Surface quality detection method for mining anchor cable steel strand after stabilization treatment

Abstract

The invention relates to the technical field of image data processing, in particular to a method for detecting the surface quality of a mining anchor cable steel strand after stabilization treatment, and the method comprises the steps: obtaining a surface image of a to-be-detected steel strand; determining a local texture collaboration degree index of each pixel point; determining a longitudinal consistency deviation index of each pixel point; determining a defect significance index of each pixel point; and identifying a defect area in the surface image of the to-be-detected steel strand based on the defect saliency index of each pixel point. According to the method, a local texture synergy degree and longitudinal consistency deviation index is constructed, and the gray change characteristics and the spatial change rate are combined, so that multi-dimensional quantitative detection of small defects on the surface of the steel strand is realized, noise and real defects are effectively distinguished, the sensitivity to low-contrast defects is improved, missing detection and false alarm are reduced, and the detection accuracy is improved. And a high-reliability quality detection method is provided for the mining anchor cable steel strand.

Description

Technical Field

[0001] The present invention relates to the technical field of image data processing, and more particularly to a method for detecting the surface quality of a stabilized mining anchor cable steel strand. Background Art

[0002] As the core load-bearing component of the support system for underground projects such as coal mine tunnels, the manufacturing quality of mining anchor cable steel strands directly determines the long-term stability and safety of the project. In the production process of steel strands, stabilization treatment is a key link. This treatment eliminates internal stress in the material through high-temperature stretching and shaping, significantly improving its mechanical properties. However, this high-temperature and high-stress process inevitably introduces various subtle defects on the surface of the steel strands, such as tiny oxidation spots, shallow scratches, and fine pitting caused by localized stress imbalances. These defects are tiny in size and have extremely low contrast with the background of the steel strands themselves, but they are potential stress concentration points and rust initiation points, posing a serious safety hazard to the long-term service performance and fatigue life of the anchor cables.

[0003] Currently, the industry is widely adopting automated inspection technologies based on machine vision to replace manual visual inspections, improving inspection efficiency and consistency. These technologies typically use industrial cameras to capture images of the steel strand surface and then utilize image processing algorithms or deep learning models to identify defects.

[0004] However, steel strands themselves are composed of multiple spirally twisted steel wires, and their surfaces naturally exhibit complex and regular periodic textures. The visual characteristics of subtle defects (such as grayscale variations and shape features) are often very similar to the texture characteristics of the steel strand itself, and are easily overwhelmed by the strong background texture. This results in high missed detection rates for traditional algorithms based on fixed thresholds, edge detection, or morphological processing. Furthermore, while deep learning-based detection methods excel at detecting specific defect types, their ability to identify small, low-contrast defects remains limited. Deep learning models are prone to overfitting normal textures, misidentifying normal texture fluctuations as defects, resulting in a high false alarm rate. Furthermore, because subtle defects are low-probability events in actual production and defect samples are scarce, the model struggles to fully learn their effective features, leading to missed detections and impacting the reliability of quality inspection for mining anchor steel strands. Summary of the Invention

[0005] In order to solve the technical problems of high missed detection rate and high false alarm rate of subtle defects caused by strong texture interference of steel strands themselves, the present invention proposes a surface quality detection method for mining anchor cable steel strands after stabilization treatment, which includes the following steps: Acquire a surface image of the steel strand to be inspected; record any pixel point in the surface image as a target pixel point, and determine a local texture coordination index of the target pixel point for characterizing the degree of consistency of the local texture direction according to the texture directions of each pixel point in a preset neighborhood of the target pixel point; determine a longitudinal consistency deviation index of the target pixel point for characterizing the degree of local grayscale mutation according to the grayscale distribution difference of multiple pixel blocks along the texture direction of the target pixel point and the local texture coordination index; determine a defect significance index of the target pixel point according to the longitudinal consistency deviation index and the spatial variation rate of the longitudinal consistency deviation index; and identify the defect area in the surface image of the steel strand to be inspected based on the defect significance index of each pixel point.

[0006] The present invention constructs a local texture synergy index by analyzing the texture direction of each pixel point in the neighborhood of the target pixel point, which can effectively quantify the directional consistency characteristics of the surface texture of the steel strand, thereby distinguishing normal texture areas from abnormal areas, and providing a reliable background reference for subsequent defect detection; by analyzing the grayscale distribution differences of multiple pixel blocks along the texture direction and combining the local texture synergy index, a longitudinal consistency deviation index is constructed, which can effectively capture the grayscale mutation characteristics along the texture direction, avoid the misjudgment that may be caused by traditional omnidirectional detection, and improve the sensitivity to subtle defects; through the defect significance index, the longitudinal consistency deviation and its spatial variation rate are comprehensively considered to realize the multi-dimensional quantitative evaluation of defect characteristics. The introduction of the spatial variation rate can effectively distinguish local noise from real defects, avoiding the false alarm problem that may be caused by a single indicator; the regular characteristics of the surface texture of the steel strand are fully utilized, and the detection ability of tiny and low-contrast defects is improved through the organic combination of texture synergy and grayscale consistency deviation, which effectively solves the problems of high missed detection rate of traditional methods and high false alarm rate of deep learning methods, and provides a more reliable technical guarantee for the quality inspection of mining anchor steel strands.

[0007] Furthermore, the method of obtaining the surface image of the steel strand to be detected includes: obtaining images of multiple perspectives distributed around the axis of the steel strand through an image acquisition device; splicing the images of multiple perspectives and performing coordinate transformation to expand the cylindrical steel strand surface image into a two-dimensional plane image to obtain the surface image of the steel strand to be detected.

[0008] Furthermore, the local texture synergy index is obtained by constructing a neighborhood window with the target pixel point as the center; calculating the local texture direction angle of each pixel point in the neighborhood window; performing frequency multiplication processing on the local texture direction angle and mapping it into a complex unit vector; performing vector summation and averaging on all complex unit vectors in the neighborhood window, and taking the modulus of the averaged vector as the local texture synergy index of the target pixel point.

[0009] The present invention can effectively capture the local directional characteristics of the surface texture of the steel strand by constructing a neighborhood window and calculating the local texture direction angle of each pixel point; the frequency doubling processing maps the texture direction angle to the complex domain, fully utilizes the periodic characteristics of the texture, and enhances the sensitivity of directional consistency detection; by mapping the texture direction angle after frequency doubling processing to a complex unit vector and performing vector summing and averaging, a quantitative index of local texture coordination is constructed, which can effectively distinguish normal areas with consistent texture directions from abnormal areas with chaotic texture directions. The closer the vector modulus is to 1, the higher the texture coordination is, and the closer it is to 0, the more chaotic the texture direction is.

[0010] Furthermore, the local texture direction angle is obtained as follows: for each pixel point in the neighborhood window, a Hessian matrix of the pixel point is constructed; the Hessian matrix of the pixel point is decomposed into eigenvalues ​​to obtain the eigenvector corresponding to the minimum eigenvalue; the direction of the eigenvector corresponding to the minimum eigenvalue is used as the local texture direction of the pixel point, and the angle between the local texture direction and the preset coordinate axis is calculated to obtain the texture direction angle of the pixel point.

[0011] Furthermore, the longitudinal consistency deviation index is obtained as follows: a rectangular pixel block is constructed with the target pixel point as the center as the center pixel block, and along the local texture direction of the target pixel point, multiple path pixel blocks of the same size as the center pixel block are constructed on the front and back sides of the center pixel block; based on the grayscale histograms of the multiple path pixel blocks, the expected grayscale histogram is calculated; based on the difference between the grayscale histogram of the center pixel block and the expected grayscale histogram, the dispersion difference moment of the target pixel point is calculated, and the longitudinal consistency deviation index of the target pixel point is determined by combining the dispersion difference moment of the target pixel point with the local texture coordination index.

[0012] The present invention constructs a central pixel block and multiple path pixel blocks along the local texture direction, thereby achieving precise tracking of the texture direction, effectively capturing the grayscale change characteristics along the texture direction, and avoiding the interference that may be caused by traditional omni-directional analysis; by calculating the expected grayscale histogram of multiple path pixel blocks as a reference benchmark, it can accurately reflect the grayscale distribution law of the normal texture area, and the difference calculation (dispersion difference moment) between the central pixel block and the expected histogram can effectively identify local grayscale anomalies, thereby improving the detection sensitivity of subtle defects; combining the dispersion difference moment with the local texture synergy index to determine the longitudinal consistency deviation index, it realizes the organic combination of texture background and grayscale anomaly. When the texture synergy is high, it indicates that the texture regularity of the area is strong, and any grayscale anomaly is more likely to be a real defect. When the texture synergy is low, it indicates that the texture of the area itself is chaotic, and a higher grayscale anomaly threshold is required to avoid false alarms.

[0013] Furthermore, the calculation to obtain the expected grayscale histogram includes: performing Gaussian weighted averaging based on the distance between each path pixel block and the center pixel block on actual grayscale histograms of the plurality of path pixel blocks to obtain the expected grayscale histogram.

[0014] Furthermore, the longitudinal consistency deviation index satisfies: Where, Pixel The longitudinal consistency deviation index, Pixel The dispersion difference moment of Pixel The local texture coordination index, is the hyperbolic tangent function.

[0015] The present invention performs nonlinear mapping of the dispersion difference moment through the hyperbolic tangent function, realizes the saturation processing of the longitudinal consistency deviation index, avoids the interference of extreme values ​​on subsequent processing, and maintains sensitivity to moderate anomalies; through the organic combination of the dispersion difference moment and texture synergy, the ability to recognize minor defects and resist noise interference is improved, effectively solving the problem of coexistence of missed detection and false alarm in the surface quality inspection of steel strands.

[0016] Furthermore, the dispersion difference moment satisfies: Where, Pixel The dispersion difference moment of is the maximum gray level, is the index of gray level, is the average gray level of the expected gray histogram of the central pixel block, is the actual histogram of the central pixel block in gray level The numerical value of is the expected grayscale histogram of the central pixel block in grayscale The numerical value of is the absolute value symbol.

[0017] The present invention calculates the absolute value of the difference between the actual histogram of the central pixel block and the expected histogram at each grayscale level, and performs weighted summation with the difference between the grayscale and the average grayscale value as the weight, thereby constructing a quantization model of the dispersion difference moment. This model can effectively measure the degree of deviation of the grayscale distribution of the central pixel block from the normal texture area. The weight factor reflects the influence of the degree of grayscale deviation from the center. The grayscale difference that deviates further from the average grayscale has a greater contribution to the dispersion difference moment, which conforms to the physical law that defect characteristics are usually manifested as extreme grayscale value changes. The absolute value operation ensures that both positive and negative differences can effectively contribute to the difference measurement.

[0018] Furthermore, the defect significance index satisfies: Where, Pixel The defect significance index, Pixel The longitudinal consistency deviation index, for The gradient vector of for The mold length.

[0019] The present invention realizes the organic combination of local anomaly intensity and spatial variation rate by taking the product of the longitudinal consistency deviation index and its gradient modulus as the defect significance index; the longitudinal consistency deviation index reflects the degree to which the pixel point deviates from the normal texture, and the gradient modulus characterizes the severity of the spatial variation of the anomaly; the product form ensures that only areas with high-intensity anomalies and significant spatial variations will obtain high significance scores, and a simple high deviation value may be caused by noise, and a simple high gradient value may be generated by a normal texture boundary; by comprehensively considering the anomaly intensity and spatial variation characteristics, the accuracy and robustness of defect detection are improved, providing reliable support for the detection of the surface quality of mining anchor steel strands.

[0020] Furthermore, the method of identifying defect areas in the surface image of the steel strand to be inspected based on the defect significance index of each pixel point includes: generating a significance map composed of the defect significance index of each pixel point; performing adaptive threshold segmentation on the significance map using the maximum inter-class variance method to obtain a binary image; in the binary image, determining the area where the pixel value is identified as a defect as the defect area in the surface image of the steel strand to be inspected, thereby completing the surface quality detection method of the mining anchor steel strand after stabilization treatment.

[0021] The present invention has the following beneficial effects: (1) The traditional algorithm is difficult to distinguish between subtle defects and the spiral texture of the steel strand. The consistency of the texture direction is quantified by the local texture coordination index. The normal texture area presents highly consistent directional characteristics due to the spiral structure, while the defect area will destroy this coordination. Combined with the longitudinal consistency deviation index to capture the grayscale mutation along the texture direction, it can effectively separate subtle defects (such as tiny scratches and pits) from the periodic texture background, solve the problem of missed detection caused by defects being submerged by the texture, and improve the recognition sensitivity of low-contrast and small-size defects.

[0022] (2) By integrating the longitudinal consistency deviation and its spatial variation rate with the defect significance index, the feature of "real defects have local grayscale mutations and continuous spatial distribution" is strengthened, while the periodic grayscale fluctuations of normal textures are weakened. Compared with the deep learning model that easily misjudges texture fluctuations as defects, it can accurately filter out pseudo-defect signals caused by uneven texture and reduce the false alarm rate. It is especially suitable for scenarios where the surface texture of steel strands is complex but the defect characteristics are weak.

[0023] (3) There is no need to rely on a large number of labeled defect samples for model training. Instead, detection indicators are constructed based on the inherent characteristics of steel strand texture (directional consistency, grayscale distribution law), which solves the problem of insufficient model learning caused by the scarcity of subtle defect samples in actual production. It has low requirements on the number of samples and can be directly applied to real-time detection on the production line, adapting to the industrial scenario needs of mining steel strand quality inspection.

[0024] (4) By accurately identifying minor surface defects, potential quality risks (such as stress concentration points and material damage) can be discovered in a timely manner after the steel strands are stabilized, thus preventing unqualified products from being used in mine anchor cable projects. This not only improves the quality control accuracy of steel strand production, but also reduces the risks of mine support failure and safety accidents caused by material defects. At the same time, it reduces the rework costs caused by missed inspections and misjudgments, thereby improving production efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 The present invention is a flowchart of the steps of a method for detecting the surface quality of a mining anchor cable steel strand after stabilization treatment according to an embodiment of the present invention. DETAILED DESCRIPTION

[0026] The following is a clear and complete description of the technical solutions in the embodiments of the present invention. The described embodiments are part of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative efforts are within the scope of protection of the present invention.

[0027] The specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0028] See also Figure 1 , which shows a flowchart of a method for detecting the surface quality of a mining anchor cable steel strand after stabilization treatment provided by one embodiment of the present invention, the method comprising the following steps: S1: Acquire a surface image of the steel strand to be inspected.

[0029] It should be noted that this step aims to obtain two-dimensional image data that comprehensively reflects the steel strand's surface condition and is easily processed. Because the steel strand is cylindrical, a single camera captured from a single angle cannot capture complete surface information. Therefore, in this embodiment, multiple industrial cameras can be arranged in a circular pattern, or a linear array camera can be used in conjunction with the steel strand's rotation to capture an annular field of view image array centered on the strand's axis, thereby obtaining full-circumferential image data of the strand's surface.

[0030] Specifically, obtaining the surface image of the steel strand to be inspected includes: Using an image acquisition device, images of multiple viewing angles distributed around the axis of the steel strand are acquired; The images from multiple perspectives are stitched together and coordinate transformed (to eliminate perspective distortion and uneven lighting problems caused by the cylindrical surface). The cylindrical steel strand surface image is expanded into a two-dimensional plane image to obtain the surface image of the steel strand to be inspected.

[0031] S2: Determine the local texture coordination index of each pixel point.

[0032] It should be noted that this step is based on a priori knowledge of the physical structure of steel strands. In defect-free areas, the surface texture of a steel strand is primarily composed of regularly arranged steel wires, so its local texture directions should exhibit a high degree of synergy. However, the presence of defects disrupts this regular arrangement, leading to chaotic texture directions. The purpose of this step is to quantify this physical "synergy" or "disorder" into a specific indicator.

[0033] Any pixel in the surface image is recorded as a target pixel, and the local texture coordination index of the target pixel, which is used to characterize the degree of consistency of the local texture direction, is determined according to the texture directions of each pixel in the neighborhood preset by the target pixel.

[0034] Specifically, first, the target pixel Build for the center Rectangular neighborhood window of (For example, ).

[0035] Next, for the neighborhood window Any pixel within , by calculating its Hessian matrix to analyze its local structure. The Hessian matrix is ​​a square matrix composed of second-order derivatives (the second-order derivatives of the pixel points are obtained by the finite difference method), which can reflect the local curvature change of the image grayscale. The eigenvector of the matrix indicates the direction in which the grayscale changes fastest and slowest. For linear or edge structures, the grayscale changes slowest along its direction and changes fastest perpendicular to its direction. Therefore, the eigenvector corresponding to the smaller eigenvalue of the Hessian matrix can be regarded as the pixel point The local texture main direction vector, the angle between this vector and the horizontal axis is recorded as .

[0036] Then, considering the direction and Physically, it represents the same linear texture. To eliminate this mathematical ambiguity, this embodiment adopts the method of angle multiplication and mapping to the complex plane; each direction angle Convert to a complex unit vector, that is ; Through this transformation, and The direction angles are mapped to the complex plane The position solves the problem of directional ambiguity.

[0037] Finally, the calculation window The vector average of the complex unit vectors corresponding to all pixels in the pixel is taken as the pixel point. Local texture coordination index , satisfy: ; Where, Neighborhood window The number of pixels within is an imaginary number, is the modulus length symbol.

[0038] Among them, in the intact steel wire texture area, the direction angles of all pixels Therefore, if the texture directions of all pixels in the neighborhood window are highly consistent, their corresponding complex vectors will point in the same direction. When the vector sum is calculated, they will accumulate in the same direction to obtain a very long resultant vector. After averaging, the end point of the average vector will still be very close to the boundary of the unit circle. Therefore, the modulus length of the vector average (i.e. ) will approach 1; on the contrary, if there are defects in the window that cause the texture direction to be disordered, the complex vectors will point in all directions. When these vectors are added together, they will cancel each other out. For example, a rightward vector will be canceled by a leftward vector, and an upward vector will be canceled by a downward vector, making Approaching 0.

[0039] S3: Determine the longitudinal consistency deviation index of each pixel point.

[0040] It's important to note that this step leverages another key physical prior: along the axial direction (i.e., the dominant structural direction) of a single steel wire, its surface material and light reflection properties should maintain a high degree of continuity and consistency in the absence of defects. Tiny oxidation spots or scratches can disrupt this consistency. This step effectively amplifies localized mutations caused by defects by comparing the characteristic differences between a single pixel and its adjacent points before and after it along the wire's path.

[0041] According to the grayscale distribution difference of multiple pixel blocks along the texture direction of the target pixel point and the local texture coordination index, a longitudinal consistency deviation index of the target pixel point for characterizing the degree of local grayscale mutation is determined.

[0042] Specifically, first, for the target pixel , building a The central pixel block (for example, ); At the same time, the local texture direction angle determined according to the above steps , sample forward and backward along this direction of the same size The path pixel blocks.

[0043] Next, we need to build an "expected" grayscale distribution model for the central pixel block. The normalized grayscale histogram of the path pixel blocks is weighted averaged and recorded as the expected grayscale histogram , satisfy: ; Where, is the number of the path pixel block, For the The weight of the grayscale histogram of the path pixel block, For the Grayscale histogram of the path pixel blocks.

[0044] Among them, the weight The design of makes the contribution of path pixel blocks closer to the center pixel block greater. For example, Gaussian weighting can be used: , For the The center point of the path pixel block to the target pixel point The distance obtained Represents the “normal” grayscale distribution in the absence of defects, as inferred from the context.

[0045] Then, calculate the grayscale histogram of the central pixel block and the expected grayscale histogram The difference between the two, this embodiment uses the dispersion difference matrix To measure, satisfy: ; Where, Pixel The dispersion difference moment of is the maximum gray level, is the index of gray level, is the average gray level of the expected gray histogram of the central pixel block, is the actual histogram of the central pixel block in gray level The numerical value of is the expected grayscale histogram of the central pixel block in grayscale The numerical value of is the absolute value symbol.

[0046] Among them, for , which calculates the pixel frequency of the actual observed gray level at each gray level (i.e. ) and the pixel frequency of the desired grayscale (i.e. ) The absolute size of the difference, which only quantifies the "magnitude of the difference" and does not care whether the difference is increasing or decreasing. , is a dynamically calculated weight. Its physical meaning is to assign an "importance" factor to each gray level difference. This importance is proportional to the distance that the gray level deviates from the "average gray level of the expected gray level". Distance to average grayscale The farther away, the greater the weight, and vice versa; this design gives extremely high importance to differences in extremely bright or dark areas, while differences near the average brightness are considered secondary information. For the overall formula, if the central pixel block is in a normal defect-free area, due to the longitudinal consistency of the grayscale of the steel strand, and Highly similar, Very small, calculated tends to 0; if the central pixel block is an area with a small defect and the path pixel block is a defect-free area, the histogram will be consistent with the expected grayscale histogram Producing a large difference, will be larger, and these differences will be Further amplification, output a large It should be added that if the center pixel block and the path pixel block are both areas with small defects, the calculated It will be very small, but you don’t need to worry about it at this time, because the purpose of this invention is to highlight “minor defects”. When the pixel blocks are all defective areas, they are already obvious in the image and do not need to be further enhanced. In addition, although It describes the characteristics of the central pixel block, but the central pixel block is based on the target pixel point Constructed, so Can be used as target pixel The characteristics of When it gets larger, the central pixel block is considered to be a defective area. As the center point of the central pixel block, the target pixel point The possibility of being a defect also increases, so Used to reflect the target pixel The possibility of a defect.

[0047] Finally, the dispersion difference moment Coordination index with local texture Fusion is performed to obtain the longitudinal consistency deviation index , satisfy: ; Where, Pixel The longitudinal consistency deviation index, is the hyperbolic tangent function.

[0048] in, Used to Normalized, the range is limited to , Serves as a weight for "structural credibility". For a point to be considered a significant deviation from the entire formula (i.e. The value is high, close to 1), and two conditions must be met at the same time: its grayscale value is significantly different from the surrounding area; and the local area structure is chaotic and uncoordinated. This can greatly improve the reliability of the indicator and effectively avoid misjudging normal texture fluctuations as deviations.

[0049] S4: Determine the defect significance index of each pixel.

[0050] It should be noted that the purpose of this step is to further purify the defect signal and suppress possible large-scale, slowly varying artifacts caused by factors such as uneven illumination. The core logic behind this is that a true, subtle defect is not only an anomaly in itself, but also appears abruptly, with a sharp boundary between the abnormal and normal areas.

[0051] A defect significance index of a target pixel is determined according to the longitudinal consistency deviation index and the spatial variation rate of the longitudinal consistency deviation index.

[0052] Specifically, the defect significance index satisfies: ; Where, Pixel The defect significance index, Pixel The longitudinal consistency deviation index, for The gradient vector of (can be approximately calculated by operators such as Sobel), for The mold length.

[0053] in, Represents a pixel The deviation intensity becomes the "initial possibility" of the defect, but this possibility is not perfect. When a real slight defect exists in an area, The pixel on the defect will have a higher value, and at the edge of the defect, The value of will change from high to low, resulting in its gradient modulus The product of the two will make the defective pixel is significantly amplified; for some pixels in large-scale, slowly changing deviation areas caused by uneven illumination, although There may be a certain value, but its gradient modulus is very small, and the multiplication results in The value will be suppressed. Therefore, only the abnormal ( The value of is large) and suddenly ( Only small defective pixels with larger values ​​of , further amplifying the “isolated and abrupt” defect characteristics of real micro-defects.

[0054] S5: Based on the defect significance index of each pixel point, the defect area in the surface image of the steel strand to be inspected is identified.

[0055] Specifically, the identifying of defect areas in the surface image of the steel strand to be inspected based on the defect significance index of each pixel point includes: Generate a saliency map consisting of defect saliency indicators for each pixel; Adopting adaptive threshold methods, such as the maximum inter-class variance method, to binarize and segment the saliency map, the masks of all defect areas can be accurately segmented. The segmented mask coordinates are back-calculated to the original, unexpanded image coordinate system through inverse coordinate transformation, so that all identified minor defects can be located and marked on the original image, completing the surface quality detection method of the mining anchor steel strand after stabilization treatment.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for detecting the surface quality of a mining anchor cable steel strand after stabilization treatment, characterized in that: include: Acquire a surface image of the steel strand to be inspected; Any pixel in the surface image is recorded as a target pixel, and a local texture coordination index for the target pixel, which is used to characterize the degree of consistency of the local texture direction, is determined based on the texture directions of each pixel in a preset neighborhood of the target pixel; Determine a longitudinal consistency deviation index for characterizing the degree of local grayscale mutation of the target pixel point based on the grayscale distribution difference of multiple pixel blocks along the texture direction of the target pixel point and the local texture coordination index; Determining a defect significance index of a target pixel point according to the longitudinal consistency deviation index and the spatial change rate of the longitudinal consistency deviation index; Based on the defect significance index of each pixel point, the defect area in the surface image of the steel strand to be inspected is identified.

2. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 1, wherein: The obtaining of the surface image of the steel strand to be inspected comprises: Using an image acquisition device, images of multiple viewing angles distributed around the axis of the steel strand are acquired; The images from multiple perspectives are spliced ​​and coordinate transformed to expand the cylindrical steel strand surface image into a two-dimensional plane image to obtain the surface image of the steel strand to be inspected.

3. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 1, wherein: The local texture coordination index is obtained as follows: Construct a neighborhood window with the target pixel as the center; Calculate the local texture direction angle of each pixel in the neighborhood window; Performing frequency multiplication processing on the local texture direction angle and mapping it into a complex unit vector; All complex unit vectors in the neighborhood window are summed and averaged, and the modulus of the averaged vector is used as the local texture coordination index of the target pixel.

4. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 3, wherein: The local texture direction angle is obtained as follows: For each pixel in the neighborhood window, construct the Hessian matrix of the pixel; Perform eigenvalue decomposition on the Hessian matrix of the pixel point to obtain the eigenvector corresponding to the minimum eigenvalue; The direction of the eigenvector corresponding to the minimum eigenvalue is used as the local texture direction of the pixel point, and the angle between the local texture direction and the preset coordinate axis is calculated to obtain the texture direction angle of the pixel point.

5. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 1, wherein: The longitudinal consistency deviation index is obtained as follows: A rectangular pixel block is constructed with the target pixel point as the center as the central pixel block, and multiple path pixel blocks of the same size as the central pixel block are constructed on both sides of the central pixel block along the local texture direction of the target pixel point; According to the grayscale histograms of the plurality of path pixel blocks, an expected grayscale histogram is calculated; Based on the difference between the grayscale histogram of the central pixel block and the expected grayscale histogram, the dispersion difference moment of the target pixel is calculated. Combined with the dispersion difference moment of the target pixel and the local texture coordination index, the longitudinal consistency deviation index of the target pixel is determined.

6. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 5, characterized in that: The calculation to obtain the expected grayscale histogram includes: The actual grayscale histograms of the multiple path pixel blocks are subjected to Gaussian weighted averaging based on the distance between each path pixel block and the central pixel block to obtain the expected grayscale histogram.

7. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 5, characterized in that: The longitudinal consistency deviation index satisfies: ； Where, Pixel The longitudinal consistency deviation index, Pixel The dispersion difference moment of Pixel The local texture coordination index, is the hyperbolic tangent function.

8. A method for detecting surface quality of a stabilized mining anchor steel strand according to claim 5 or 7, characterized in that: The dispersion difference moment satisfies: ； Where, Pixel The dispersion difference moment of is the maximum gray level, is the index of gray level, is the average gray level of the expected gray histogram of the central pixel block, is the actual histogram of the central pixel block in gray level The numerical value of is the expected grayscale histogram of the central pixel block in grayscale The numerical value of is the absolute value symbol.

9. The method for detecting surface quality of a stabilized mining anchor steel strand according to claim 1, wherein: The defect significance index satisfies: ； Where, Pixel The defect significance index, Pixel The longitudinal consistency deviation index, for The gradient vector of for The mold length.

10. The method for detecting the surface quality of a stabilized mining anchor steel strand according to claim 1, wherein: The method of identifying defect areas in the surface image of the steel strand to be inspected based on the defect significance index of each pixel point includes: Generate a saliency map consisting of defect saliency indicators for each pixel; The maximum inter-class variance method is used to perform adaptive threshold segmentation on the saliency map to obtain a binary image; In the binary image, the area with pixel values ​​identified as defects is determined as the defect area in the surface image of the steel strand to be inspected, thereby completing the surface quality inspection method of the mining anchor steel strand after stabilization treatment.

Images