according to claim 1, characterized in that, The step of performing anomaly removal processing on the first identification result and the second identification result, and outputting the detection result, includes: The feature sets of the first recognition result and the second recognition result are obtained respectively to obtain the feature set of the first recognition result and the feature set of the second recognition result; Normal distribution models are constructed for the first recognition result feature set and the second recognition result feature set, respectively, as follows: Where, x i ∈X,X={x1,…x N } represents either the first identification result feature set or the second identification result feature set; N is the number of features in the first or second identification result feature set; μ is the normal distribution value corresponding to the i-th first identification result feature or the i-th second identification result feature; Features that deviate from the preset proportion range of the normal distribution center are filtered as outliers to obtain the detection results after anomaly removal.

14. An automatic logging cuttings identification system, characterized in that, The system includes: The acquisition unit is used to acquire rock cutting samples and perform sample preparation processing on the rock cutting samples to obtain detection rock cuttings; the registration unit is used to acquire white light dry image and fluorescence image of the detection rock cuttings respectively, and perform image registration on the fluorescence image based on the white light dry image; The identification unit is used for: Based on the white light dry image, rock cutting instance segmentation result recognition is performed to obtain a first recognition result; Based on the registered fluorescence image, rock cutting fluorescence imaging result recognition is performed to obtain a second recognition result; The anomaly removal unit is used to perform anomaly removal processing on the first identification result and the second identification result, and output the detection result.

15. The system according to claim 14, characterized in that, The acquisition unit includes: The rock cuttings retrieval and washing module is used to collect rock cuttings samples into the sampler and perform cleaning treatment on the rock cuttings samples; The rock cuttings leveling module is used to perform leveling treatment on the cleaned rock cuttings samples; The rock cuttings drying module is used to dry the leveled rock cuttings samples. The rock cuttings conveying module is used to connect the rock cuttings washing module, the rock cuttings leveling module and the rock cuttings drying module to realize the transfer of rock cuttings samples.

16. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the automatic logging cuttings identification method as described in any one of claims 1-13.

Images