Method and system for detecting installation quality of middle part of belt conveyor

Abstract

The application belongs to the technical field of image data processing, and particularly relates to a method and system for detecting installation quality of a middle part of a belt conveyor, which comprises the following steps: obtaining initial direction vectors of each pixel point of an image of the middle part of the belt conveyor through frequency domain analysis; for each pixel point, obtaining an anisotropic kernel function based on direction similarity and iteratively updating the initial direction vector to obtain a final direction vector; obtaining an axis positioning score based on the divergence of the normal vector to form an axis positioning graph; extracting a main central axis skeleton point set of the axis positioning graph and fitting a straight line, comparing an error of the skeleton point to the fitted straight line with a preset tolerance, and determining whether the installation quality is qualified or not. The application solves the problems of linear feature breakage and unreliable direction field caused by the middle gap in an industrial image, and realizes robust and high-precision automatic detection of the straightness of the belt conveyor.

Description

Technical Field

[0001] This invention relates to the field of image data processing technology. More specifically, this invention relates to a method and system for detecting the installation quality of the middle section of a belt conveyor. Background Technology

[0002] Belt conveyors are indispensable continuous material transport equipment in industrial fields such as mining and ports. Their installation quality, especially the straightness of the intermediate longitudinal beams, directly affects the smoothness of the conveyor's operation, service life, and production safety. Traditional manual inspection methods are inefficient, subjective, and pose safety risks. Using machine vision technology for automated and high-precision installation quality inspection has become a trend in the industry.

[0003] In actual testing scenarios, industrial environments are complex, with strong noise interference such as dust, uneven lighting, and reflections. The longitudinal beams of belt conveyors are usually spliced ​​together from multiple segments, and there are physical gaps between the segments. These factors cause the linear features of the longitudinal beams in the acquired images to be broken, blurred, or submerged by noise, making it difficult to extract the main central axis of the longitudinal beams.

[0004] Existing technologies often employ local orientation estimation methods based on frequency domain analysis to obtain the orientation of linear structures in images. This method analyzes the spectral characteristics of each local region of the image by performing Fourier transform to determine the dominant texture orientation. However, such methods rely only on local information. When dealing with the physical connection gaps of longitudinal beams or background areas without obvious texture, the calculated orientation is unreliable or even random due to the lack of effective structural information within the local neighborhood window. This results in the obtained orientation field being broken at key gap locations, failing to form a global orientation field that runs through the entire longitudinal beam. Consequently, it is difficult to accurately fit the complete central axis and evaluate its straightness, severely limiting the robustness and accuracy of the detection. Summary of the Invention

[0005] To address the technical problem of existing technologies that rely solely on local information and cannot accurately fit the complete central axis, the present invention provides solutions in the following aspects.

[0006] In a first aspect, the present invention provides a method for inspecting the installation quality of the middle section of a belt conveyor, comprising: acquiring a grayscale image of the middle section of the belt conveyor; designating any pixel of the grayscale image as a target pixel; applying a window function to the neighborhood window of the target pixel and obtaining an initial direction vector of the target pixel based on Fourier transform; obtaining an anisotropic kernel function between the target pixel and its neighborhood window pixels based on the distance between the target pixel and its neighborhood window pixels and the similarity of the initial direction vectors of the target pixel and its neighborhood window pixels, iteratively updating the initial direction vector of the target pixel using the anisotropic kernel function to obtain the final direction vector of the target pixel; obtaining the normal vector of the target pixel based on the final direction vector of the target pixel; obtaining the axis positioning score of the target pixel based on the divergence of the normal vector of the target pixel, and assembling the axis positioning scores of all pixels into an axis positioning map; extracting the main central axis skeleton point set of the axis positioning map and fitting it as a straight line using the least squares method, calculating the error from the skeleton point to the fitted straight line; and comparing the error with a preset tolerance to determine whether the installation quality of the middle section of the belt conveyor is qualified.

[0007] This invention employs a technical approach that obtains the initial direction from frequency domain analysis and then iteratively optimizes it through anisotropic smoothing. This approach can intelligently propagate and fill the gaps in the characteristic interruption area of ​​the conveyor body with accurate directional information, thereby extracting the continuous main centerline covering the entire length of the longitudinal beam. This provides reliable technical support for achieving automatic and high-precision installation quality inspection.

[0008] Preferably, applying a window function to the neighborhood window of the target pixel includes applying a Hanning window function to the neighborhood window of the target pixel.

[0009] This invention improves the accuracy of initial direction vector calculation by applying the Hanning window function before Fourier transform, effectively suppresses spectral leakage artifacts caused by image neighborhood window truncation, and makes the frequency domain energy more accurately correspond to the real structural direction, providing high-quality input for subsequent iterative optimization, thereby enhancing the robustness of the entire detection method.

[0010] Preferably, the anisotropic kernel function between the target pixel and its neighboring window pixels satisfies the expression: ;in, Represents the target pixel. Its neighboring window pixels Anisotropic kernel function values ​​between; Indicates the target pixel; Represents the pixels of the neighboring window; It is the number of iterations; Represents pixels In the The direction vector at the next iteration; Represents pixels In the The direction vector at the next iteration; Indicates the first hyperparameter; Indicates the second hyperparameter; Represents the dot product operation of vectors; Represents the natural exponential function; Represents the absolute value symbol.

[0011] The anisotropic kernel function of this invention can effectively bridge the structural discontinuities of the conveyor longitudinal beam in the image, forming a continuous and smooth final orientation field. When iteratively updating the orientation, the anisotropic kernel function will adaptively assign higher weights to the neighboring window pixels that are consistent with the orientation of the center point, so that the correct orientation information can flow from the structural entity to the gap region, while suppressing the interference of background noise. Finally, an orientation field that remains continuous and accurate even at the physical gap is obtained, laying the foundation for subsequent main centerline positioning.

[0012] Preferably, the first hyperparameter is set to 0.8.

[0013] Preferably, the second hyperparameter is set to 2.0.

[0014] Preferably, obtaining the normal vector of the target pixel based on the final direction vector of the target pixel includes: rotating the final direction vector of the target pixel clockwise by 90° to obtain the normal vector of the target pixel.

[0015] Preferably, the axis positioning score of the target pixel satisfies the expression: ;in, Represents the target pixel. Represents the target pixel. The axis positioning score; and express Within a central oversized neighborhood window, two sub-neighborhood windows are defined by the line representing the final direction vector of the target pixel. These oversized neighborhood windows are defined by the expected width of the longitudinal beam. A neighborhood window whose side length is twice that of the neighborhood window; Represents a sub-neighborhood window Pixels within; Represents a sub-neighborhood window Pixels within; and Representing pixels normal vector and pixel The normal vector; This represents the divergence.

[0016] This invention utilizes the divergence characteristics of the normal vector field to clearly highlight the main central axis of the longitudinal beam on the axis positioning map with a high-brightness response band. The obtained axis positioning score expression can capture the center of the divergence extreme value bands that are positive and negative symmetrical on both sides of the longitudinal beam. The score is maximized only when the pixel is located on the true main central axis, so that the position of the main central axis is clearly marked, reducing the difficulty and uncertainty of subsequent extraction.

[0017] Preferably, extracting the main central axis skeleton point set of the axis positioning map includes: binarizing the axis positioning map to obtain a binary image; applying a morphological skeleton extraction algorithm to the binary image to obtain a main central axis skeleton with a single pixel width; performing connected component analysis on the thinned main central axis skeleton, and retaining the longest main central axis skeleton line segment as the skeleton point set of the main central axis.

[0018] This invention applies a morphological skeleton extraction algorithm to obtain the main central axis of the axis positioning map, effectively avoiding inaccurate gradient directions caused by local grayscale fluctuations or noise interference on the axis positioning map, thereby preventing the breakage or deviation of the main central axis refinement result. By retaining the longest skeleton line segment through connected component analysis, the main central axis can be accurately separated, and isolated short skeletons formed by noise points can be removed, ensuring the accuracy of the final skeleton point set, thus providing a more reliable data foundation for subsequent straightness error calculation.

[0019] Preferably, the step of calculating the error from the skeleton point to the fitted line and comparing the error with a preset tolerance to determine whether the installation quality of the middle part of the belt conveyor is qualified includes: the error from the skeleton point to the fitted line is the root mean square of the vertical distance from all skeleton points to the fitted line; when the root mean square is less than or equal to the preset tolerance, the installation quality is qualified; when it is greater than the preset tolerance, the installation quality is unqualified.

[0020] Secondly, the present invention provides a belt conveyor intermediate section installation quality inspection system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned belt conveyor intermediate section installation quality inspection method is implemented.

[0021] By adopting the above technical solution, a computer program for the above-mentioned method for quality inspection of the middle section of a belt conveyor is generated and stored in a memory so that it can be loaded and executed by a processor. This allows for the creation of terminal equipment based on the memory and processor, making it convenient to use.

[0022] The beneficial effects of this invention are as follows:

[0023] This invention utilizes reliable directional information from the physical components of the conveyor to intelligently propagate and fill the gaps with the correct structural orientation, thereby forming a continuous and smooth directional field covering the entire length of the longitudinal beam. This eliminates the constraint of the physical continuity of the structure on straightness detection, improving the robustness and reliability of the evaluation results under complex working conditions.

[0024] This invention constructs a positioning score that maximizes the opposite divergence response on both sides of the conveyor edge, which can highlight the position of the main central axis on the axis positioning map in a high-brightness, high-contrast form. Then, the high-brightness response band is refined into a precise line with a single pixel width through a non-maximum suppression algorithm, providing an accurate geometric basis for subsequent quantitative calculations, thereby improving the accuracy of the entire detection method.

[0025] This invention compares the extracted main central axis skeleton point set with the ideal straight line fitted by the least squares method, and calculates the root mean square error value as the error of the straightness of the longitudinal beam. This provides a unified and repeatable judgment standard for the installation quality acceptance of belt conveyors, and meets the needs of automated inspection in industrial sites. Attached Figure Description

[0026] Figure 1 This is a flowchart illustrating a method for inspecting the installation quality of the middle section of a belt conveyor according to the present invention. Detailed Implementation

[0027] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, 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 creative effort are within the scope of protection of the present invention.

[0028] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0029] This invention discloses a method for inspecting the installation quality of the middle section of a belt conveyor, referring to... Figure 1 This includes steps S1-S4:

[0030] S1. Obtain a grayscale image of the middle part of the belt conveyor; denote any pixel of the grayscale image as the target pixel; apply a window function to the neighborhood window of the target pixel and obtain the initial direction vector of the target pixel based on Fourier transform.

[0031] It should be noted that this invention calculates an initial structural orientation for each pixel in the image, reflecting its local dominant texture, providing a foundation for subsequent orientation field smoothing. Due to the presence of strong noise interference such as dust and glare in industrial environments like underground coal mines, this invention employs a frequency domain analysis-based method to improve the robustness of the orientation calculation.

[0032] Specifically, grayscale images of the middle region of the belt conveyor are acquired using image acquisition equipment such as industrial cameras. ; Record pixels For the target pixel, extract an area centered on the target pixel with a size of [value missing]. neighborhood window The size of the neighborhood window should be large enough to encompass local texture features without obscuring the orientation of different structures. A window function is applied to preprocess the pixel grayscale values ​​within the neighborhood window to reduce potential spectral leakage during subsequent Fourier transforms. A two-dimensional discrete Fourier transform (2D-DFT) is performed on the windowed neighborhood window data to obtain its corresponding spectrum. Linear structures such as conveyor beams exhibit regular textures in the spatial domain, which correspond to specific energy concentration areas in the frequency domain. In the spectrum, the coordinates of the point with the strongest energy (excluding the center point) are identified and located. The direction vector of this coordinate is perpendicular to the dominant texture direction within the neighborhood window in the spatial domain. A unit vector orthogonal to this direction is calculated and defined as the target pixel. Initial direction vector at The initial direction vectors of all pixels form the initial direction field. The initial direction vectors of pixels located in the longitudinal beam region are relatively accurate and consistent, while the initial direction vectors of pixels located in the gaps between segments and the background region may appear as noise or have no clear direction.

[0033] It should be noted that the neighborhood window size in this embodiment... Select 15 pixels; select Hanning window as the window function. In other embodiments, implementers can select the neighborhood window size and window function according to actual needs.

[0034] S2. Based on the distance between the target pixel and its neighboring window pixels and the similarity of the initial direction vectors of the target pixel and its neighboring window pixels, obtain the anisotropic kernel function between the target pixel and its neighboring window pixels. Use the anisotropic kernel function to iteratively update the initial direction vector of the target pixel to obtain the final direction vector of the target pixel.

[0035] It should be noted that the initial orientation field is reliable in the longitudinal beam, but it lacks orientation information for the critical gap region. This invention utilizes known reliable orientation information and propagates the correct structural orientation to the gap region through an orientation-adaptive smoothing process, while suppressing background noise interference, to generate a final orientation field that is continuous and smooth throughout the entire longitudinal beam path, including all gaps.

[0036] Specifically, in the first In the next iteration, the target pixel The direction vector satisfies the expression:

[0037]

[0038] The anisotropic kernel function between the target pixel and its neighboring window pixels satisfies the following expression:

[0039]

[0040] in, Represents the target pixel. Its neighboring window pixels Anisotropic kernel function values ​​between; Indicates the target pixel; Indicates A neighborhood window centered on the center; Represents the pixels of the neighboring window; It is the number of iterations; Represents the target pixel. In the The direction vector at the next iteration; Represents the neighborhood window pixels In the The direction vector at the next iteration; This represents the first hyperparameter used to control the weight decay of spatial distance; The second hyperparameter represents the effect used to regulate directional consistency. Represents the dot product operation of vectors; Represents the natural exponential function; Indicates the absolute value symbol; This indicates a normalization operation, which converts the composite vector within the parentheses into a unit vector.

[0041] More specifically, spatial weight term It is a standard Gaussian function that makes the spatial distance from the target pixel point... The closer the neighboring window pixels Obtain higher weights; directional consistency weighting term This measures the similarity between the direction vectors of the target pixel and the pixels in its neighboring window. When the two direction vectors are completely identical, the absolute value of the dot product is... The weight is the largest; when the two directions are perpendicular, the dot product is... The weight is .

[0042] When performing a smoothing calculation on a target pixel located in the gap region, pixels from the two longitudinal beams within its neighborhood window will receive higher weights because their structural orientation aligns with the main direction of the beams, and their correct orientation information will effectively fill the gap. Pixels from the background region with random orientations, however, will have extremely low weights because their orientations do not align with the beam orientations, and will hardly participate in the weighted average.

[0043] After several iterations, the initial direction vector of the target pixel becomes the final direction vector. The final direction vectors of all pixels in the image form a continuous and smooth final direction field throughout the longitudinal beam and its connecting gaps. .

[0044] It should be further explained that the first hyperparameter is used to control the weight decay of spatial distance in the anisotropic kernel function. When the first hyperparameter increases, pixels that are spatially farther away will receive higher weights; conversely, they will be mainly affected by nearest neighbor points. In this embodiment, the first hyperparameter is set to 0.8. In other embodiments, implementers can set the first hyperparameter according to the actual implementation situation. If the physical gap of the longitudinal beam is very wide, the first hyperparameter can be appropriately increased, for example, to 1.0, so that pixels in the gap area can refer to reliable directional information of the longitudinal beam further away. If the image noise is mainly manifested as local noise, or the longitudinal beam structure is tight, it is more appropriate to keep a smaller first hyperparameter, for example, 0.7, to focus on nearest neighbor smoothing and prevent irrelevant noise points in the distance from causing interference.

[0045] The second hyperparameter is used to regulate the influence of directional consistency in the anisotropic kernel function. When the second hyperparameter is greater than 1, the weights of neighboring window points with slight directional deviations will decrease sharply, thereby strengthening the requirement for directional consistency. In this embodiment, a value of 2.0 is chosen, indicating a very strict requirement for directional consistency. This is suitable for situations where the longitudinal beam structure itself is relatively clear and the initial orientation field quality is good, effectively suppressing background noise interference. Implementers can set the second hyperparameter according to the actual implementation situation. If the image noise interference is extremely strong, causing a large deviation in the initial orientation field itself, an excessively high second hyperparameter may over-suppress, leading to difficulties in information propagation. In this case, the second hyperparameter can be appropriately lowered, for example, set between 1.0 and 1.5, to increase the fault tolerance of iterative updates.

[0046] It should be further explained that the purpose of iteration is to allow reliable directional information from the longitudinal beams to intelligently propagate and fill the gap regions where features are interrupted. The choice of the number of iterations is a balance: too few iterations may result in directional information failing to fully traverse the physical gaps of the longitudinal beams, making the directional field in the gap regions unreliable; too many iterations will significantly increase computation time and may lead to over-smoothing. In this embodiment, the number of iterations is set to 5. Implementers can adjust the number of iterations according to the actual implementation situation. For example, when the gaps between the longitudinal beams to be processed are large or the noise is strong, the number of iterations can be appropriately increased, such as 5 to 10, to ensure sufficient convergence of the directional field; if the longitudinal beam continuity is good and the gaps are small, the number of iterations can be reduced, such as 3, to improve efficiency.

[0047] S3. Obtain the normal vector of the target pixel based on the final direction vector of the target pixel; obtain the axis positioning score of the target pixel based on the divergence of the normal vector of the target pixel, and assemble the axis positioning scores of all pixels into an axis positioning map.

[0048] It should be noted that, in obtaining a continuous final direction field Subsequently, in order to find the main centerline of the longitudinal beam and complete the longitudinal beam straightness detection, this invention generates a scalar map so that the pixels located on the main centerline of the longitudinal beam have the highest response value.

[0049] Within the longitudinal beam region, due to its straight structure, the normal vectors at all points point in the same direction and have the same magnitude. Therefore, the normal vector field in this region is constant, and its divergence is zero. However, at the two edges of the longitudinal beam, i.e., the region transitioning from the longitudinal beam to the background, the normal vectors undergo a sharp and opposite change in direction. This creates a region with positive divergence on one edge and a region with negative divergence on the other edge. The main central axis of the longitudinal beam is located precisely in the middle of these two extreme divergence zones. By utilizing this characteristic, the present invention can accurately locate the main central axis of the longitudinal beam.

[0050] Specifically, for target pixels Obtain the axis positioning score. The axis positioning score of the target pixel satisfies the expression:

[0051]

[0052] in, Represents the target pixel. Represents the target pixel. The axis positioning score; and express Within the central supersized neighborhood window, there are two sub-neighborhood windows separated by the line to which the final direction vector of the target pixel belongs; Represents a sub-neighborhood window Pixels within; Represents a sub-neighborhood window Pixels within; and Representing pixels and The normal vector of a pixel is obtained by rotating the final direction vector of the pixel clockwise by 90°. This represents the divergence.

[0053] More specifically, using image processing software, open one or more representative images of the longitudinal beam, ensuring the beam area is clearly visible in the image. Use the software's measurement tool to measure the vertical pixel distance between the two edges of the beam. Perform multiple measurements at different locations in the image, such as the upper, middle, and lower parts of the beam. Then calculate the average of these measurements as the expected width of the beam. In this embodiment, the image processing software uses an interactive window from OpenCV, and the measurement tool uses a ruler tool. The implementer can choose the appropriate image processing software and measurement tool based on the actual situation.

[0054] The diameter of the oversized neighborhood window should be larger than the maximum expected width of the longitudinal beam being measured in the image. In this embodiment, the expected width is used. The core requirement for setting the diameter of the oversized neighborhood window is that it must be larger than the maximum expected width of the measured longitudinal beam in the image. This is because the calculation of the axis positioning score relies on capturing the divergence extrema bands that are symmetrically positive and negative at the edges of the longitudinal beam. If the diameter of the oversized neighborhood window is smaller than the width of the longitudinal beam, the two half-neighborhood windows will not be able to simultaneously and completely contain the divergence responses on both sides when calculating the score, leading to the failure of the main central axis positioning. Choosing 1.5 times instead of exactly 1.0 times is to provide a safety margin, ensuring that the method still works reliably even if the actual width of the longitudinal beam in the image is slightly larger than the expected width. If the width of the longitudinal beam varies greatly or the expected width is inaccurate, a larger multiple, such as 2.0 times, may be needed to ensure robustness.

[0055] More specifically, when the target pixel is located exactly on the main central axis of the longitudinal beam, its oversized neighborhood window is divided into two halves: a half-neighborhood window. The entire divergence response band generated by one side edge will be negative, and the sum of its divergences will be a negative number with a large absolute value; while the other half of the neighborhood window The positive divergence response band completely encompassing the other edge is summed to a large positive number. Multiplying these two results in a large negative number, which is then inverted to make... It becomes a very large positive value. When the center pixel... When the line deviates from the main central axis, the dividing line shifts accordingly. For example, if... Towards If it moves to one side, then The mixture will contain some regions of positive divergence, causing the absolute value of its sum to decrease, while... This may result in the loss of some positive divergence regions, reducing the sum of these regions and causing a significant decrease in the absolute value of the product. The value also decreases accordingly. Therefore, the target pixel will only generate the maximum axis localization score at the position that maximizes the product of the sum of the divergences on both sides of the neighborhood window, i.e., on the main central axis.

[0056] The axis localization scores of all pixels in the image constitute the axis localization map. The axis positioning diagram appears as a bright, continuous response band, with its ridgeline precisely corresponding to the main centerline of the longitudinal beam.

[0057] S4. Extract the main centerline skeleton point set of the axis positioning diagram and fit it into a straight line using the least squares method. Calculate the error from the skeleton point to the fitted straight line. Compare the error with the preset tolerance to determine whether the installation quality of the middle part of the belt conveyor is qualified.

[0058] It should be noted that the present invention uses an axis positioning diagram. The geometric information contained therein is transformed into quantitative straightness assessment results.

[0059] Specifically, the axis positioning image is binarized, converting the high-brightness, wide response band into a binary image. A morphological skeleton extraction algorithm is applied to this binary image, iteratively stripping pixels from the region edges until the wide binary image is thinned into a single-pixel-width main central axis skeleton. Connected component analysis is performed on the thinned skeleton image, retaining the longest skeleton line segment as the skeleton point set of the main central axis. The extracted main central axis skeleton point set is then fitted with a line using the least squares method to obtain a mathematical straight line that best represents the main central axis of the longitudinal beam. The perpendicular distance from each skeleton point to the fitted line is calculated, and the root mean square (RMS) value of all these distances is used as the error in the straightness of the longitudinal beam. In this embodiment, the morphological skeleton extraction algorithm uses the Zhang-Suen thinning algorithm, with a length threshold of 150 pixels. Implementers can choose the morphological skeleton extraction algorithm and length threshold according to actual needs.

[0060] Furthermore, the calculated straightness error is compared with the preset tolerance standard. If the error is less than or equal to the tolerance, the installation quality of the middle part of the belt conveyor is deemed qualified; if the error is greater than the tolerance, the installation quality of the middle part of the belt conveyor is deemed unqualified.

[0061] This completes a method for inspecting the installation quality of the middle section of a belt conveyor.

[0062] This invention also discloses a belt conveyor intermediate section installation quality inspection system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, a belt conveyor intermediate section installation quality inspection method according to the present invention is implemented.

[0063] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A method for inspecting the installation quality of the middle section of a belt conveyor, characterized in that, include: Acquire a grayscale image of the middle section of the belt conveyor; denote any pixel in the grayscale image as the target pixel. After applying a window function to the neighborhood window of the target pixel, the initial direction vector of the target pixel is obtained based on Fourier transform; Based on the distance between the target pixel and its neighboring window pixels and the similarity of the initial direction vectors of the target pixel and its neighboring window pixels, the anisotropic kernel function between the target pixel and its neighboring window pixels is obtained, including: ;in, For target pixel Its neighboring window pixels Anisotropic kernel function values ​​between; For the target pixel; For the neighboring window pixels; It is the number of iterations; For pixels In the The direction vector at the next iteration; For pixels In the The direction vector at the next iteration; Indicates the first hyperparameter; Indicates the second hyperparameter; Represents the dot product operation of vectors; Represents the natural exponential function; The absolute value sign is represented by the anisotropic kernel function, which iteratively updates the initial direction vector of the target pixel to obtain the final direction vector of the target pixel. The normal vector of the target pixel is obtained from its final direction vector; the axis localization score of the target pixel is obtained from the divergence of its normal vector. ; in, For target pixel The axis positioning score; , for Within the centered super-sized neighborhood window, two sub-neighborhood windows are separated by the line representing the final direction vector of the target pixel. The super-sized neighborhood window is defined by the expected width of the longitudinal beam. A neighborhood window whose side length is twice that of the neighborhood window; For child neighborhood windows Pixels within; For child neighborhood windows Pixels within; , Each pixel normal vector, pixel The normal vector; To calculate the divergence, the axis positioning scores of all pixels are used to construct an axis positioning map. Extract the main centerline skeleton point set from the axis positioning diagram and fit it to a straight line using the least squares method. Calculate the error between the skeleton points and the fitted straight line. Compare the error with the preset tolerance to determine whether the installation quality of the middle part of the belt conveyor is qualified: the error between the skeleton points and the fitted straight line is the root mean square of the vertical distance from all skeleton points to the fitted straight line. If the root mean square is less than or equal to the preset tolerance, the installation quality is qualified; if it is greater than the preset tolerance, the installation quality is unqualified.

2. The method for inspecting the installation quality of the middle section of a belt conveyor according to claim 1, characterized in that, The application of a window function to the neighborhood window of the target pixel includes: Apply the Hanning window function to the neighborhood window of the target pixel.

3. The method for inspecting the installation quality of the middle section of a belt conveyor according to claim 1, characterized in that, The first hyperparameter is set to 0.

8.

4. The method for inspecting the installation quality of the middle section of a belt conveyor according to claim 1, characterized in that, The second hyperparameter is set to 2.

0.

5. A method for inspecting the installation quality of the middle section of a belt conveyor according to claim 1, characterized in that, The step of obtaining the normal vector of the target pixel based on the final direction vector of the target pixel includes: Rotate the final direction vector of the target pixel 90° clockwise to obtain the normal vector of the target pixel.

6. A method for inspecting the installation quality of the middle section of a belt conveyor according to claim 1, characterized in that, Extracting the main centerline skeleton point set of the axis positioning map, including: The axis positioning map is binarized to obtain a binary image; a morphological skeleton extraction algorithm is applied to the binary image to obtain a main central axis skeleton with a single pixel width; connected component analysis is performed on the thinned main central axis skeleton, and the longest main central axis skeleton line segment is retained as the skeleton point set of the main central axis.

7. A quality inspection system for the middle section of a belt conveyor, characterized in that, include: A processor and a memory, the memory storing computer program instructions that, when executed by the processor, implement a method for detecting the installation quality of the intermediate section of a belt conveyor according to any one of claims 1-6.

Images