A method for detecting surface defects of silicate fireproof plate based on machine vision

Abstract

The present application belongs to the technical field of image processing, and particularly relates to a silicate fireproof plate surface defect detection method based on machine vision, which comprises the following steps: first, a line array camera is used in cooperation with a double-channel light source to respectively acquire a luminosity image representing material albedo and a topography image representing surface micro relief; then, a micro texture index is constructed by combining gradient amplitude and gray deviation to generate a defect probability graph that suppresses mineral interference; finally, suspected areas are verified by using the cumulative characteristics of edge texture gradient to calculate the penetration and diffusion confidence. The present application utilizes the difference between the penetration characteristics of real defects and the physical truncation characteristics of natural minerals, improves the accuracy of distinguishing the two in gray scale, reduces the false positive rate, and improves the detection accuracy.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a machine vision-based method for detecting surface defects in silicate fireproof boards. Background Technology

[0002] Silicate fireproof boards are important building materials used in decoration and fire protection engineering, typically belonging to the category of porous media materials. Their surface microstructure is relatively complex, mainly composed of a crisscrossing fiber network and distributed micropores. In actual industrial production and processing, the substrate of silicate fireproof boards often contains dark or black solid mineral particles embedded within it due to the characteristics of the natural mineral raw materials. Simultaneously, due to mechanical lubrication and gluing processes on the production line, the surface of the boards is easily contaminated by oil, glue, and other fluid or semi-fluid substances, resulting in appearance defects. To ensure product quality, automated detection and removal of these surface defects are necessary.

[0003] Existing detection methods typically employ grayscale thresholding algorithms, gradient-based edge detection algorithms, or basic morphological processing. Grayscale thresholding algorithms mark dark areas as defects by setting specific brightness limits; while edge detection algorithms delineate the target contour by calculating abrupt changes in pixel grayscale.

[0004] However, in the defect detection scenario of silicate fireproof boards, the aforementioned traditional algorithms have certain shortcomings and are difficult to meet the requirements of high-precision detection. Naturally occurring mineral particles and defects such as oil stains and glue spots usually appear as black or dark spots with low grayscale under visible light imaging, so grayscale-based analysis has a certain false detection rate. In addition, existing technologies ignore the differences in microscopic morphology after the interaction between defective substances and the substrate. Real oil stain defects, as fluids, are permeable and fill the micropores on the surface of the board, making its surface micro-texture tend to be smooth. Mineral particles, on the other hand, are embedded solids, and their surfaces often retain their original roughness or have sharp physical fractures, resulting in poor detection results. Summary of the Invention

[0005] To address the technical problem of insufficient accuracy in detecting surface defects in silicate fireproof boards, this invention provides a machine vision-based method for detecting surface defects in silicate fireproof boards, comprising:

[0006] A photometric image containing material albedo information and a morphological image containing surface micro-undulation information are acquired. The micro-texture index of the morphological image is calculated to obtain a micro-texture index map. The value of the micro-texture index is positively correlated with the weighted average of the gradient magnitudes of all pixels in a local neighborhood of the morphological image, and the weight of the weighted average is negatively Gaussian correlated with the difference between the corresponding pixel grayscale value and the local neighborhood grayscale average value. Based on the photometric image and the micro-texture index map, a defect probability map is obtained that is positively correlated with the inversion value of the grayscale value of the photometric image and negatively correlated with the micro-texture index. Connected component segmentation is performed on the defect probability map to obtain suspected connected components. The penetration and diffusion confidence of each suspected connected component is obtained. The penetration and diffusion confidence is positively correlated with the mean of the defect probability within the corresponding connected component and the reciprocal of the cumulative gradient value of the micro-texture index along the edge of the connected component. In response to the penetration and diffusion confidence being greater than a first determination threshold, the corresponding suspected connected component is determined to be a real defect and located.

[0007] This invention combines photometric and morphological images, utilizing a Gaussian negative correlation between weighted mean and grayscale difference to calculate a microtexture index. This allows for precise assessment of surface microroughness, thereby identifying the smooth features caused by fluid defects filling micropores. Furthermore, by introducing a penetration-diffusion confidence level, this invention differentiates the gentle edge gradients generated by real fluid defects in porous media from the sharp edges of solid minerals, eliminating interference from minerals with high grayscale similarity and improving detection accuracy and robustness.

[0008] Preferably, the acquisition of the photometric image and the morphological image includes:

[0009] A single scan of the silicate fireproof board was performed using an imaging system comprising a linear array camera and a dual-channel light source; channel A used a high-angle coaxial light source to acquire the photometric image; channel B used a low-angle grazing light source to acquire the morphological image.

[0010] Preferably, the microtexture index satisfies the expression:

[0011] ;

[0012] In the formula, This represents the microtexture index of the c-th pixel in the topography image; This represents the number of pixels in the local window of the c-th pixel in the topographic image; This represents the gradient magnitude of the i-th local window pixel at the c-th pixel of the topography image; This represents the grayscale value of the i-th local window pixel at the c-th pixel of the topographic image; This represents the average gray value of the local window at the c-th pixel in the topographic image. The standard deviation constant representing the inherent texture noise of the board material; Represents the natural exponential function; Represents the absolute value symbol.

[0013] This invention achieves robust extraction of texture features by constructing weights using an adaptive Gaussian function based on local neighborhood gray-level differences through a micro-texture index expression. This expression assigns a larger weight to the gradient magnitude when the pixel gray-level is close to the local mean, and reduces the weight when there are abrupt changes in gray-level. This suppresses inherent texture noise in the board material while capturing the smooth changes in micro-texture caused by fluid filling micropores, thus obtaining subtle differences in surface roughness.

[0014] Preferably, obtaining the standard deviation constant of the inherent texture noise of the board material includes:

[0015] Select a defect-free normal board area, calculate the standard deviation of pixel grayscale within the normal board area, and use it as the standard deviation constant.

[0016] Preferably, obtaining the local window of the c-th pixel in the shape image includes:

[0017] Centered on the c-th pixel of the topography image, obtain a neighborhood region of a preset size, which is denoted as the local window of the c-th pixel of the topography image.

[0018] Preferably, the defect probability satisfies the expression:

[0019] ;

[0020] In the formula, This represents the defect probability of the c-th pixel in the photometric image; This represents the grayscale value of the c-th pixel in the photometric image; This represents the microtexture index of the c-th pixel in the topography image; This represents the preset texture suppression coefficient; This represents the preset nonlinear adjustment factor.

[0021] This invention uses the texture index as a suppression term. When the roughness of the region is high, the probability of defects is automatically reduced. Only when the region is both black and smooth will a high probability be output, thereby suppressing pseudo-defects in black ore and preserving the response of real defects.

[0022] Preferably, the step of performing connected component segmentation on the defect probability graph to obtain suspected connected components includes:

[0023] The defect probability map is thresholded using the Otsu method to obtain a binarized image; the foreground pixels in the binarized image are clustered using the 8-connected component labeling algorithm to obtain each suspected connected component.

[0024] Preferably, the penetration diffusion confidence level satisfies the expression:

[0025] ;

[0026] In the formula, Indicates the first The penetration and diffusion confidence of a suspected connected domain; Indicates the first The average defect probability value within a suspected connected domain; Indicates the first The number of edge contour pixels of a suspected connected region; On the microtexture index map, corresponding to the first... The gradient value of the microtexture index of the u-th edge contour pixel of a suspected connected domain; This indicates the microtexture index map corresponding to the first... A set of edge contour pixels of a suspected connected domain, and a set of corresponding microtexture indices; Represents the maximum value function; The sign indicating the magnitude of a vector; Represents the normalization function; It represents a tiny positive value.

[0027] This invention measures the sharpness of edges by calculating the cumulative magnitude of the micro-texture gradient at the edges of suspected connected domains. It utilizes the physical difference between fluid defect edges, where the transition zone causes a gentle gradient change, and solid mineral edges, where the sharpness causes abrupt gradient changes, to map edge gradient features into confidence values, thereby distinguishing between embedded solid minerals and attached fluid stains.

[0028] Preferably, the method further includes classifying the actual defects into different types:

[0029] True defects with a penetration and diffusion confidence level greater than or equal to the second judgment threshold are judged as serious defects and defined as deep adhesive spots; true defects with a penetration and diffusion confidence level less than the second judgment threshold but greater than the first judgment threshold are judged as minor defects and defined as shallow oil stains.

[0030] This invention subdivides the detection results into deep adhesive stains and shallow oil stains, providing not only a determination of the existence of defects but also information on the attributes of the defects. This helps the production line to take different handling measures according to the type of defect, such as wiping, scrapping, or downgrading, thereby improving the level of precision in quality control.

[0031] Preferably, the positioning includes:

[0032] Obtain the center coordinates of the bounding rectangle of the actual defect, which will be used as the location of the actual defect.

[0033] The beneficial effects of this invention are as follows:

[0034] (1) The present invention separates the blackness information of the material from the morphology information of the surface through a dual-channel imaging scheme based on physical light field decoupling, thereby eliminating the visual interference caused by natural mineral spots on the surface of silicate fireproof board and improving the accuracy of distinguishing minerals and oil stains in a single gray dimension.

[0035] (2) This invention identifies defects by analyzing the changes in micro-texture caused by fluid filling porous media. Even when the oil stains are very light in color or have dried and the gray level is not obvious, they can still be effectively detected by the change in texture features, thus improving the sensitivity of detection.

[0036] (3) By analyzing the distribution characteristics of edge gradients, this invention can identify and eliminate solid mineral interference with sharp edge characteristics, accurately lock fluid defects with gradual diffusion characteristics, thereby reducing the false alarm rate on the production line and ensuring the reliability of the detection results. Attached Figure Description

[0037] Figure 1 This is a flowchart illustrating a machine vision-based method for detecting surface defects in silicate fireproof boards according to the present invention.

[0038] Figure 2 This is a schematic diagram illustrating the microtexture index.

[0039] Figure 3 This is a schematic representation of the defect probability map. Detailed Implementation

[0040] This invention discloses a machine vision-based method for detecting surface defects in silicate fireproof boards, referring to... Figure 1 This includes steps S1-S4:

[0041] S1: Construct a physical light field decoupled imaging system that includes a linear array camera and a dual-channel light source, and control the linear array camera to perform a single scan of the silicate fireproof board in conjunction with the dual-channel light source; obtain a photometric image characterizing the albedo of the material under high-angle coaxial illumination conditions, and obtain a morphological image characterizing the micro-undulations of the surface under low-angle grazing illumination conditions.

[0042] It should be noted that silicate fireproof boards, as a porous medium, have a surface composed of complex fibers and micropores. In actual production, natural minerals, as part of the substrate, are embedded solids, while real defects such as oil stains and glue are usually surface-adhered fluids or semi-fluids. In conventional single-light imaging, black natural minerals and black oil stains are difficult to distinguish in grayscale, leading to a high false alarm rate. Therefore, this invention first decomposes color and shape, utilizing the physical characteristics of optical imaging to obtain the material blackness and surface roughness of the object separately.

[0043] Specifically, a physical light field decoupled imaging system is constructed, comprising a linear scan camera and a dual-channel light source. The linear scan camera, in conjunction with the dual-channel light source, performs a single scan of a silicate fireproof board, including:

[0044] A high-resolution line scan camera is used as the image acquisition unit, and in conjunction with a time-division stroboscopic controller, two images are output in a single scan: Channel A is set as the photometric imaging channel, using a high-angle coaxial light source with perpendicular incident light; Channel B is set as the topographic imaging channel, using a low-angle grazing light source, for example, with an incident angle of [missing information]. .

[0045] Preferably, obtaining a photometric image characterizing the material's albedo under high-angle coaxial illumination conditions and obtaining a morphological image characterizing the surface's micro-undulations under low-angle grazing illumination conditions includes:

[0046] A photometric image is obtained using channel A, denoted as Under this light field, the diffuse reflection component of the object's surface dominates, mainly reflecting the albedo information of the object's surface. Black natural minerals and black oil stains both appear as dark spots with low grayscale.

[0047] The topographic image is obtained using channel B, denoted as Under this light field, as light passes over the surface, the micropores and raised particles of the substrate will produce significant shadows, while the areas filled by the fluid will have fewer shadows due to the smooth surface, mainly reflecting the micro-texture information of the object's surface.

[0048] At this point, both the photometric and morphological images have been obtained.

[0049] S2: Based on the gradient magnitude of the neighboring pixels of the topography image, and combined with the Gaussian weighted model based on gray-level deviation, obtain the micro-texture index that characterizes micro-roughness, and obtain the micro-texture index map of the topography image.

[0050] It should be noted that while the human eye can observe texture differences in morphological images, machines require accurate evaluation metrics. Real defects, such as oil stains, are fluid and can penetrate and fill the micropores on the surface of silicate fireproof boards, resulting in a lower micro-roughness in that area compared to normal areas. To evaluate this characteristic, this invention constructs a micro-texture index, introduces gradient amplitude to characterize the severity of texture changes, and utilizes grayscale deviation to construct Gaussian weights to highlight the texture features of the microporous structure, thereby accurately quantifying the micro-roughness of the surface.

[0051] Specifically, based on the gradient magnitude of neighboring pixels in the topography image, and combined with a Gaussian weighted model based on gray-level deviation, a microtexture index characterizing micro-roughness is obtained, resulting in a microtexture index map of the topography image, including:

[0052] A local window is defined for any pixel in the topographic image, representing a neighborhood region centered on that pixel. For example, the size of the local window is... Pixel.

[0053] Obtain the average grayscale value of the local window of the pixel.

[0054] A standard deviation constant for the inherent texture noise of the pre-defined board material is set. It should be noted that this standard deviation constant is used to normalize grayscale differences. It is obtained by selecting a defect-free, normal board material area and calculating the standard deviation of pixel grayscale values ​​within that area, using this standard deviation constant as the standard deviation constant. For example, the standard deviation constant is set to... .

[0055] It should be noted that simply summing gradient magnitudes is susceptible to interference from isolated noise points and cannot distinguish between background texture and abrupt changes. By constructing weights using a Gaussian function, a larger weight is assigned when the pixel's grayscale value is close to the local mean, representing a flat region or uniform background, allowing the gradient features at that location to be referenced more extensively. Conversely, the weight is reduced when the pixel's grayscale value deviates from the mean, representing noise or unstructured extrema. This allows for robust extraction of the micro-texture index reflecting surface roughness.

[0056] The microtexture index of any pixel in a topographic image satisfies the expression:

[0057] ;

[0058] In the formula, This represents the microtexture index of the c-th pixel in the topography image; This represents the number of pixels in the local window of the c-th pixel in the topographic image; This represents the gradient magnitude of the i-th local window pixel at the c-th pixel of the topography image; This represents the grayscale value of the i-th local window pixel at the c-th pixel of the topographic image; This represents the average gray value of the local window at the c-th pixel in the topographic image. The standard deviation constant representing the inherent texture noise of the board material; Represents the natural exponential function; Represents the absolute value symbol.

[0059] In the formula, It represents the degree of gradient oscillation of the local pixel of the c-th pixel in the topography image, reflecting the neighborhood roughness of the c-th pixel in the topography image; By constructing an adaptive Gaussian weighting function, the confidence of different local window pixels of the c-th pixel in the topography image is weighted. The microtexture index is formed by weighting the gradients of different local window pixels of the c-th pixel in the topography image using the Gaussian weighting function. The larger the value, the larger the gradient of the neighboring pixels of the c-th pixel in the topography image, and the more the gray level deviates from the normal situation. Therefore, the microtexture index is larger.

[0060] The microtexture indexes of all pixels in the topography image are sorted according to the pixel coordinates to obtain the microtexture index map of the topography image.

[0061] It should be noted that, as Figure 2 This is a microtexture index map, which shows the microtexture index of each pixel in the topography image. The color bars range from 0 to 100.

[0062] At this point, the microtexture index map of the topographic image has been obtained.

[0063] S3: Set the texture suppression coefficient and nonlinear adjustment factor, and combine the pixel gray values ​​of the microtexture index map and the photometric image to obtain a defect probability map that suppresses mineral interference.

[0064] It should be noted that photometric images cannot distinguish between minerals and stains, and morphological images cannot distinguish between flat stains and relatively flat substrate areas. Therefore, this invention uses a microtexture index map to correct photometric images. Considering that although natural minerals are black, their surfaces are still rough, this invention constructs a defect probability index so that high roughness features can suppress the scores of photometric anomalies, retaining only areas with abnormal color and texture as suspected defects.

[0065] Specifically, by setting a texture suppression coefficient and a nonlinear adjustment factor, and combining the pixel grayscale values ​​of the microtexture index map and the photometric image, a defect probability map suppressing mineral interference is obtained, including:

[0066] Set the texture suppression coefficient, denoted as . This is used to control the suppression strength of the microtexture index on the photometric score. For example, Set as It should be noted that if Setting it too high, such as This can lead to an overly strong inhibition effect, causing some slightly rough, shallow oil stains to be misidentified as minerals and suppressed, resulting in missed detection; if The setting is too small, such as If the inhibition is insufficient, the minerals cannot be effectively filtered out, leading to false alarms.

[0067] Set a nonlinear adjustment factor, denoted as... This is used to adjust the model's sensitivity to changes in texture energy. For example, Set as It should be noted that, The steepness of the suppression curve is determined by the square relationship, which enhances the suppression effect of high microtexture index, causing the suppression term to decay rapidly in areas with slightly higher roughness.

[0068] It should be noted that the purpose of constructing the defect probability is to output the probability value, which is such that the probability approaches the maximum value when the gloss is darker and the texture is smoother; conversely, if the texture is rough, the probability will decrease even if the gloss is satisfied.

[0069] The defect probability of any pixel in a photometric image satisfies the expression:

[0070] ;

[0071] In the formula, This represents the defect probability of the c-th pixel in the photometric image; This represents the grayscale value of the c-th pixel in the photometric image; This represents the microtexture index of the c-th pixel in the topography image; Indicates the texture suppression coefficient; This represents a nonlinear adjustment factor.

[0072] In the formula, These constitute photometric anomalies; the larger the value, the darker the color and the higher the likelihood of it being an anomaly. The texture suppression term is constructed, and the suppression logic is implemented using an inverse proportional function structure. When the value is large, the roughness is high, corresponding to natural minerals. The denominator increases, causing the texture suppression term to approach a certain value. This makes the defect probability approach the value. It is judged as a false defect; when When the value is small, the surface is smooth, corresponding to real defects, and the denominator approaches the value. Texture suppression term approaches This allows the complete retention of the score for the photometric anomaly item, which is then judged as a true defect.

[0073] Based on coordinate order, the defect probabilities of all pixels in the photometric image are used to construct a defect probability map.

[0074] It should be noted that, as Figure 3 This is a defect probability map, which shows the defect probability of each pixel in the photometric image. The color bars range from 0 to 50.

[0075] At this point, the defect probability map has been obtained.

[0076] S4: Perform connected component segmentation on the defect probability map to obtain suspected connected components and their edge contours; based on the gradient change characteristics of the microtexture index map at the edge contours, obtain the penetration and diffusion confidence of each suspected connected component, and combine it with the average probability value to complete the verification of the real defect.

[0077] It should be noted that while the defect probability map can identify most minerals, a small number of interferences with indistinct edge features may still exist. This invention further introduces depth scene features and the capillary penetration effect. The bond between solid minerals and the substrate is physically truncated, with its roughness changing abruptly from present to absent, resulting in sharp boundaries. In contrast, oil or adhesives undergo capillary penetration in porous media, causing micropores to transition from complete filling to partial filling, and then back to unfilled areas, with roughness energy exhibiting a gradual gradient decay at the boundaries. Therefore, this invention establishes a penetration-diffusion confidence index for the final detection and verification of defects.

[0078] Specifically, the defect probability map is segmented into connected components to obtain suspected connected components and their edge contours, including:

[0079] Threshold segmentation and connected component labeling are performed on the defect probability map to obtain several suspected connected components; the edge contour pixel set of any suspected connected component is obtained; the average defect probability value within the suspected connected component is calculated as the basic confidence level of the suspected connected component. The threshold segmentation is performed using Otsu's method, and the connected component labeling uses the 8-connected component labeling algorithm to cluster the foreground pixels in the binary mask, dividing spatially adjacent pixel sets into independent suspected connected components.

[0080] It should be noted that, in order to distinguish between sharp boundaries and permeable boundaries, it is necessary to calculate the gradient modulus of the microtexture index map at the edge. If the gradient modulus is large, it indicates that the roughness has changed abruptly, and it is more likely to be a mineral. If the gradient modulus is small, it indicates that the roughness is gradual, and it is more likely to be oil.

[0081] Preferably, based on the gradient change characteristics of the microtexture index map at the edge contour, the penetration and diffusion confidence of each suspected connected domain is obtained, and the verification of the real defect is completed by combining the average probability value, including:

[0082] The penetration-diffusion confidence of any suspected connected component satisfies the expression:

[0083] ;

[0084] In the formula, Indicates the first The penetration and diffusion confidence of a suspected connected domain; Indicates the first The average defect probability value within a suspected connected domain; Indicates the first The number of edge contour pixels of a suspected connected region; On the microtexture index map, corresponding to the first... The gradient value of the microtexture index of the u-th edge contour pixel of a suspected connected domain; This indicates the microtexture index map corresponding to the first... A set of edge contour pixels of a suspected connected domain, and a set of corresponding microtexture indices; Represents the maximum value function; The sign indicating the magnitude of a vector; Represents the normalization function; This indicates a small positive value, used to prevent the denominator from being zero. For example... .

[0085] In the formula, Indicates the first The cumulative gradient magnitude of the microtexture index of all edge contour pixels of a suspected connected region reflects the... The degree of roughness change at the edges of a suspected connected region; the larger the value, the more severe the roughness change at the edges of the suspected connected regions. The more drastic the change in the microscopic texture of the edge of a suspected connected domain; This is a normalized benchmark for the severity of the abrupt change, representing the maximum possible mutation. If it is an embedded solid mineral, its boundaries are relatively sharp, and the gradient modulus of the microtexture index map is large, leading to... Increase The smaller the value, the lower the confidence level of penetration and diffusion; if it is penetrating oil, there is a transition zone at its boundary, and the gradient modulus of the microtexture index map is smaller and gentler. Smaller Item close This allows for maintaining a high confidence level in penetration and diffusion.

[0086] A first and a second judgment threshold are set. Suspected connected regions with a penetration-diffusion confidence level greater than or equal to the first judgment threshold are recorded as true defects. The center coordinates of the bounding rectangle of each true defect are obtained as the defect location. True defects with a penetration-diffusion confidence level less than the second judgment threshold are marked as minor defects, and their defect type is determined to be shallow oil contamination. True defects with a penetration-diffusion confidence level greater than or equal to the second judgment threshold are marked as severe defects, and their defect type is determined to be deep adhesive residue. An example first judgment threshold is 0.6, and the second judgment threshold is 0.8.

[0087] This completes the defect detection of the silicate fireproof board surface.

[0088] 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 method for detecting surface defects in silicate fireproof boards based on machine vision, characterized in that, include: Acquire photometric images containing material albedo information and morphological images containing surface micro-undulation information; Calculate the microtexture index of the topography image. , In the formula, This represents the number of pixels in the local window of the c-th pixel in the topographic image; This represents the gradient magnitude of the i-th local window pixel at the c-th pixel of the topography image; This represents the grayscale value of the i-th local window pixel at the c-th pixel of the topographic image; This represents the average gray value of the local window at the c-th pixel in the topographic image. The standard deviation constant representing the inherent texture noise of the board material; Represents the natural exponential function; The absolute value sign is used to obtain the micro-texture index map. The value of the micro-texture index is positively correlated with the weighted average of the gradient magnitudes of all pixels in the local neighborhood of the topography image. The weight of the weighted average is Gaussian negatively correlated with the difference between the gray value of the corresponding pixel and the gray value of the local neighborhood. Based on the photometric image and microtexture index map, a defect probability map is obtained that is positively correlated with the inversion value of the gray level of the photometric image and negatively correlated with the microtexture index. The defect probability graph is segmented into connected components to obtain suspected connected components; the penetration and diffusion confidence of each suspected connected component is obtained. , ; In the formula, Indicates the first The average defect probability value within a suspected connected domain; Indicates the first The number of edge contour pixels of a suspected connected region; On the microtexture index map, corresponding to the first... The gradient value of the microtexture index of the u-th edge contour pixel of a suspected connected domain; This indicates the microtexture index map corresponding to the first... A set of edge contour pixels of a suspected connected domain, and a set of corresponding microtexture indices; Represents the maximum value function; The sign representing the magnitude of a vector; Represents the normalization function; The value represents a small positive value. The penetration and diffusion confidence is positively correlated with the mean of the probability of defects within the corresponding connected region and the reciprocal of the gradient accumulation of the microtexture index along the edge of the connected region. In response to the penetration and diffusion confidence being greater than the first determination threshold, the corresponding suspected connected region is determined to be a real defect and located.

2. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The acquisition of the photometric image and the morphological image includes: A single scan of the silicate fireproof board was performed using an imaging system comprising a linear array camera and a dual-channel light source; channel A used a high-angle coaxial light source to acquire the photometric image; channel B used a low-angle grazing light source to acquire the morphological image.

3. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The standard deviation constant of the inherent texture noise of the board material is obtained, including: Select a defect-free normal board area, calculate the standard deviation of pixel grayscale within the normal board area, and use it as the standard deviation constant.

4. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The acquisition of the local window of the c-th pixel in the topography image includes: Centered on the c-th pixel of the topography image, obtain a neighborhood region of a preset size, which is denoted as the local window of the c-th pixel of the topography image.

5. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The defect probability satisfies the expression: ； In the formula, This represents the defect probability of the c-th pixel in the photometric image; This represents the grayscale value of the c-th pixel in the photometric image; This represents the microtexture index of the c-th pixel in the topography image; This represents the preset texture suppression coefficient; This represents the preset nonlinear adjustment factor.

6. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The step of segmenting the defect probability graph into suspected connected components includes: The defect probability map is thresholded using the Otsu method to obtain a binarized image; the foreground pixels in the binarized image are clustered using the 8-connected component labeling algorithm to obtain each suspected connected component.

7. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, This also includes classifying the actual defects into different types: True defects with a penetration and diffusion confidence level greater than or equal to the second judgment threshold are judged as serious defects and defined as deep adhesive spots. True defects with a penetration and diffusion confidence level less than the second judgment threshold but greater than the first judgment threshold are judged as minor defects and defined as shallow oil stains.

8. The method for detecting surface defects of silicate fireproof boards based on machine vision according to claim 1, characterized in that, The positioning includes: Obtain the center coordinates of the bounding rectangle of the actual defect, which will be used as the location of the actual defect.

Images