Corrugated paper defect detection method and system based on machine vision

Abstract

The invention relates to the technical field of defect detection, in particular to a corrugated paper defect detection method and system based on machine vision. The method comprises the following steps: acquiring a corrugated paper image and converting the corrugated paper image into a frequency domain; identifying a periodic texture peak point based on amplitude spectrum angle energy distribution, generating a texture suppression filter, and combining with a multi-scale Gaussian high-pass filter to obtain a composite filter bank; the filter bank is used for filtering the frequency domain signals and performing energy weighted fusion; performing inverse transformation on the fused signal to obtain an enhanced image; and segmenting the enhanced image to obtain candidate defects, and verifying by combining the gradient information of the original image to determine a final defect. According to the scheme, the texture background can be adaptively inhibited, the multi-scale defect is enhanced, the false alarm rate is effectively reduced in combination with gradient check, and the detection accuracy and robustness are improved.

Description

Technical Field

[0001] This invention relates to the field of defect detection technology. More specifically, this invention relates to a machine vision-based method and system for detecting defects in corrugated paper. Background Technology

[0002] During the production and processing of corrugated paper, various defects such as dirt, holes, wrinkles, and scratches often occur on the surface due to factors such as raw materials, equipment, or processes. When using machine vision for defect detection, the periodic and directional strong texture background inherent in the corrugated paper surface manifests as high-frequency noise in the acquired images. Because the characteristics of this high-frequency noise are similar to or overlap with the characteristics of many minute defects in the frequency domain, defect segmentation and identification become extremely difficult. Therefore, existing conventional image processing methods, such as thresholding or edge detection techniques, are prone to misjudging textures as defects due to strong interference from the textures when processing images with such strong texture backgrounds, or are unable to effectively extract the true defect information.

[0003] To address the aforementioned issues, existing technologies employ homomorphic filtering, a classic frequency domain processing technique. This technique can effectively suppress the impact of uneven illumination on image quality while enhancing local contrast, thereby highlighting image details, by separating the illuminance and reflectance components of an image. Specifically, in the application of corrugated paper defect detection, traditional homomorphic filtering methods typically employ a standard Gaussian high-pass filter. Its principle is to filter out low-frequency components representing the background and illumination, while enhancing high-frequency components, including defects and textures, to highlight the defects.

[0004] However, the aforementioned method using traditional homomorphic filtering still has significant shortcomings. First, this method cannot effectively distinguish between defect information and periodic texture information. Because the Gaussian high-pass filter indiscriminately enhances all high-frequency signals, it enhances textures equally while enhancing defect features, resulting in background textures still severely interfering with defect identification. Second, the key parameters of the filters used in traditional homomorphic filtering usually need to be manually set based on the operator's experience. When the texture period, direction, or lighting conditions of the corrugated paper on the production line change, the fixed parameters are difficult to adapt to, failing to achieve the ideal filtering effect and potentially causing technical problems such as loss of defect information or insufficient texture suppression. Summary of the Invention

[0005] The purpose of this invention is to propose a method and system for detecting defects in corrugated paper based on machine vision, so as to solve the problem of inaccurate detection of defects in corrugated paper in the prior art; to this end, this invention provides solutions in the following two aspects.

[0006] In a first aspect, the present invention provides a machine vision-based method for detecting defects in corrugated paper, comprising the following steps: A corrugated paper image to be detected is acquired, and logarithmic and Fourier transforms are performed on the image to obtain a frequency domain signal. The amplitude spectrum of the frequency domain signal is calculated, and multiple main direction frequency peaks representing periodic textures are identified based on the angular energy distribution of the amplitude spectrum. The average distance from each peak to the origin of the frequency domain is calculated, and a texture suppression filter with a valley value at the corresponding position is generated based on the position of the peak. A set of Gaussian high-pass filters with different cutoff frequencies is generated based on the average distance, and the texture suppression filter is multiplied by each Gaussian high-pass filter to obtain multiple composite filters. The multiple composite filters are used to process the frequency domain signal. The signal is filtered to obtain multiple filtered frequency domain signals. A normalized fusion weight is determined based on the total energy of each filtered frequency domain signal, and the multiple filtered frequency domain signals are weighted and fused to obtain the final frequency domain signal. An inverse Fourier transform and exponential transform are performed on the final frequency domain signal to obtain an enhanced image. The gradient magnitude map of the corrugated paper image to be detected is calculated, and threshold segmentation is performed on the enhanced image to obtain candidate defect regions. For any candidate defect region, if the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold, then the candidate defect region is determined to be the final defect region.

[0007] Preferably, the step of identifying multiple main direction frequency peak points representing periodic textures based on the amplitude spectrum angular energy distribution includes: dividing the amplitude spectrum into 180 angular sectors centered on the frequency domain origin, with each sector having an angular range of 1°; calculating the sum of the energy of all frequency points within each sector to obtain the total energy of each sector; calculating the average value of the total energy of all sectors, and determining the sectors with a total energy greater than 1.2 times the average value as main direction sectors; and within each main direction sector, searching for the frequency point with the largest amplitude as the main direction frequency peak point.

[0008] Preferably, generating a texture suppression filter with a valley value at the corresponding position based on the position of the peak point includes: multiplying multiple Gaussian band-stop filters to obtain the texture suppression filter, for the first... Peak points The Gaussian band-stop filter corresponding to the peak point The transfer function is: ,in, For point to peak point distance, is the standard deviation of the Gaussian function.

[0009] Preferably, the cutoff frequencies of the set of Gaussian high-pass filters with different cutoff frequencies are set according to different scaling factors based on the average distance.

[0010] Preferably, three Gaussian high-pass filters are configured, with their cutoff frequencies set to 0.2 times, 0.5 times, and 1.0 times the average distance, respectively.

[0011] Preferably, determining the normalized fusion weights based on the total energy of each filtered frequency domain signal includes: for the first... The filtered frequency domain signal Calculate the total energy of the frequency domain signal. : ; Calculate the first Normalized fusion weights of individual signals : ,in This is the index for all filtered frequency domain signals.

[0012] Preferably, the step of thresholding the enhanced image to obtain candidate defect regions includes: using the Niblack thresholding algorithm, setting a window of size 15×15 pixels, and determining the candidate defect region based on the average value of the pixels within the window. and standard deviation Calculate the segmentation threshold The formula for calculating the segmentation threshold is: ,in, The segmentation threshold is... This represents the average value of the pixels within the window. is the standard deviation of the pixels within the window.

[0013] Preferably, the step of calculating the gradient magnitude map of the corrugated paper image to be detected includes: using the Sobel operator to calculate the gradient in the horizontal direction and the gradient in the vertical direction of the corrugated paper image to be detected; calculating the gradient magnitude of each pixel based on the gradient in the horizontal direction and the gradient in the vertical direction, and the gradient magnitudes of all pixels together constitute the gradient magnitude map.

[0014] Preferably, determining the candidate defect region as the final defect region when the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold includes: calculating the average gradient value of all pixels in the entire gradient magnitude map. Set the preset gradient threshold to 1.5 times.

[0015] In the second aspect, a machine vision-based corrugated paper defect detection system includes: The processor; the memory storing computer instructions for machine vision-based corrugated paper defect detection, which, when executed by the processor, cause the system to perform the machine vision-based corrugated paper defect detection method described above.

[0016] The beneficial effects of this invention are as follows: By analyzing the angular energy distribution of the amplitude spectrum, it can automatically identify and accurately suppress the unique periodic textures in corrugated paper images, effectively solving the problem of traditional fixed filters failing due to texture changes, and exhibiting strong adaptability; at the same time, this invention employs a multi-scale composite filter bank combined with an energy-weighted fusion strategy, which can integrate and enhance real defect information at different scales while suppressing strong texture backgrounds; moreover, by using the gradient information of the original image to perform secondary verification on the candidate defects segmented from the enhanced image, it can effectively eliminate false defects caused by residual textures or filtering artifacts, thereby significantly reducing the false alarm rate and greatly improving the accuracy and robustness of defect detection. Attached Figure Description

[0017] Figure 1 The flowchart illustrating the steps of the machine vision-based corrugated paper defect detection method in this embodiment is shown in the schematic diagram. Figure 2 The schematic diagram illustrates the structural block diagram of the machine vision-based corrugated paper defect detection system in this embodiment. Detailed Implementation

[0018] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0019] like Figure 1 As shown in this embodiment, a machine vision-based method for detecting defects in corrugated paper includes the following steps: Step S1: Acquire the corrugated paper image to be detected, perform logarithmic transformation and Fourier transform on the image to obtain a frequency domain signal; calculate the amplitude spectrum of the frequency domain signal, identify multiple main direction frequency peak points representing periodic texture based on the angular energy distribution of the amplitude spectrum, calculate the average distance from the peak point to the frequency domain origin, and generate a texture suppression filter with a valley value at the corresponding position according to the position of the peak point.

[0020] Specifically, images of the corrugated paper surface are captured using industrial cameras on the production line as the images to be inspected. Each pixel value in the image is incremented by 1 and its natural logarithm is taken to perform a logarithmic transformation, making it easier to separate the low-frequency components representing background lighting from the high-frequency components representing object details (texture and defects) in the frequency domain. A two-dimensional discrete Fourier transform is then performed on the logarithmically transformed image, shifting the zero-frequency component of the image to the center of the spectrum to obtain a complex frequency domain signal. This makes the corrugated paper texture appear as several highly concentrated bright spots in the frequency domain.

[0021] Calculate the amplitude spectrum of the frequency domain signal. Convert the Cartesian coordinates of the spectrum to polar coordinates, and divide the angular range of 0 to 180 degrees into several small angular intervals. Accumulate the amplitude values ​​of all frequency points within each angular interval to obtain the angular energy distribution curve. Find the points with the highest energy on this curve; the angles corresponding to these points are the main directions of the texture. Along each main direction, search for local maxima on the amplitude spectrum from the origin in the frequency domain; these points are the frequency peak points of the main direction. Calculate the Euclidean distances from all found peak points to the origin in the frequency domain, and then calculate the arithmetic mean of these Euclidean distances to obtain the average distance.

[0022] For each peak point and its conjugate symmetric point in the frequency domain, a Gaussian band-stop filter is constructed such that the center of the filter is located at the peak point. By multiplying all the Gaussian band-stop filters point by point, a texture suppression filter with a minimum value (i.e., a valley value) at all peak points is generated.

[0023] In an optional embodiment, identifying multiple main direction frequency peaks representing periodic textures based on the amplitude spectrum angular energy distribution includes: dividing the amplitude spectrum into 180 angular sectors centered on the frequency domain origin, with each sector having an angular range of 1°; calculating the sum of the energy of all frequency points within each sector to obtain the total energy of each sector; calculating the average value of the total energy of all sectors, and determining the sectors with a total energy greater than 1.2 times the average value as main direction sectors; and within each main direction sector, searching for the frequency point with the largest amplitude as the main direction frequency peak point.

[0024] Specifically, the amplitude spectrum of 512×512 pixels is divided into 180 identical sector regions with the center point as the origin, and each sector represents an angle range of 1°. For example, a round cake is cut into 180 slices, and each slice is an angle sector.

[0025] Next, the energy of each sector is calculated. For example, for a sector with an angle of 30°, all frequency points falling within that sector are identified, the amplitude of each point is squared, and then summed to obtain the total energy of the sector. For instance, if there are three points in the sector with amplitudes of 10, 20, and 5, the total energy is 525. This process is repeated for all 180 sectors, resulting in a list of 180 energy values.

[0026] Next, the average of the 180 energy values ​​is calculated, and a threshold of 1.2 times the average is set. The total energy of each sector is then checked. If the total energy of a sector is greater than 1.2 times the average, it is identified as the primary direction sector. For example, if the average of the 180 energy values ​​is 400, and the total energy of the previously calculated 30° sector is 525, which is greater than 1.2 times the average, then the 30° sector is identified as the primary direction sector.

[0027] Finally, find the frequency point with the largest amplitude within the main direction sector and designate this point as a main direction frequency peak point. Repeat this operation for all designated main direction sectors to find all peak points representing texture directions.

[0028] In an optional embodiment, generating a texture suppression filter with a valley value at the corresponding position based on the position of the peak point includes: multiplying multiple Gaussian band-stop filters to obtain the texture suppression filter, for the first... Peak points The Gaussian band-stop filter corresponding to the peak point The transfer function is: , in, For point to peak point distance, is the standard deviation of the Gaussian function.

[0029] For example, assuming the identified texture peak point has coordinates (80, 100) in the frequency domain, a filter is created to suppress that specific frequency and the frequencies surrounding it. Using the Gaussian band-stop filter formula, the standard deviation is set to 5, which determines the size of the suppression region; the smaller the value, the narrower the suppression range.

[0030] Create a filter template of the same size as the frequency domain image, for example, 512×512. For each point on the template, calculate its distance to the peak point (80, 100). For example, for point (81, 102), the square of the distance to the peak point is 5. Substituting this distance into the formula, the calculated filter value will be very close to 0; while for points far from the peak point, such as point (300, 300), the calculated filter value will be very close to 1, thus generating a filter with a deep valley at (80, 100) and a flat surface elsewhere.

[0031] If multiple peaks are identified, such as one at (-80, -100), a similar band-stop filter is generated for that point. Multiplying these two or more filters point-by-point yields a texture suppression filter. This filter creates deep valleys at all identified peak locations, effectively suppressing all major texture frequencies simultaneously.

[0032] Step S2: Generate a set of Gaussian high-pass filters with different cutoff frequencies based on the average distance, and multiply the texture suppression filter with each Gaussian high-pass filter to obtain multiple composite filters.

[0033] Specifically, using the average distance calculated in step S1 as a reference, a set of Gaussian high-pass filters with different cutoff frequencies are generated. The cutoff frequencies of the Gaussian high-pass filters are set according to different scaling factors based on the average distance.

[0034] In this embodiment, three Gaussian high-pass filters are set, and their cutoff frequencies are set to 0.2 times, 0.5 times, and 1.0 times the average distance, respectively.

[0035] For each cutoff frequency, a corresponding Gaussian high-pass filter is generated using the standard formula for a Gaussian high-pass filter. The previously generated texture suppression filter is then multiplied point-by-point by the three Gaussian high-pass filters to obtain three composite filters.

[0036] For example, if the average distance from all texture peaks to the origin in the frequency domain is 150 pixels, the cutoff frequencies of three Gaussian high-pass filters are set using this average distance, wherein the cutoff frequencies of the three Gaussian high-pass filters are set to 30, 75, and 150, respectively.

[0037] Based on the aforementioned cutoff frequencies, three Gaussian high-pass filters are generated. The first filter removes extremely low-frequency information with a frequency radius less than 30, the second removes low-frequency signals with a frequency radius less than 75, and the third removes mid-to-low-frequency signals with a frequency radius less than 150. This results in three frequency domain signals of different scales that retain different high-frequency details, preparing for subsequent fusion and enhancement of defect information.

[0038] Step S3: Filter the frequency domain signal using the multiple composite filters to obtain multiple filtered frequency domain signals. Determine the normalized fusion weight based on the total energy of each filtered frequency domain signal, and perform weighted fusion on the multiple filtered frequency domain signals to obtain the final frequency domain signal.

[0039] Specifically, the original frequency domain signal is multiplied point-by-point by multiple composite filters to obtain multiple filtered frequency domain signals. For each filtered frequency domain signal, the total energy of the signal is calculated, which is done by summing the squares of the values ​​of all pixels in the amplitude spectrum. Based on the magnitude of each energy value, a normalized weight is calculated for each. Taking three composite filters as an example, the corresponding frequency domain signals are as follows: , , ,Signal weight = The signal was calculated in the same way. weight and signal weight ,in , , Frequency domain signals , , The total energy. The three filtered frequency domain signals. , , Multiply by the corresponding normalized weights respectively , , The results are added together point by point to obtain the final frequency domain signal after fusion.

[0040] In an optional embodiment, the normalized fusion weights are determined based on the total energy of each filtered frequency domain signal, including: for the first... The filtered frequency domain signal Calculate the total energy of the frequency domain signal. : ; Calculate the first Normalized fusion weights of individual signals : ,in This is the index for all filtered frequency domain signals.

[0041] The normalized fusion weights reflect the relative amount of information contained in each filtered signal; signals with higher energy will have a larger weight in subsequent fusion.

[0042] Step S4: Perform inverse Fourier transform and exponential transform on the final frequency domain signal to obtain an enhanced image; calculate the gradient magnitude map of the corrugated paper image to be detected, and perform threshold segmentation on the enhanced image to obtain candidate defect regions; for any candidate defect region, when the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold, the candidate defect region is determined to be the final defect region.

[0043] Specifically, a two-dimensional inverse discrete Fourier transform is performed on the final frequency domain signal to obtain a spatial domain image. Perform an exponential operation on each pixel value of the image, i.e., e^(-e^(-e)). The result is raised to the power of 1, minus 1, and the initial logarithmic transform is reversed to obtain an image with enhanced defect features. The original image of the corrugated paper to be detected. The Sobel operator is applied to calculate the gradients of the corrugated paper image to be detected in the horizontal and vertical directions, respectively. The square root of the sum of the squares of the gradients in the two directions is then calculated to obtain the gradient magnitude map. Simultaneously, the enhanced image... Threshold segmentation is performed using the local mean method. For each pixel, the mean value of pixels within its neighborhood window is calculated as a threshold. Pixels with values ​​higher than this threshold are marked as foreground pixels, forming a binary image. Connectivity analysis is performed on this binary image, identifying each independent white region as a candidate defect region. For each candidate defect region, the corresponding pixel set in the gradient magnitude map is found, and the average gradient value of that pixel is calculated. This average gradient value is compared with a pre-set gradient threshold. If the average gradient value is greater than the threshold, the candidate region is confirmed as a true final defect region; otherwise, it is determined as a false defect and discarded.

[0044] To process the enhanced image pixel by pixel, in an optional embodiment, thresholding is performed on the enhanced image to obtain candidate defect regions, including: The Niblack thresholding algorithm is used, with a window size of 15×15 pixels. The thresholding is based on the average value of the pixels within the window. and standard deviation Calculate the segmentation threshold The formula for calculating the segmentation threshold is: ,in, The segmentation threshold is... This represents the average value of the pixels within the window. is the standard deviation of the pixels within the window.

[0045] Specifically, a neighborhood window of size 15×15 is taken around a certain pixel in the image.

[0046] Calculate the average grayscale value of all pixels within the 15×15 window. and standard deviation The standard deviation reflects the dispersion of pixel gray values ​​within a window. If there are edges or textures in the area, the standard deviation will be larger.

[0047] The segmentation threshold for the center pixel is calculated using the Niblack formula. The original grayscale value of the center pixel is compared with the segmentation threshold. If the grayscale value is less than the segmentation threshold, the center pixel is marked as a candidate defect point and set to black; if it is greater than or equal to the segmentation threshold, it is marked as background and set to white. This process is repeated for each pixel in the image to generate a binary image with only black and white, where the black areas are the candidate defect areas.

[0048] In an optional embodiment, calculating the gradient magnitude map of the corrugated paper image to be detected includes: using the Sobel operator to calculate the gradient in the horizontal direction and the gradient in the vertical direction of the corrugated paper image to be detected; calculating the gradient magnitude of each pixel based on the gradient in the horizontal direction and the gradient in the vertical direction, and the gradient magnitudes of all pixels together constitute the gradient magnitude map.

[0049] Specifically, prepare the original corrugated cardboard image and use two 3×3 Sobel operator templates, one for detecting horizontal edges and the other for detecting vertical edges, calculating the gradient pixel by pixel. For example, for a specific pixel in the image, place the horizontal Sobel operator over that pixel, multiply the 3×3 coefficients by the corresponding pixel grayscale value, and then sum them to obtain the horizontal gradient component at that point. Using the same method, the vertical gradient component at that point is calculated using the vertical Sobel operator. The horizontal and vertical gradient components represent the degree and direction of grayscale change at a point in the horizontal and vertical directions, respectively.

[0050] The total gradient magnitude at that point is calculated using the formula described above. This calculation is then applied to every pixel of the original image to obtain a new image, the gradient magnitude map. The brightness of each point on this map represents the edge intensity at the corresponding location in the original image.

[0051] In an optional embodiment, when the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold, the candidate defect region is determined to be the final defect region, including: calculating the average gradient value of all pixels in the entire gradient magnitude map. Set the preset gradient threshold to 1.5 times.

[0052] Specifically, the entire gradient magnitude map is analyzed, the sum of the gradient values ​​of all pixels in the map is calculated, and then divided by the total number of pixels to obtain the global average gradient value. .

[0053] The gradient threshold is set based on the global average gradient. The threshold is set to... The threshold value is 1.5 times the set threshold value, and can be used to distinguish between real defect edges and normal texture edges of corrugated paper itself. Each previously segmented candidate defect region is examined. For each candidate region, all pixels corresponding to it on the gradient magnitude map are found, and the average gradient value of the pixels within the candidate region is calculated. If the average gradient value is greater than the set threshold, the candidate region is determined to be a real final defect region. Conversely, if the average gradient value of the candidate region is less than the set threshold, it is considered to be just ordinary background texture and is excluded.

[0054] This invention also provides a machine vision-based corrugated paper defect detection system. For example... Figure 2 As shown, the system includes a processor and a memory, the memory storing computer program instructions, which, when executed by the processor, implement the machine vision-based corrugated paper defect detection method according to the present invention.

[0055] 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 therefore will not be described in detail here.

[0056] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented by computer-readable / executable instructions stored or otherwise maintained on such a computer-readable medium.

[0057] In the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.

[0058] While various embodiments of the invention have been shown and described in this specification, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention.

Claims

1. A machine vision-based method for detecting defects in corrugated paper, characterized in that, Includes the following steps: The process involves acquiring an image of the corrugated paper to be detected, performing a logarithmic transform and a Fourier transform on the image to obtain a frequency domain signal, calculating the amplitude spectrum of the frequency domain signal, identifying multiple main direction frequency peaks representing periodic textures based on the angular energy distribution of the amplitude spectrum, calculating the average distance from the peaks to the origin of the frequency domain, and generating a texture suppression filter with valleys at the corresponding positions based on the positions of the peaks. A set of Gaussian high-pass filters with different cutoff frequencies is generated based on the average distance, and the texture suppression filter is multiplied by each Gaussian high-pass filter to obtain multiple composite filters; The frequency domain signal is filtered by the multiple composite filters to obtain multiple filtered frequency domain signals. The normalized fusion weight is determined based on the total energy of each filtered frequency domain signal, and the multiple filtered frequency domain signals are weighted and fused to obtain the final frequency domain signal. The final frequency domain signal is subjected to inverse Fourier transform and exponential transform to obtain an enhanced image; Calculate the gradient magnitude map of the corrugated paper image to be detected, and perform threshold segmentation on the enhanced image to obtain candidate defect regions; for any candidate defect region, if the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold, then the candidate defect region is determined to be the final defect region.

2. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The identification of multiple principal direction frequency peaks representing periodic textures based on the amplitude spectrum angular energy distribution includes: The amplitude spectrum is divided into 180 angular sectors centered on the origin of the frequency domain, with each sector having an angular range of 1°. Calculate the sum of the energy at all frequency points within each sector to obtain the total energy of each sector; Calculate the average total energy of all sectors, and determine the sectors with total energy greater than 1.2 times the average as the main direction sectors; Within each main direction sector, the frequency point with the largest amplitude is searched as the main direction frequency peak point.

3. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The step of generating a texture suppression filter with a valley value at the corresponding position based on the position of the peak point includes: The texture suppression filter is obtained by multiplying multiple Gaussian band-stop filters. For the first... Peak points The Gaussian band-stop filter corresponding to the peak point The transfer function is: ， in, For point to peak point distance, is the standard deviation of the Gaussian function.

4. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The cutoff frequencies of the set of Gaussian high-pass filters with different cutoff frequencies are set according to the average distance using different scaling factors.

5. The machine vision-based defect detection method for corrugated paper according to claim 4, characterized in that, The Gaussian high-pass filter is set to three, and its cutoff frequencies are set to 0.2 times, 0.5 times, and 1.0 times the average distance, respectively.

6. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The determination of normalized fusion weights based on the total energy of each filtered frequency domain signal includes: For the The filtered frequency domain signal Calculate the total energy of the frequency domain signal. : ； Calculate the first Normalized fusion weights of individual signals : ,in This is the index for all filtered frequency domain signals.

7. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The step of thresholding the enhanced image to obtain candidate defect regions includes: The Niblack thresholding algorithm is used, with a window size of 15×15 pixels. The thresholding is based on the average value of the pixels within the window. and standard deviation Calculate the segmentation threshold The formula for calculating the segmentation threshold is: ,in, The segmentation threshold is... This represents the average value of the pixels within the window. is the standard deviation of the pixels within the window.

8. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The calculation of the gradient magnitude map of the corrugated paper image to be detected includes: The Sobel operator is used to calculate the gradients in the horizontal and vertical directions of the corrugated paper image to be detected. Based on the gradients in the horizontal and vertical directions, the gradient magnitude of each pixel is calculated, and the gradient magnitudes of all pixels together constitute a gradient magnitude map.

9. The machine vision-based defect detection method for corrugated paper according to claim 1, characterized in that, The step of determining the candidate defect region as the final defect region when the average gradient value of the candidate defect region at the corresponding position in the gradient magnitude map is greater than a preset gradient threshold includes: Calculate the average gradient value of all pixels in the entire gradient magnitude map. ; Set the preset gradient threshold to 1.5 times.

10. A machine vision-based corrugated paper defect detection system, characterized in that, include: processor; A memory storing computer instructions for corrugated paper defect detection using machine vision, which, when executed by the processor, cause the system to perform the machine vision method for corrugated paper defect detection according to any one of claims 1-9.

Images