Representative area-based high-efficiency cataloging method and system for drilling core hyperspectrum

Abstract

The application belongs to the technical field of geological exploration, and specifically discloses a drilling core hyperspectral efficient logging method and system based on a representative area, which comprises the following steps: obtaining segmented hyperspectral image data of a drilling core; splicing the hyperspectral images of all segments into a depth-continuous core mosaic image; performing mask processing on the obtained core mosaic image to determine the boundary of a single core; finding the center position and the area width of the optimal representative area in each core, respectively; performing mineral mapping on the representative area and cutting, splicing the cut representative area of the mineral mapping according to the depth; calculating the relative content of different minerals in each depth interval to obtain the depth and relative content profile of each mineral. The application can screen out a high-spectral quality core area with a small degree of distortion and complete spectral information, greatly reduce the amount of data to be processed, and ensure the spectral effectiveness of the area.

Description

Technical Field

[0001] This invention relates to the field of geological exploration technology, and in particular to a method and system for efficient hyperspectral logging of borehole cores from representative areas. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Borehole core logging is a core foundational task in geological exploration. It involves the systematic observation, description, recording, and analysis of core samples retrieved from boreholes, and has crucial supporting value for mineral resource assessment, hydrogeological and engineering geological evaluation, and oil and gas exploration.

[0004] Traditional core logging relies heavily on the visual observation and experience of geologists, which is highly subjective. Hyperspectral imaging technology, however, can simultaneously record corresponding spectral information while acquiring target image information. Identifying minerals based on their unique spectral characteristics is akin to reading their "chemical fingerprints," resulting in highly objective mineral identification results and effectively eliminating inconsistencies and subjective biases inherent in human descriptions. This technology can perform non-destructive scanning of core surfaces at pixel-level resolution, accurately capturing subtle geological features easily missed by the naked eye or traditional methods, such as alteration zoning, fine veins, and sedimentary laminae.

[0005] However, while hyperspectral imaging technology acquires massive amounts of data, it also brings a series of technical challenges: On the one hand, the massive amount of data puts considerable pressure on computing resources. A single 1-meter core often contains tens of thousands of pixels and hundreds of bands of data. Traditional whole-processing methods for mineral mapping are extremely time-consuming and cannot meet the needs of rapid cataloging in practical engineering. On the other hand, when acquiring hyperspectral images of cores, the geometric relationship between the light source, probe, and core affects the signal intensity of each band as the angle between the core and the vertical line of the spectrometer probe increases, exacerbating spectral distortion. Traditional methods require complex correction of the entire area of ​​data, which has limited effectiveness and is costly. In addition, non-rock impurities (wooden boxes, labels) between cores interfere, further introducing irrelevant spectral information and affecting the effective interpretation of the data. Summary of the Invention

[0006] To address the aforementioned issues, this invention proposes a highly efficient hyperspectral logging method and system for borehole cores based on representative regions. This method can focus the analysis on key geological units such as ore-bearing layers, alteration zones, and transition zones between different lithologies. It not only significantly reduces the amount of data and alleviates computational complexity, but also significantly improves the algorithm's efficiency when processing the massive amount of band and pixel information contained in hyperspectral data.

[0007] In some implementations, the following technical solutions are adopted: A high-efficiency hyperspectral logging method for borehole cores based on representative regions includes: Acquire segmented hyperspectral image data of borehole cores; A linear regression method was used to establish the conversion relationship between pixel coordinates and the actual depth of the core, and the hyperspectral images of all segments were stitched together into a depth-continuous core mosaic. The obtained core mosaic was masked to remove the wooden box, the label recording the borehole number information, and the locations with reflectivity below the set threshold. For the processed image, calculate the average reflectance of each column in the corresponding column interval of a single rock core. The column index with the lowest average reflectance is the boundary of the single rock core. Based on the boundary segmentation of a single core, the center position and width of the optimal representative region are found in each core, and the index is recorded. Mineral mapping is performed on representative areas and then cropped. The representative areas of the cropped mineral mapping are then stitched together according to depth. The relative content of different minerals in each depth interval is calculated to obtain a depth and relative content profile for each mineral.

[0008] As a further approach, a linear regression method is used to establish the conversion relationship between pixel coordinates and the actual depth of the core sample, specifically: Multiple calibration points were selected on each core sample, and the actual depth of each calibration point on the core sample and the pixel coordinates along the length of the core sample were recorded. Using pixel coordinates as the independent variable and actual depth as the dependent variable, a linear regression model was established. The model was fitted using multiple calibration points to obtain the conversion coefficient between the pixel coordinates and actual length of each core sample. The converted images were stitched together sequentially according to the actual depth of the core, with a filler band remaining at the joint between adjacent segments, to obtain a panoramic hyperspectral mosaic of the core with continuous depth.

[0009] As a further step, the obtained core mosaic is masked to remove the wooden box, the label recording borehole number information, and locations with reflectivity below a set threshold. Specifically: Morphological operations are used to extract connected regions from an image, identifying peripheral connected regions with areas smaller than a set threshold and irregular shapes. Calculate the ratio of the average reflectance of the area near the wood absorption valley to the reflectance at the wood absorption valley. If the ratio is greater than a set value, mark this area as a wooden box. Calculate the ratio of the average reflectance of the area near the absorption valley of the tag to the reflectance at the absorption valley of the tag. If the ratio is greater than a set value, then this area is marked as a tag. Mark the pixels of the outer connected region, the wooden box, and the label as 0, and update the mask; Calculate the average reflectance of each pixel across the entire band, mark pixels with an average reflectance lower than a set threshold as 0, and update the mask again; finally, obtain the core mask image with the wooden box removed, the label recording the borehole number information, and the reflectance lower than the set threshold.

[0010] As a further solution, the method for determining the boundary of the single core sample is as follows: Determine the total number of columns W in the hyperspectral image and the total number N of core samples to be identified, and calculate the theoretical spacing between adjacent core samples. ; For the nth core, in its corresponding column interval Calculate the average reflectance of each column of b bands. ; Within the column interval, find the two column indices with the smallest average reflectance, which are the left and right boundaries of the nth core.

[0011] As a further step, the optimal representative region's center location is identified within each core sample, specifically: A single rock core is divided into multiple columns according to pixels; Calculate the average reflectance of each column of b bands in the current single core sample. As a measure of signal strength, calculate the standard deviation of reflectivity across all bands in each column. As noise intensity; Calculate the ratio of signal strength to noise intensity. The calculated ratios and average reflectance of each column are weighted and summed, and the column corresponding to the maximum value is selected as the center column of the representative area of ​​the current single core.

[0012] As a further approach, the method for determining the optimal width of the representative region is as follows: Define the candidate region within the candidate width set with the center column c as the axis of symmetry. The candidate width set is the set of widths from 1 to the theoretical distance D between adjacent cores; Calculate candidate regions separately Internal and external areas The spectral statistics include the within-class mean, within-class variance, combined between-class variance, and combined within-class variance; Construct a width optimization objective function based on the combined inter-class variance and the combined intra-class variance; The optimal width parameter is obtained by iterating through the candidate width set and solving the objective function for width optimization.

[0013] As a further solution, the width optimization objective function Specifically: ; in, To combine the variance between classes, To combine the within-class variance, This is the balance coefficient.

[0014] As a further step, the relative content of different minerals within each depth range is calculated, specifically: Based on the conversion relationship between pixel coordinates and actual core depth, the depth range corresponding to the representative area of ​​each core segment is determined. The representative regions are sorted according to their depth. For the k-th depth interval Count the number of pixels for each mineral i within this interval. relative content ; This represents the total number of all valid mineral pixels within the k-th depth interval; All depth segments Arranged in depth order, the depth-relative content profile of mineral i is obtained.

[0015] In other embodiments, the following technical solutions are adopted: A high-efficiency hyperspectral logging system for borehole cores from representative regions includes: The data acquisition module is used to acquire segmented hyperspectral image data of borehole cores; The core mosaic construction module is used to establish the conversion relationship between pixel coordinates and the actual depth of the core using a linear regression method, and to stitch together the hyperspectral images of all segments into a depth-continuous core mosaic. The image processing module is used to perform masking on the obtained core mosaic, removing the wooden box, the label recording the borehole number information, and the locations with reflectivity below a set threshold. The boundary segmentation module is used to calculate the average reflectance of each column in the corresponding column interval of a single rock core for the processed image. The column index with the lowest average reflectance is the boundary of the single rock core. The representative region determination module is used to find the center position and width of the optimal representative region within each core based on the boundary segmentation of a single core, and record the index. The mineral content calculation module is used to fill and crop representative areas with minerals, and then stitch the representative areas of the cropped mineral filling according to depth; it calculates the relative content of different minerals in each depth interval to obtain the depth and relative content profile of each mineral.

[0016] In other embodiments, the following technical solutions are adopted: A terminal device includes a processor and a memory, the processor being used to implement instructions; the memory being used to store multiple instructions adapted for loading and execution by the processor of the above-described high-efficiency hyperspectral logging method for borehole cores based on representative regions.

[0017] Compared with the prior art, the beneficial effects of the present invention are: (1) The present invention adopts the representative region (REU) center positioning technology to solve the problem that the large scale of full core data (tens of thousands of pixels × hundreds of bands) leads to high computational resource pressure and long compilation time. The method of the present invention can screen out high-spectral quality core regions with small distortion and complete spectral information, greatly reduce the amount of data to be processed, and at the same time ensure the spectral validity of the region, thereby improving compilation efficiency and avoiding invalid spectral interference.

[0018] As the angle of the core deviates from the vertical of the spectrometer probe increases, the signal intensity at each wavelength weakens (the spectral distortion effect increases). The representative region of the core has the strongest signal intensity and the smallest spectral distortion effect. This invention uses the representative region as the core to replace traditional full-data processing. It solves the pain points of traditional full-data hyperspectral processing mode, such as "excessive time consumption (unable to meet the needs of rapid engineering logging), spectral distortion interference, and spectral contamination by impurities," and achieves the goals of "high efficiency (significantly reduced data volume), accuracy (low distortion + no impurity interference), and suitability for actual engineering needs" in core hyperspectral logging.

[0019] (2) The present invention determines the center column by using the weighted result of signal-to-noise ratio and signal strength. This method takes into account both the intensity characteristics and stability characteristics of the spectrum. It avoids selecting columns with large noise or drastic spectral fluctuations by simply relying on signal strength, and also avoids selecting columns with weak signals and insufficient representativeness by simply relying on signal-to-noise ratio. This ensures that the center column has both typical spectral characteristics and good spectral stability. At the same time, the process is completely data-driven and does not require human experience intervention, which significantly improves the objectivity and repeatability of center selection.

[0020] This invention constructs a width optimization objective function based on the difference between comprehensive inter-class variance and comprehensive intra-class variance. The optimal width is determined by iteratively solving for the width that maximizes the objective function value. This method ensures significant spectral differences between the selected region and the external region by maximizing inter-class variance, thus enhancing the representativeness of the region. It also ensures high consistency of spectral characteristics across columns within the region by minimizing intra-class variance, preventing the mixing of different mineral layers or noisy regions into the same area and guaranteeing the purity of the aggregated spectrum. Furthermore, it introduces a balance coefficient to adjust the weights of inter-class and intra-class variances, allowing for flexible width selection to adapt to the spectral characteristics of different core structures. Finally, it automatically determines the width through iterative optimization, avoiding the drawbacks of fixed widths leading to over-aggregation in densely bedding areas and loss of detail, or excessively narrow widths in homogeneous areas and increased redundant computation. This achieves an optimal balance between spectral fidelity, computational efficiency, and spatial structure preservation, significantly improving the adaptability and engineering applicability of representative region extraction.

[0021] (3) The edge regions of the core contain angular distortion and impurity interference, and these edge regions correspond precisely to the boundaries with the lowest reflectivity. This invention locates the boundaries using average reflectivity, accurately identifying and discarding edge parts that need to be discarded, and automatically retaining the central region with the best signal quality, thus meeting the experimental requirements for signal purity. Using the average value of reflectivity, a physical signal, as the criterion, the method is based entirely on data calculation, avoiding the subjectivity of manual visual boundary identification, resulting in more stable and repeatable results. This method can automatically complete the boundary calculation of batch cores through algorithms, eliminating the need for manual labeling of each core, significantly improving efficiency when processing large amounts of core data, and is suitable for large-scale experimental scenarios. Even when the core surface texture is uneven and the impurity distribution is complex, the average statistical reflectivity can still stably capture boundary features, making it more robust than methods that rely on local texture.

[0022] Other features and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0023] Figure 1 This is a flowchart of the high-efficiency hyperspectral logging method for borehole cores based on representative regions in an embodiment of the present invention; Figure 2 This is an example of a core mosaic in an embodiment of the present invention; Figure 3 This is a schematic diagram of multiple columns of pixels divided from a single rock core in an embodiment of the present invention; Figure 4 This is a schematic diagram of core boundary identification in an embodiment of the present invention. Detailed Implementation

[0024] It should be noted that the following detailed description is illustrative and intended to provide further explanation of the invention. Unless otherwise specified, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0025] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0026] Example 1 In one or more embodiments, a high-efficiency hyperspectral logging method for borehole cores based on representative regions is disclosed, combined with... Figure 1 Specifically, it includes the following processes: S101: Acquire segmented hyperspectral image data of borehole cores.

[0027] In this embodiment, a hyperspectral imager is used to acquire hyperspectral images of each core segment. During hyperspectral image acquisition, contaminants on the core surface are first removed, and metadata such as borehole number, depth range, and scanning parameters are recorded simultaneously during core acquisition. The data format of the acquired three-dimensional hyperspectral data cube is H×W×B; where H is the height of the hyperspectral data volume, W is the width of the hyperspectral data volume, and B is the number of bands.

[0028] S102: A linear regression method was used to establish the conversion relationship between pixel coordinates and the actual depth of the core, and the hyperspectral images of all segments were stitched together into a depth-continuous core mosaic.

[0029] Because the core samples were taken in segments, this embodiment stitches together the hyperspectral images of all segments into a single, depth-continuous core mosaic, such as... Figure 2 As shown, the vertically stacked core boxes are converted into a horizontal grid pattern while maintaining continuity in the depth direction of the cores. Using the borehole depth "from shallow to deep" as the sole criterion, all core boxes are sorted (corresponding to the order of box 1, box 2... box n in the diagram), determining the position of each box in the final mosaic. A grid layout is adopted from left to right and from top to bottom. Figure 2 The example is 2 rows and 3 columns (the number of rows and columns can be adjusted according to the number of boxes), ensuring that the columns (①-⑤) of each core box are aligned horizontally and vertically. Here, ①-⑤ represent different core segments (each core segment is considered as one core).

[0030] To avoid conversion errors caused by differences in shooting distance and angle during segmented core imaging, this embodiment establishes a dynamic conversion relationship between pixel coordinates and actual core depth. The specific process is as follows: S1021: Select 20 calibration points for every 1m of core sample, and record the actual depth of each calibration point on the core segment (the distance from the beginning of the segment, which can be determined directly by measuring the core box scale and ruler) and the pixel coordinates along the length of the core in the hyperspectral image.

[0031] S1022: Using pixel coordinate x as the independent variable and actual length L as the dependent variable, the (x) of the 20 calibration points... i ,L i Substituting the data into the linear regression model, the optimal linear equation L=k×x+b was obtained by fitting the model. S1023: Extract the conversion coefficients. The slope k of the fitted equation is the pixel-to-length conversion coefficient of the 1m core segment. The coefficients are calculated independently for each 1m segment, which can eliminate the systematic error caused by segmented shooting. The intercept b is an auxiliary correction term. If there is a systematic deviation (such as image cropping causing the starting pixel to not correspond to the core endpoint), b will be automatically fitted as a correction value to ensure the accuracy of the conversion. S1024: The converted images are stitched together sequentially according to the actual depth of the core, and a 1-pixel fill band is reserved at the joint between adjacent segments to prevent grayscale abrupt changes caused by overlap error. Finally, a panoramic hyperspectral mosaic of the core with continuous depth information corresponding to "pixel coordinates - actual depth" is generated.

[0032] S103: Perform masking on the obtained core mosaic to remove the wooden box, the label recording the borehole number information, and the locations where the reflectivity is lower than the set threshold.

[0033] Core samples are typically placed in wooden boxes, and their distribution along the length of the image is usually elongated. The wooden boxes are located on either side or along the top and bottom edges of the core samples. The spectral characteristics of the wooden boxes and the core samples differ significantly (wood exhibits strong absorption valleys at positions such as 1720), and their spatial morphology is fixed (the outer region of the core sample). Therefore, this embodiment employs a combined masking process of "spatial constraint + spectral recognition." The specific process is as follows: Morphological operations are used to extract connected regions from the image, and small, irregularly shaped peripheral connected regions are removed (typically, the horizontal area of ​​a core box is only about 5 pixels; if the horizontal pixel count of this area is ≤5, it is removed).

[0034] Since wood has a strong absorption valley at 1720 nm, the ratio of the average reflectance at nearby locations of 1700 nm and 1800 nm to the reflectance at 1720 nm is calculated: If satisfied If the pixel value of the outer connected region and the wooden box is zero, then this region is marked as the wooden box. Mark the pixels of the outer connected region and the wooden box as 0, and update the mask.

[0035] Drill hole number labels are mostly paper stickers or plastic covers (with a noticeable absorption valley at 1200nm). Therefore, the ratio of the average reflectance at the adjacent locations of 1100nm and 1300nm to the reflectance at 1200nm is calculated: If satisfied This area is then marked as the borehole number label area. The larger the value, the more obvious the absorption valley at 1200nm. Mark the pixels in the labeled area as 0 and update the mask again.

[0036] Statistically analyze the reflectance distribution of the remaining region after removing the masked area (0), and take 1 / 3 of the mean reflectance as the threshold T. Calculate the average reflectance of each pixel across the entire band in the remaining region after removing the masked area: ,in is the average reflectance of a single pixel across all wavelengths, and b is the total number of wavelengths across all wavelengths of the hyperspectral spectrum. For this pixel in the j-th band (corresponding wavelength) Reflectance under ( ).

[0037] Pixels with an average reflectance less than the threshold T are selected as low reflectance regions. Pixels in the low reflectance regions are marked as 0, and the mask is updated again.

[0038] The final result was a core mask image with the wooden box, the label recording the borehole number information, and the area with reflectivity below the set threshold removed.

[0039] Compared to conventional masking methods that rely heavily on single grayscale, morphological, or fixed thresholds, which are prone to misidentification or missed detection, this embodiment employs a multi-dimensional masking technology combining spatial constraints and spectral recognition. Spatially, it utilizes the fixed characteristics of the wooden box's "peripheral distribution and regular shape" to accurately locate the box through connected region analysis. Spectrally, it leverages the unique spectral characteristics of wood (strong absorption valley at 1720nm) and the label (absorption valley at 1200nm) to differentiate interference items from the core material, significantly reducing the false recognition rate. Instead of using a fixed threshold, a threshold T (1 / 3 of the average reflectance) is dynamically generated by statistically analyzing the reflectance distribution of the remaining region. This adaptive logic can adapt to the reflectance differences of different core images, avoiding the failure of fixed thresholds in complex scenarios and broadening its applicability. Ultimately, it eliminates non-rock impurities and low-quality areas at the source, providing "clean data" for subsequent boundary identification and REU screening, reducing invalid data and indirectly lowering subsequent computational pressure.

[0040] S104: For the processed image, calculate the average reflectance of each column in the corresponding column interval of a single rock core. The column index with the lowest average reflectance is the boundary of the single rock core.

[0041] Specifically, within the region containing each core sample, a single core sample is divided into multiple columns according to pixels, with each column consisting of a specific column of pixels, such as... Figure 3 As shown, Figure 3 The rock core ① is divided into multiple columns according to pixels.

[0042] For the nth core sample, within its corresponding column interval, calculate the average spectral reflectance for each column with b bands. The index of the column with the lowest reflectance is the boundary of the core sample, denoted as . , where n is the nth core sample.

[0043] The specific process is as follows: S1041: Determine the total number of columns W in the hyperspectral image, the total number N of core samples to be identified, and calculate the theoretical spacing between adjacent core boundaries. ; S1042: For the nth core, in its corresponding column interval The average reflectance of each column is calculated as follows: Where b is the band number, It is the reflectance of the x-th column and j-th band, where j is the band index and x is the column index.

[0044] S1043: Within this interval, find the two column indices with the lowest average reflectance; these are the left and right boundaries of the nth core sample. The left boundary is... The right boundary is .

[0045] Figure 4 A schematic diagram of core boundary identification is provided. Figure 4 The left side shows the false color of the original core hyperspectral image, and the right side shows the average reflectance curve of the row corresponding to the center point of the left square. The horizontal axis is the number of pixel sample coordinates, and the vertical axis is the average reflectance of b bands. It can be seen that the green triangle indicates the column position with the minimum average reflectance, and the left and right boundaries of each core can be clearly seen.

[0046] S105: Based on the boundary segmentation of a single core, find the center position and width of the optimal representative region within each core and record the index.

[0047] Specifically, the process of finding the center location of the optimal representative region is as follows: S1051: The average reflectance of each column calculated in step S1042 is used as the signal strength; S1052: Calculate the standard deviation of reflectivity for all bands in each column as the noise intensity: ; Where b is the number of bands, It is the reflectance of the x-th column and j-th band, where j is the band index and x is the column index.

[0048] S1053: Calculate the ratio of signal strength to noise strength as the signal-to-noise ratio. ; S1054: Perform a weighted summation of the signal-to-noise ratio and signal strength for each column, and select the column corresponding to the maximum value as the center column c of the representative region. ; in, These are all weighting coefficients, which can be selected based on experience as 0.3 and 0.7 respectively.

[0049] After obtaining the center column position of the representative region, the width of the representative region is further determined.

[0050] If a fixed width is used, the spectral heterogeneity and bedding thickness differences at different locations in the core are significant. A fixed width will lead to "over-aggregation of bedding regions (loss of fine structure) and narrowing of uniform regions (increasing redundant calculations)," failing to adapt to the spatial / spectral differences of the core. Furthermore, a fixed width only focuses on computational efficiency, sacrificing spectral fidelity. Excessive spectral differences within a region mean that the aggregated representative region cannot reflect the true mineral characteristics of the original core, directly causing errors in subsequent interpretation results.

[0051] To address the aforementioned issues, this embodiment utilizes intra-class variance and inter-class variance to calculate the width *w* of the representative region. Intra-class variance constrains the "consistency of spectral characteristics within the region" (the smaller the intra-class variance, the smaller the spectral differences between columns within the region), while inter-class variance distinguishes the "spectral specificity between the region and the outside world" (ensuring the representative region reflects local core characteristics). The combination of these two methods allows the aggregated region to accurately match the spectral characteristics of the original data, avoiding interpretation bias. The variance index quantifies the "reasonable degree of aggregation," preventing region fragmentation due to excessively narrow width (increasing redundant calculations) and avoiding the introduction of irrelevant data due to excessively wide width (slowing down processing speed), thus meeting the time-sensitive requirements of "rapid cataloging" in engineering projects. Combining the required scale of the working conditions (matching the spatial dimensions of the actual core texture and bedding) avoids the problem of mismatch between the width and the actual core structure.

[0052] In this embodiment, a candidate region is defined within the candidate width set with the center column c as the axis of symmetry. The candidate width set refers to the width set from 1 to the theoretical spacing D. A systematic algorithm is needed to balance spectral fidelity (ensuring consistency between the spectral characteristics of the aggregated region and the original data), computational efficiency (avoiding information loss due to over-aggregation or redundancy due to insufficient aggregation), and spatial structure preservation (maintaining the spatial distribution characteristics of texture, bedding, etc., within the core). The specific process is as follows: Calculate candidate regions separately Internal and external areas The spectral statistics include: within-class mean, within-class variance, combined between-class variance, and combined within-class variance; the specific calculation formulas are as follows: Candidate region Internal and external areas The within-class means are as follows: ; in, For pixel i at wavelength The reflectivity of the following These represent the total number of pixels in the inner and outer regions, respectively. Candidate region Internal and external areas The intra-class variances are as follows: ; The specific variance between classes is as follows: ; The comprehensive within-class variance is as follows: ; in, These represent the pixel ratios of the candidate region and its outer region, respectively.

[0053] Construct a width optimization objective function based on the combined inter-class variance and the combined intra-class variance: ; in, This is the internal and external variance balance coefficient, with a value ranging from 0.3 to 0.7.

[0054] Within the preset search interval (candidate width set), the objective function is... By iterating through the problem, we can obtain the optimal width parameter that maximizes the objective function. .

[0055] In this embodiment, the width optimization objective function ensures that the selected region has a significant spectral difference from the external region by maximizing the inter-class variance, thereby enhancing the representativeness of the region. By minimizing the intra-class variance, it ensures that the spectral characteristics of each column within the region are highly consistent, thus avoiding the mixing of different mineral layers or noisy regions into the same region and ensuring the purity of the aggregated spectrum. By introducing a balance coefficient to adjust the weights of inter-class and intra-class variances, the width selection can be flexibly adapted to the spectral characteristics of different core structures. By automatically determining the width through traversal optimization, it avoids the defects of excessive aggregation in densely bedding areas with fixed widths leading to loss of details and excessive narrowness in uniform areas leading to redundant calculations. It achieves the optimal balance between spectral fidelity, computational efficiency, and spatial structure preservation, significantly improving the adaptability and engineering applicability of representative region extraction.

[0056] S106: Perform mineral mapping and cropping on representative areas, and stitch together the representative areas of the cropped mineral mapping according to depth; calculate the relative content of different minerals in each depth interval to obtain the depth and relative content profile of each mineral.

[0057] In this embodiment, the hyperspectral images of the core after the above processing are subjected to noise reduction, endmember identification, and mineral content calculation, and the mapping results of different minerals are exported.

[0058] The noise reduction process utilizes a denoising network that combines spectral and spatial information. In the spatial domain, a nonlocal mean algorithm is used to calculate the similarity weights of neighboring pixel blocks to restore details; in the spectral domain, spectral angle matching is introduced to measure the similarity of spectral vectors; finally, a filtering model is constructed by combining the two types of weights.

[0059] Among them, nonlocal similarity in the spatial domain is used to find pixel blocks with similar structures in the image, and noise is suppressed by weighted averaging, aiming to preserve the texture, bedding and other details of the rock core at the same time.

[0060] The similarity weights of neighboring pixel blocks are calculated as follows: S1061: Pixel block selection: For the current pixel x, define a neighborhood block of fixed size, denoted as B(x); at the same time, within a larger search window, extract a neighborhood block of the same size B(y) for each candidate pixel y. S1062: Perform similarity calculation, calculate the Euclidean distance between two blocks B(x) and B(y), and measure the spatial gray-level similarity: ; in, , respectively, are the original reflectance values ​​of sub-pixel i of pixel block B(x) and sub-pixel j of pixel block B(y); S1063: Convert distance to similarity weights: h is the spatial domain smoothing parameter; S1064: Spectral vector similarity is measured using spectral angle matching. The cosine of the angle between the x and y spectral vectors of a pixel is calculated using the following formula: ,in For pixel x, the complete spectral vector y is the complete spectral vector of pixel y.

[0061] S1065: The cosine of the included angle is used as the weight for spectral similarity, and the formula is as follows: ,in represents the spectral similarity weights between pixels x and y.

[0062] S1066: Normalize spatial weights Weighted by spectral similarity Direct multiplication yields the combined spatial-spectral weight. This combined weight is then normalized, and the weighted average of the original reflectance of all candidate pixels is calculated. The result is the final reflectance of the target pixel after denoising, thus achieving comprehensive denoising using both spectral and spatial information. This joint weighting strategy avoids both the spectral blurring that may result from relying solely on spatial information and the spatial context that may be ignored when relying solely on spectral information. Ultimately, it effectively removes noise while maximizing the preservation of spatial detail and spectral fidelity of the core image.

[0063] In this embodiment, the process of endmember identification and mineral content calculation is as follows: First, using the pixel purity index, the data is projected onto a random vector multiple times, and pixels that frequently appear at extreme value positions are recorded as "pure pixels" to achieve endmember extraction. Calculate the location of different absorption valleys Absorption depth Half height and width and the symmetry of the absorption valley .

[0064] Specifically, the location of the absorption valley In the hyperspectral reflectance curve, find the wavelength corresponding to the minimum reflectance value; this wavelength is the location of the absorption valley.

[0065] Absorption depth First, determine the baseline of the absorption valley (where the baseline is a straight line connecting the two shoulder points of the absorption valley), then calculate the reflectance of the baseline at the location of the absorption valley, and then read the reflectance at the bottom of the valley. The absorption depth is then the absolute value of the ratio of the reflectance at the location of the absorption valley to the reflectance at the bottom of the valley.

[0066] Furthermore, calculate the half-peak height, which is half the ratio of baseline reflectance to valley reflectance; let the left and right intersection points at the half-peak height be respectively... The left and right half widths are Symmetry is .

[0067] Furthermore, construct shape feature vectors. Different shape parameters contribute differently to the results, and need to be differentiated by setting weights.

[0068] Let the shape feature vector of the endmember be... The shape feature vector of the hyperspectral image of the representative area of ​​the core is Then the weighted distance between the two can be written as: Where M is the total dimension of the feature vectors, The weight of the j-th feature dimension satisfies , The size is assigned different weights based on experience.

[0069] Weighted shape distance Compared with the residual in traditional SFF By combining these factors, a comprehensive matching index S is constructed, and the matching degree between each input terminus and all pixels in the image is calculated: ; in, This is the balance coefficient between morphological information and the overall fitting residual.

[0070] Furthermore, the reciprocal of the comprehensive matching index S is normalized and used as a weight in the linear combination equation of the feature vectors, as follows: For each mineral i, it satisfies the linear combination equation of the mineral's eigenvectors: ; in, It represents the content of the i-th mineral in the representative region.

[0071] Finally, the content of each mineral in the representative area was calculated and applied to the panoramic hyperspectral core image to obtain a mineral content distribution map (i.e., mineral mapping of the representative area).

[0072] In the representative region of a single core obtained based on the center column and optimal width parameters, there may be areas marked as 0 by a mask, such as cracks, remnants of wooden boxes, and shadows. In order to reduce the impact of these areas on the final mineral content calculation results, this embodiment trims the representative region. The specific process is as follows: In the optimal representative region (REU) of each core sample, the percentage of invalid pixels (mask=0: fractures, remnants of wooden boxes, shadows, etc.) was calculated: ; in, , These represent the number of all invalid pixels (mask=0) within the REU region and the total number of pixels in the entire REU region, respectively.

[0073] like If the REU segment is not representative, it is considered to be skipped or the REU boundary is adjusted, and the area is expanded to the left and right by 5-10 pixels for re-verification to ensure that the final cropped area meets the quality requirements.

[0074] Finally, the obtained mineral mapping results are mapped onto REU regions that meet the quality requirements.

[0075] It should be noted that when the window clipping touches the core boundary, the candidate region satisfies... The pixel columns are constrained to prevent the selected candidate region from exceeding the core boundary.

[0076] In this embodiment, the relative content of different minerals within each depth range is calculated, specifically as follows: Based on the conversion relationship between pixel coordinates and actual core depth, the depth range corresponding to the representative area of ​​each core segment is determined. The representative regions are sorted according to their depth. For the k-th depth interval ,in To allow for adjustments based on user needs, the number of pixels for each mineral i within this interval is counted. relative content ; This represents the total number of all valid mineral pixels within the k-th depth interval; All depth segments Arranged in depth order, the depth-relative content profile of mineral i is obtained.

[0077] Based on the mineral mapping of different minerals on the panoramic hyperspectral core image and the depth-relative content profile of different minerals, efficient hyperspectral logging of borehole cores is achieved.

[0078] Example 2 In one or more embodiments, a high-efficiency hyperspectral logging system for borehole cores based on representative regions is disclosed, specifically including: The data acquisition module is used to acquire segmented hyperspectral image data of borehole cores; The core mosaic construction module is used to establish the conversion relationship between pixel coordinates and the actual depth of the core using a linear regression method, and to stitch together the hyperspectral images of all segments into a depth-continuous core mosaic. The image processing module is used to perform masking on the obtained core mosaic, removing the wooden box, the label recording the borehole number information, and the locations with reflectivity below a set threshold. The boundary segmentation module is used to calculate the average reflectance of each column in the corresponding column interval of a single rock core for the processed image. The column index with the lowest average reflectance is the boundary of the single rock core. The representative region determination module is used to find the center position and width of the optimal representative region within each core based on the boundary segmentation of a single core, and record the index. The mineral content calculation module is used to fill and crop representative areas with minerals, and then stitch the representative areas of the cropped mineral filling according to depth; it calculates the relative content of different minerals in each depth interval to obtain the depth and relative content profile of each mineral.

[0079] It should be noted that the specific implementation methods of the above modules are exactly the same as those in Example 1, and will not be described in detail again.

[0080] Example 3 In one or more embodiments, a terminal device is disclosed, comprising a processor and a memory, wherein the processor is used to implement instructions; and the memory is used to store multiple instructions adapted to be loaded by the processor and executed by the processor to perform the high-efficiency hyperspectral recording method for borehole cores based on representative regions as described in Embodiment 1.

[0081] It should be understood that in this embodiment, the processor can be a central processing unit (CPU), or it can be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or other conventional processor.

[0082] Memory may include read-only memory and random access memory, and provides instructions and data to the processor. A portion of memory may also include non-volatile random access memory. For example, memory may also store information about the device type.

[0083] In the implementation process, each step of the above method can be completed by the integrated logic circuits in the processor hardware or by software instructions.

[0084] While the specific embodiments of the present invention have been described above in conjunction with the accompanying drawings, this is not intended to limit the scope of protection of the present invention. Those skilled in the art should understand that various modifications or variations that can be made by those skilled in the art without creative effort based on the technical solutions of the present invention are still within the scope of protection of the present invention.

Claims

1. A high-efficiency hyperspectral logging method for borehole cores from representative regions, characterized in that, include: Acquire segmented hyperspectral image data of borehole cores; A linear regression method was used to establish the conversion relationship between pixel coordinates and the actual depth of the core, and the hyperspectral images of all segments were stitched together into a depth-continuous core mosaic. The obtained core mosaic was masked to remove the wooden box, the label recording the borehole number information, and the locations with reflectivity below the set threshold. For the processed image, calculate the average reflectance of each column in the corresponding column interval of a single rock core. The column index with the lowest average reflectance is the boundary of the single rock core. Based on the boundary segmentation of a single core, the center position and width of the optimal representative region are found in each core, and the index is recorded. Mineral mapping is performed on representative areas and then cropped. The cropped representative areas of the mineral mapping are then stitched together according to depth. The relative content of different minerals in each depth interval is calculated to obtain a depth and relative content profile for each mineral. Specifically, the optimal representative region's center location is found within each core sample, as follows: A single rock core is divided into multiple columns according to pixels; Calculate the average reflectance of each column of b bands in the current single core sample. As a measure of signal strength, calculate the standard deviation of reflectivity across all bands in each column. As noise intensity; Calculate the ratio of signal strength to noise intensity. The weighted sum of the ratios and average reflectances calculated for each column is then used to select the column corresponding to the maximum value as the center column of the representative area of ​​the current single core. The method for determining the optimal width of the representative region is as follows: Define the candidate region within the candidate width set with the center column c as the axis of symmetry. The candidate width set is the set of widths from 1 to the theoretical distance D between adjacent cores; Calculate candidate regions separately Internal and external areas The spectral statistics include the within-class mean, within-class variance, combined between-class variance, and combined within-class variance; A width optimization objective function is constructed based on the combined inter-class variance and the combined intra-class variance. The optimal width parameter is obtained by iterating through the candidate width set and solving the objective function for width optimization.

2. The method for efficient hyperspectral logging of borehole cores based on representative regions as described in claim 1, characterized in that, A linear regression method was used to establish the conversion relationship between pixel coordinates and the actual depth of the core sample, specifically: Multiple calibration points were selected on each core sample, and the actual depth of each calibration point on the core sample and the pixel coordinates along the length of the core sample were recorded. Using pixel coordinates as the independent variable and actual depth as the dependent variable, a linear regression model was established. The model was fitted using multiple calibration points to obtain the conversion coefficient between the pixel coordinates and actual length of each core sample. The converted images were stitched together sequentially according to the actual depth of the core, with a filler band remaining at the joint between adjacent segments, to obtain a panoramic hyperspectral mosaic of the core with continuous depth.

3. The method for efficient hyperspectral logging of borehole cores based on representative regions as described in claim 1, characterized in that, The obtained core mosaic was masked to remove the wooden box, the label recording borehole number information, and locations with reflectivity below a set threshold. Specifically: Morphological operations are used to extract connected regions from an image, identifying peripheral connected regions with areas smaller than a set threshold and irregular shapes. Calculate the ratio of the average reflectance of the area near the wood absorption valley to the reflectance at the wood absorption valley. If the ratio is greater than a set value, mark this area as a wooden box. Calculate the ratio of the average reflectance of the area near the absorption valley of the tag to the reflectance at the absorption valley of the tag. If the ratio is greater than a set value, then this area is marked as a tag. Mark the pixels of the outer connected region, the wooden box, and the label as 0, and update the mask; Calculate the average reflectance of each pixel across the entire band, mark pixels with an average reflectance lower than a set threshold as 0, and update the mask again; The final result is a core mask image with the wooden box removed, the borehole number information recorded, and the reflectivity below a set threshold.

4. The method for efficient hyperspectral logging of borehole cores based on representative regions as described in claim 1, characterized in that, The method for determining the boundary of the single rock core is as follows: Determine the total number of columns W in the hyperspectral image and the total number N of core samples to be identified, and calculate the theoretical spacing between adjacent core samples. ; For the nth core, in its corresponding column interval Calculate the average reflectance of each column of b bands. ; Within the column interval, find the two column indices with the smallest average reflectance, which are the left and right boundaries of the nth core.

5. The method for efficient hyperspectral logging of borehole cores based on representative regions as described in claim 1, characterized in that, The width optimization objective function Specifically: ； in, To combine the variance between classes, To combine the within-class variance, This is the balance coefficient.

6. The method for efficient hyperspectral logging of borehole cores based on representative regions as described in claim 1, characterized in that, Calculate the relative abundance of different minerals within each depth range, specifically: Based on the conversion relationship between pixel coordinates and actual core depth, the depth range corresponding to the representative area of ​​each core segment is determined. The representative regions are sorted according to their depth. For the k-th depth interval Count the number of pixels for each mineral i within this interval. relative content ; This represents the total number of all valid mineral pixels within the k-th depth interval; All depth segments Arranged in depth order, the depth-relative content profile of mineral i is obtained.

7. A system for implementing the high-efficiency hyperspectral logging method for borehole cores based on representative regions as described in claim 1, characterized in that, include: The data acquisition module is used to acquire segmented hyperspectral image data of borehole cores; The core mosaic construction module is used to establish the conversion relationship between pixel coordinates and the actual depth of the core using a linear regression method, and to stitch together the hyperspectral images of all segments into a depth-continuous core mosaic. The image processing module is used to perform masking on the obtained core mosaic, removing the wooden box, the label recording the borehole number information, and the locations with reflectivity below a set threshold. The boundary segmentation module is used to calculate the average reflectance of each column in the corresponding column interval of a single rock core for the processed image. The column index with the lowest average reflectance is the boundary of the single rock core. The representative region determination module is used to find the center position and width of the optimal representative region within each core based on the boundary segmentation of a single core, and record the index. The mineral content calculation module is used to fill and crop representative areas with minerals, and then stitch the representative areas of the cropped mineral filling according to depth; it calculates the relative content of different minerals in each depth interval to obtain the depth and relative content profile of each mineral.

8. A terminal device comprising a processor and a memory, the processor for implementing instructions; the memory for storing multiple instructions, characterized in that, The instructions are adapted to be loaded by a processor and executed by the high-efficiency hyperspectral logging method for borehole cores based on any one of claims 1-6.

Images