Rock debris particle image instance segmentation method under white light condition
Through the efficient image segmentation method and superpixel merging technology based on graphs, the accuracy problem of rock chip particle image instance segmentation under white light conditions is solved, and the image segmentation of rock chip particle with high accuracy and strong adaptability is achieved, and manual labeling is avoided and high cost issues.
Patent Information
- Application Number
- CN202310600274.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-25
- Publication Date
- 2024-11-26
AI Technical Summary
It is difficult to accurately segment the example segmentation of rock chip particle images under white light conditions in the prior art. The traditional algorithm is not accurate, deep learning methods require a large amount of data annotation and are costly. The existing methods have limitations in applicability and accuracy.
Superpixel extraction and merging are used to calculate the dissimilarity through the European distance, and post-processing and merging are performed to achieve example segmentation of rock chip particles under white light conditions.
It improves the accuracy of instance segmentation of rock cutting particle images, reduces the cost of manual labeling, adapts to various types of rock cuttings and different lighting conditions, and has high accuracy and robustness.
Smart Images

Figure CN119027435A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of computer vision and image processing, and is a method for segmenting rock cuttings particle images under white light conditions. Background Art
[0002] Rock cuttings are rock particles that are ground or broken by the drill bit during drilling. They are brought out of the well by the circulating flushing fluid and are the basis for understanding the properties of rock formations, oil and gas displays, and compiling geological profiles. The rock cuttings returned from the well are generally aggregates, which are composed of multiple rock cutting particles. When describing rock cuttings, it is necessary to carefully observe each specific small particle to determine the color, sorting, roundness, and other information of the formation it represents.
[0003] In the process of intelligent description of rock cuttings based on image recognition, it is necessary to distinguish different rock cuttings from images containing a large number of rock cuttings, and accurately perform rock cutting instance segmentation of the image (identify the image area where each rock cutting is located) in order to perform image feature analysis on a single rock cutting particle. If the rock cutting instance segmentation granularity is too coarse, multiple rock cutting particles are identified as one rock cutting particle, and the features of multiple rock cutting particles will interfere with each other, causing confusion in image feature analysis and affecting the analysis results. If the rock cutting instance segmentation granularity is too fine, one rock cutting particle is segmented into multiple rock cutting particles, which will result in incomplete rock cutting particle image features, too many rock cutting particles, and affect the accuracy of extracting the shape and size features of rock cutting particles.
[0004] Traditional image segmentation algorithms have difficulty accurately segmenting large quantities of small, messy rock fragments. Although deep learning-based methods have high accuracy in various instance segmentation tasks, they require the collection and annotation of a large amount of data, which is not feasible in the problem involved in this patent.
[0005] A Chinese patent document with publication number CN113781503A discloses a rock chip image segmentation method and device. This method is an image segmentation method for rock chip, which uses an image processing method to generate a contour point sequence, and then identifies the rock chip based on the concave-convex characteristics of the contour point sequence. However, this technology has limitations in its scope of application, and is more dependent on the stable contour point sequence extraction results and the concave-convex characteristics of the segmentation target, and cannot achieve good results in the instance segmentation problem of rock chip particles.
[0006] A Chinese patent document with publication number CN112348831A discloses a shale SEM image segmentation method based on machine learning. This method is a shale SEM image segmentation method based on filter features and random forest models. This method is applicable to single-channel SEM images and has not been verified to be effective under white light conditions. In addition, this method is mainly aimed at segmenting shale pores, organic matter, inorganic skeletons and other targets, and has limited applicability to the instance segmentation problem of rock debris particles.
[0007] The Chinese patent document with publication number CN114022680A discloses a method for segmenting rock cuttings images under PDC drill bit conditions with an improved deep learning algorithm. The method is an instance segmentation method based on the deep learning model MTLU-Net++, which adopts a multi-task learning framework. The method requires a large number of rock cutting particle images to be collected, rock cutting particle annotation, and then the model training is performed using a high-performance floating point unit, which has a high implementation cost. In particular, the rock cutting particle annotation work requires a lot of labor costs for fine-grained annotation.
[0008] The Chinese patent document with publication number CN109859224A discloses a conveyor belt rock particle image segmentation method. Similar to CN114022680A, this method requires high labor costs for labeling and high computational costs. In addition, this method reduces the image resolution in order to increase the recognition speed, which may result in the inability to accurately identify rock particles under certain conditions, especially when it is necessary to accurately identify their volume and position. Summary of the invention
[0009] The present invention provides a method for instance segmentation of rock cuttings particle images under white light conditions, which overcomes the deficiencies of the above-mentioned prior art and can effectively solve the problem that the traditional image segmentation algorithm has low accuracy in instance segmentation of rock cuttings images.
[0010] The technical solution of the present invention is achieved by the following measures: A method for segmenting rock debris particle images under white light conditions is performed according to the following steps:
[0011] S1, superpixel extraction, collects rock cuttings particle images under white light conditions, and uses an efficient graph-based image segmentation method to perform superpixel segmentation on the rock cuttings particle images;
[0012] S2, superpixel color similarity merging, uses the superpixel merging method based on color similarity to post-process the segmentation results of the graph-based efficient image segmentation method and merge the superpixels belonging to the same rock fragment;
[0013] S3, superpixel morphological similarity merging, uses a superpixel merging method based on morphological similarity to merge superpixels with different colors belonging to the same rock cuttings, completing the instance segmentation of rock cuttings particle images under white light conditions.
[0014] The following are further optimizations and / or improvements to the above technical solutions:
[0015] In the above step S1, the specific operation of performing superpixel segmentation on the image by the efficient graph-based image segmentation method is as follows:
[0016] S11, for each pixel in the rock particle image, calculates the dissimilarity between it and the eight adjacent pixels. The dissimilarity is calculated using the Euclidean distance of the RGB three-channel pixel values. i With a neighbor v j The connection is recorded as an edge e of the graph G, and the weight of the edge is the dissimilarity. For each edge v i Assign a tag, initially Id(v i );
[0017] S12, sort all the edges of all graphs G in ascending order according to dissimilarity to obtain E = {e1, e2, ..., e N};
[0018] S13, ask the edge e with the smallest weight in E = <v i , v j >, if the merge condition Id(v i )≠Id(v j ) and the dissimilarity of e is greater than Id(v i )、Id(v j ) internal dissimilarity, then execute step S14;
[0019] S14, Id(v j ) is assigned to Id(v i ), update the dissimilarity threshold of this class, and set e i Removed from E;
[0020] S15, if the number of elements of E is greater than or equal to the preset number of instances, execute step S13, otherwise end.
[0021] In the above step S2, the specific operation of post-processing the segmentation result of the graph-based efficient image segmentation method using the superpixel merging method based on color similarity is as follows:
[0022] In S21, for each superpixel obtained by performing superpixel segmentation calculation on the rock cuttings particle image using the graph-based efficient image segmentation method in step S1, a color distribution feature is calculated;
[0023] S22, for every two adjacent superpixels S i and Sj Calculate their color similarity respectively;
[0024] S23, assuming that step S1 extracts M1 superpixels in total, and the color similarity of adjacent superpixels coexisting in P1 is calculated according to a preset threshold τ color , 0≤τ color <1, select the color with the largest similarity in group P1 According to the selected high-similarity superpixel combinations, superpixel merging is performed respectively. The superpixel merging is to take S i and S j The union of S is added to the superpixel segmentation result, and S is removed i and S j , after superpixels are merged, M2 superpixels remain, M2≤M1.
[0025] In the above step S3, the specific operation of merging superpixels with different colors belonging to the same rock fragment using the superpixel merging method based on morphological similarity is as follows:
[0026] S31, for each superpixel S calculated in step S2, calculate the upper, lower, left and right boundaries S of the superpixel region ymin ,S ymax ,S xmin ,S xmax ;
[0027] S32, for each pair of adjacent superpixels S i and S j Calculate its morphological similarity ShapeSim i,j ;
[0028] S33, assuming that after step S2, there are M2 superpixels remaining, and there are P2 pairs of adjacent superpixels. According to a preset threshold τ shape , 0≤τ shape <1, select the morphological similarity in group P2 greater than τ shape and merge these superpixels, leaving M3 superpixels after merging.
[0029] In the above step S21, the specific operation of calculating the color distribution feature is to count the RGB three-channel colors of all pixels in the area covered by the superpixel, obtain a distribution histogram and perform a normalization equation:
[0030]
[0031] In formula 1:
[0032] is the number of pixels in the superpixel with color c and non-pixel value k (0≤k≤255), where c is one of R, G, and B.
[0033] is a vector obtained by splicing color histograms.
[0034] The final distribution histogram is a vector obtained by splicing the color histograms of three channels.
[0035] In the above step S22, the color similarity ColorSim i,j is calculated according to the following formula:
[0036]
[0037] In formula 2:
[0038] i < j, and S i and S j are adjacent,
[0039] h i is the color histogram vector extracted from the i-th superpixel in step S21.
[0040] h j is the color histogram vector extracted from the j-th superpixel in step S21.
[0041] In the above step S32, the shape similarity ShapeSim i,j is calculated according to the following formula:
[0042]
[0043] In formula 3: i < j, and S i and S j are adjacent,
[0044] S i,xmax and S i,xmin are the right and left boundary coordinates of the i-th superpixel extracted in step S31, respectively.
[0045] S i,ymax and S i,ymin are the lower and upper boundary coordinates of the i-th superpixel extracted in step S31, respectively.
[0046] S j,xmax and S j,xmin are the right and left boundary coordinates of the j-th superpixel extracted in step S31, respectively.
[0047] S j,ymax and S j,ymin are the lower and upper boundary coordinates of the j-th superpixel extracted in step S31, respectively.
[0048] This patent studies the instance image features of rock cuttings particle images under white light conditions and invents an instance segmentation method for rock cuttings particle images under white light conditions based on superpixel segmentation and merging, which solves the problem that it is difficult to accurately perform instance segmentation of rock cuttings particle images under white light conditions. The invention has the advantages of high accuracy, no need for manual data annotation, adaptability to various types of rock cuttings, insensitivity to shooting environments such as lighting angles, and strong robustness. BRIEF DESCRIPTION OF THE DRAWINGS
[0049] Attached Figure 1 This is the rock cuttings particle image processed by step S1 in the rock cuttings particle image instance segmentation method under white light conditions in embodiment 2 of the present invention.
[0050] Attached Figure 2 This is the rock cuttings particle image processed by step S2 in the rock cuttings particle image instance segmentation method under white light conditions in Example 3 of the present invention.
[0051] Attached Figure 3 This is the rock cuttings particle image processed by step S3 in the rock cuttings particle image instance segmentation method under white light conditions in embodiment 4 of the present invention.
[0052] Attached Figure 4 This is the rock cuttings particle image processed by step S1 in the rock cuttings particle image instance segmentation method under white light conditions in Example 8 of the present invention.
[0053] Attached Figure 5 This is the rock cuttings particle image processed by step S2 in the rock cuttings particle image instance segmentation method under white light conditions in Example 8 of the present invention.
[0054] Attached Figure 6 This is the rock cuttings particle image processed by step S3 in the rock cuttings particle image instance segmentation method under white light conditions in Example 8 of the present invention. DETAILED DESCRIPTION
[0055] The present invention is not limited by the following embodiments, and specific implementation methods can be determined based on the technical solution of the present invention and actual conditions.
[0056] The present invention will be further described below in conjunction with embodiments:
[0057] Example 1: The example segmentation method of rock debris particle image under white light conditions is performed according to the following steps:
[0058] S1, superpixel extraction, collect rock cuttings particle images under white light conditions, and use the efficient graph-based image segmentation method to perform superpixel segmentation on the rock cuttings particle images;
[0059] S2, superpixel color similarity merging, uses the superpixel merging method based on color similarity to post-process the segmentation results of the graph-based efficient image segmentation method and merge the superpixels belonging to the same rock fragment;
[0060] S3, superpixel morphological similarity merging, uses a superpixel merging method based on morphological similarity to merge superpixels with different colors belonging to the same rock cuttings, completing the instance segmentation of rock cuttings particle images under white light conditions.
[0061] Embodiment 2: As an optimization of the above embodiment, in step S1, the specific operation of performing superpixel segmentation on the image by the efficient graph-based image segmentation method is as follows:
[0062] S11, for each pixel in the rock particle image, calculates the dissimilarity between it and the eight adjacent pixels. The dissimilarity is calculated using the Euclidean distance of the RGB three-channel pixel values. i With a neighbor v j The connection is recorded as an edge e of the graph G, and the weight of the edge is the dissimilarity. For each edge v i Assign a tag, initially Id(v i );
[0063] S12, sort all the edges of all graphs G in ascending order according to dissimilarity to obtain E = {e1, e2, ..., e N};
[0064] S13, ask the edge e with the smallest weight in E = <v i , v j >, if the merge condition Id(v i )≠Id(v j ) and the dissimilarity of e is greater than Id(v i )、Id(v j ) internal dissimilarity, then execute step S14;
[0065] S14, Id(v i ) is assigned to Id(v i ), update the dissimilarity threshold of this class, and set e i Removed from E;
[0066] S15, if the number of elements of E is greater than or equal to the preset number of instances, execute step S13, otherwise end.
[0067] The present invention uses an efficient graph-based image segmentation method (Efficient Graph-based Image Segmentation) to perform superpixel segmentation on an image. This method can be referred to in the literature reference: Felzenszwalb, Pedro F., and Daniel P. Huttenlocher. "Efficient graph-based image segmentation." International journal of computer vision 59.2 (2004): 167-181. The result after the above steps S11 to S15 is a set of multiple pixels, that is, the image segmentation result after the step S1 of the present invention is called a superpixel. Figure 1 shown.
[0068] Embodiment 3: As an optimization of the above embodiment, in step S2, the specific operation of post-processing the segmentation result of the graph-based efficient image segmentation method using the superpixel merging method based on color similarity is as follows:
[0069] In S21, for each superpixel obtained by performing superpixel segmentation calculation on the rock cuttings particle image using the graph-based efficient image segmentation method in step S1, a color distribution feature is calculated;
[0070] S22, for every two adjacent superpixels S i and S j Calculate their color similarity respectively;
[0071] S23, assuming that step S1 extracts M1 superpixels in total, and the color similarity of adjacent superpixels coexisting in P1 is calculated according to a preset threshold τ color , 0≤τ color <1, select the color with the largest similarity in group P1 According to the selected high-similarity superpixel combinations, superpixel merging is performed respectively. The superpixel merging is to take S i and S j The union of S is added to the superpixel segmentation result, and S is removed i and S j , after superpixels are merged, M2 superpixels remain, M2≤M1.
[0072] Depend on Figure 1It can be seen that after the instance segmentation of the rock chip image, there is a serious over-segmentation phenomenon, that is, the same rock chip instance is segmented into several superpixels. Therefore, the present invention uses a superpixel merging method based on color similarity to post-process the segmentation results of the efficient image segmentation method based on the graph, and merge the superpixels belonging to the same rock chip. After the above steps S21 to S23, after the color similarity is merged, as shown in Figure 2 shown.
[0073] Embodiment 4: As an optimization of the above embodiment, in step S3, the specific operation of merging superpixels with different colors belonging to the same rock fragment by using the superpixel merging method based on morphological similarity is as follows:
[0074] S31, for each superpixel S calculated in step S2, calculate the upper, lower, left and right boundaries S of the superpixel region ymin , S ymax , S xmin , S xmax ;
[0075] S32, for each pair of adjacent superpixels S i and S i Calculate its morphological similarity ShapeSim i,j ;
[0076] S33, assuming that after step S2, there are M2 superpixels remaining, and there are P2 pairs of adjacent superpixels. According to a preset threshold τ shape , 0≤τ shape <1, select the morphological similarity in group P2 greater than τ shape and merge these superpixels, leaving M3 superpixels after merging.
[0077] In the rock cutting particle image processed by step S2, although some adjacent superpixels have certain color differences, they are mutually inclusive in morphology and essentially belong to different parts of the same rock cutting. Therefore, this patent uses a superpixel merging method based on morphological similarity to merge superpixels with different colors belonging to the same rock cutting. The final output of the rock cutting instance segmentation result of the present invention is M3 superpixels, such as Figure 3 shown.
[0078] Embodiment 5: As an optimization of the above embodiment, in step S21, the specific operation of calculating the color distribution feature is to count the RGB three-channel colors of all pixels in the area covered by the superpixel, obtain a distribution histogram and perform a normalization equation:
[0079]
[0080] In formula 1:
[0081] is the number of pixels with color c and non-pixel value k (0 ≤ k ≤ 255) in a superpixel, where c is one of R, G, B.
[0082] is the vector obtained by splicing color histograms.
[0083] The final distribution histogram is the vector obtained by splicing the color histograms of the three channels.
[0084] Example 6: As an optimization of the above example, in step S22, the color similarity ColorDim i,j is calculated according to the following formula:
[0085]
[0086] In formula 2:
[0087] i < j, and S i is adjacent to S j adjacent,
[0088] h i is the color histogram vector of the i-th superpixel extracted in step S21.
[0089] h j is the color histogram vector of the j-th superpixel extracted in step S21.
[0090] Example 7: As an optimization of the above example, in step S32, the shape similarity ShapeSim i,j is calculated according to the following formula:
[0091]
[0092] In formula 3: i < j, and S i and S j adjacent,
[0093] S i,xmax and S i,xmin are the right and left boundary coordinates of the i-th superpixel extracted in step S31, respectively.
[0094] S i,ymax and S i,ymin are the lower and upper boundary coordinates of the i-th superpixel extracted in step S31, respectively.
[0095] S j,xmax and S j,xmin are the right and left boundary coordinates of the j-th superpixel extracted in step S31, respectively.
[0096] Sj,ymax , S j,ymin They are respectively the lower and upper boundary coordinates of the j-th superpixel extracted in step S31.
[0097] In the present invention, when processing a digital image, a plane rectangular coordinate system is established with the upper left corner of the image as the origin, the x-axis from left to right, and the y-axis from top to bottom for analysis.
[0098] Example 8: Using the rock cuttings particle image instance segmentation method under white light conditions of the present invention, an image of rock cuttings particles under white light conditions is segmented by instance:
[0099] S1, superpixel extraction, collect rock cuttings particle images under white light conditions, and use the efficient graph-based image segmentation method (Efficient Graph-based Image Segmentation) to perform superpixel segmentation on the rock cuttings particle images. After superpixel segmentation in step S1, the obtained over-segmented superpixels are as follows Figure 4 As shown, in step S1, a total of M1 superpixels M1=9328 are extracted.
[0100] S2, superpixel color similarity merging, using the superpixel merging method based on color similarity, post-processes the segmentation results of the graph-based efficient image segmentation method to merge superpixels belonging to the same rock fragment. In step S2, set τ color =0.1%. After merging the superpixel color similarity, the number of merged superpixels M2 = 4127 is obtained. Figure 5 shown.
[0101] S3, superpixel morphological similarity merging, using the superpixel merging method based on morphological similarity, merges superpixels with different colors belonging to the same rock fragment to complete the instance segmentation of the rock fragment particle image under white light conditions. In step S3, set τ shape =0.7, after merging the superpixel color similarity, the number of merged superpixels M3 = 1258 is obtained. Figure 6 shown.
[0102] According to the estimation of real data sampling, after the instance segmentation of the rock cuttings particle image under white light conditions using the present invention, the accuracy rate of the rock cuttings instance segmentation (the correct proportion in the segmentation results) is 98.4%, and the recall rate (the correct proportion of the real rock cuttings segmented correctly) is 99.1%.
[0103] The present invention studies the instance image features of rock cuttings particle images under white light conditions, and provides a method for instance segmentation of rock cuttings particle images under white light conditions based on superpixel segmentation and merging, which solves the problem that it is difficult to accurately perform instance segmentation of rock cuttings particle images under white light conditions. The present invention has the advantages of high accuracy, no need for manual data annotation, adaptability to various types of rock cuttings, insensitivity to shooting environments such as illumination angles, and strong robustness.
[0104] The above technical features constitute the embodiments of the present invention, which have strong adaptability and implementation effect. Non-essential technical features can be added or reduced according to actual needs to meet the requirements of different situations.
Claims
1. A method for instance segmentation of rock debris particle images under white light conditions, characterized in that Follow these steps: S1, superpixel extraction, collects rock cuttings particle images under white light conditions, and uses an efficient graph-based image segmentation method to perform superpixel segmentation on the rock cuttings particle images; S2, superpixel color similarity merging, uses the superpixel merging method based on color similarity to post-process the segmentation results of the graph-based efficient image segmentation method and merge the superpixels belonging to the same rock fragment; S3, superpixel morphological similarity merging, uses a superpixel merging method based on morphological similarity to merge superpixels with different colors belonging to the same rock cuttings, completing the instance segmentation of rock cuttings particle images under white light conditions.
2. The method for instance segmentation of rock debris particle images under white light conditions according to claim 1, characterized in that In the above step S1, the specific operation of performing superpixel segmentation on the image by the efficient image segmentation method based on the graph is: S11, for each pixel in the rock particle image, calculates the dissimilarity between it and the eight adjacent pixels. The dissimilarity is calculated using the Euclidean distance of the RGB three-channel pixel values. i With a neighbor v j The connection is recorded as an edge e of the graph G, and the weight of the edge is the dissimilarity. For each edge v i Assign a tag, initially Id(v i ); S12, sort all the edges of all graphs G in ascending order according to dissimilarity to obtain E = {e1, e2, ..., e N }; S13, ask the edge e with the smallest weight in E = <v i , v j >, if the merge condition Id(v i )≠Id(v j ) and the dissimilarity of e is greater than Id(v i )、Id(v j ) internal dissimilarity, then execute step S14; S14, Id(v j ) is assigned to Id(v i ), update the dissimilarity threshold of this class, and set e i Removed from E; S15, if the number of elements of E is greater than or equal to the preset number of instances, execute step S13, otherwise end.
3. The method for instance segmentation of rock debris particle images under white light conditions according to claim 1 or 2, characterized in that In step S2, the specific operation of post-processing the segmentation result of the graph-based efficient image segmentation method using the superpixel merging method based on color similarity is as follows: In S21, for each superpixel obtained by performing superpixel segmentation calculation on the rock cuttings particle image using the graph-based efficient image segmentation method in step S1, a color distribution feature is calculated; S22, for every two adjacent superpixels S i and S j Calculate their color similarity respectively; S23, assuming that step S1 extracts M1 superpixels in total, and the color similarity of adjacent superpixels coexisting in P1 is calculated according to a preset threshold τ color , 0≤τ color <1, select the color with the largest similarity in group P1 According to the selected high-similarity superpixel combinations, superpixel merging is performed respectively. The superpixel merging is to take S i and S j The union of S is added to the superpixel segmentation result, and S is removed i and S i , after superpixels are merged, M2 superpixels remain, M2≤M1.
4. The method for instance segmentation of rock debris particle images under white light conditions according to any one of claims 1 to 3, characterized in that In step S3, the specific operation of merging superpixels with different colors belonging to the same rock fragment using the superpixel merging method based on morphological similarity is as follows: S31, for each superpixel S calculated in step S2, calculate the upper, lower, left and right boundaries S of the superpixel region ymin , S ymax , S xmin , S xmax ; S32, for each pair of adjacent superpixels S i and S j Calculate its morphological similarity ShapeSim i,j ; S33, assuming that after step S2, there are M2 superpixels remaining, and there are P2 pairs of adjacent superpixels. According to a preset threshold τ shape , 0≤τ shape <1, select the morphological similarity in group P2 greater than τ shape and merge these superpixels, leaving M3 superpixels after merging.
5. The method for instance segmentation of rock debris particle images under white light conditions according to claim 3, characterized in that In step S21, the specific operation of calculating the color distribution feature is to count the RGB three-channel colors of all pixels in the area covered by the superpixel, obtain a distribution histogram and perform a normalization equation: In formula 1: is the number of pixels with color c and non-pixel value k in the superpixel, where c is one of R, G, and B, 0≤k≤255, is the vector obtained by concatenating the color histograms, The final distribution histogram is the vector obtained by concatenating the color histograms of the three channels.
6. The method for instance segmentation of rock debris particle images under white light conditions according to claim 3, characterized in that In step S22, the color similarity ColorSim i,j Calculate as follows: In formula 2: i<j, and S i With S j Adjacent, h i is the color histogram vector of the i-th superpixel extracted in step S21, h j is the color histogram vector of the j-th superpixel extracted in step S21.
7. The method for instance segmentation of rock debris particle images under white light conditions according to claim 4, characterized in that In step S32, the morphological similarity ShapeSim i,j Calculate as follows: In formula 3: i<j, and S i and S j Adjacent, S i,xmax , S i,xmin are the right and left boundary coordinates of the i-th superpixel extracted in step S31, S i,ymax , S i,ymin are the lower and upper boundary coordinates of the i-th superpixel extracted in step S31, S j,xmax , S j,xmin are the right and left boundary coordinates of the j-th superpixel extracted in step S31, S j,ymax , S j,ymin They are respectively the lower and upper boundary coordinates of the j-th superpixel extracted in step S31.
Citation Information
Patent Citations
Ore particle segmentation method
CN108470173A
Rock debris lithology identification method under condition of PDC drill bit matched with lithology feature library
CN111709423A
First arrival pickup method and device based on deep learning network
CN112464725A
Cited By
Quality detection method of cashmere product
CN120318215A
A quality detection method for cashmere products
CN120318215B
