A clean board surface defect detection method and system

Abstract

The application relates to the technical field of surface defect detection, and discloses a clean plate surface defect detection method and system, wherein the method comprises the following steps: acquiring an original gray image, performing transformation decomposition based on guided filtering and local statistical enhancement; performing component fusion by adopting exponential inverse transformation and a difference enhancement method; generating a binary preliminary mask; extracting particle defect features and generating a thermal distribution map. Compared with the prior art which mainly depends on global high-pass filtering or fixed threshold segmentation, especially under the condition that the illumination condition is extremely easy to fluctuate, the surface exists complex texture such as wire drawing or embossing, and the like, and high-reliability particle contamination defect detection cannot be realized, since the application introduces guided filtering and local statistical enhancement, and combines multi-scale morphological enhancement and geometric feature screening, the influence of a complex texture background and uneven illumination can be effectively inhibited, and high-sensitivity and high-accuracy detection of tiny particle defects is realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of clean plate surface defect detection, and particularly relates to a clean plate surface defect detection method and system. BACKGROUND

[0002] At present, as an important wall and partition material in clean rooms, operating rooms, pharmaceutical workshops and other high-standard environments, clean plates have corrosion resistance, antibacterial, mildew resistance and other properties, and the detection and control requirements for surface defects are extremely strict. Traditional image detection methods are mostly based on global threshold segmentation, fixed edge detection operators (such as Sobel operator, Canny operator) or simple gray difference strategies. These methods mainly rely on the premise of global uniform illumination and simple background texture. However, in actual application, the clean plate surface often has complex texture design such as wire drawing and embossing to enhance mechanical strength or prevent bacterial adhesion, and the illumination conditions in the field environment are prone to strong fluctuations, such as multi-angle strong reflection, local shadow, mirror highlight and the like. These factors seriously interfere with the stability and accuracy of traditional image detection algorithms.

[0003] For example, when there are fine particle-like defects on the surface of the clean plate and the surrounding is accompanied by obvious texture fluctuations or high reflection spots, the traditional image detection method based on global threshold or single-scale filtering is easy to misidentify the background texture as a defect, resulting in a significant increase in false detection rate, or missing detection due to the suppression of defect signals. At the same time, some detection methods based on frequency domain filtering or fixed templates also have the problem of insufficient sensitivity to small particles, which cannot effectively extract the characteristics of small particle contaminants with small size and low gray scale contrast, limiting the precision of clean plate surface quality evaluation.

[0004] The prior art cannot fully meet the needs of high-precision image detection of particle micro-defects in complex texture background and dynamic lighting environment. Therefore, there is an urgent need for an image detection method that can still achieve high sensitivity, low false detection rate and high stability under complex surface texture, variable lighting and local reflection conditions, to significantly improve the reliability and practicality of clean plate surface defect detection, and meet the strict quality control requirements of clean environments. SUMMARY

[0005] In view of the above technical deficiencies, the purpose of the present application is to provide a clean plate surface defect detection method, which aims to solve the technical problem that the prior art mainly relies on global high-pass filtering or fixed threshold segmentation, especially under the condition that the illumination condition is prone to fluctuation and the surface has complex texture such as wire drawing or embossing in clean rooms, operating rooms or pharmaceutical workshops, and high-reliability particle contamination defect detection cannot be achieved.

[0006] To solve the above technical problems, the technical scheme adopted by the present application is as follows: the present application provides a clean plate surface defect detection method,

[0007] The clean plate surface defect detection method comprises:

[0008] Step S10: Obtain an original gray image of the clean plate surface, perform guided filtering and local statistical enhancement based transformation decomposition processing on the original gray image to obtain a background light component and a weak reflection detail component of the clean plate surface; perform component fusion processing on the background light component and the weak reflection detail component by using exponential inverse transformation and differential enhancement method to obtain reflection enhancement fusion data S;

[0009] Step S20: input the reflection enhancement fusion data S into an i-th circular structural element SE i with a radius of r i and a category number of n to perform pixel-by-pixel closed operation processing respectively to obtain an optimized enhancement image D;

[0010] Step S30: perform linear gray stretching on the optimized enhancement image D to obtain a gray stretched image D', calculate the mean value μ N and the standard deviation σ N in a local window size N×N, and generate an adaptive threshold T based on the mean value μ N and the standard deviation σ N , and generate a binary preliminary mask M including particle signal information according to the gray stretched image D' and the adaptive threshold T;

[0011] Step S40: input the binary preliminary mask M, first use a morphological opening operation method to remove isolated noise, then extract the area, perimeter, major axis and minor axis of the connected domain in the binary preliminary mask, calculate the circularity C and the aspect ratio R, and retain the area meeting the preset particle-like dirt size rule to obtain a binary optimized mask M f ;

[0012] Step S50: take the binary optimized mask M f as a final particle defect mask, and calculate the center of gravity coordinates, area, gray intensity, circularity and aspect ratio of each particle defect, and generate a corresponding thermal distribution map.

[0013] Preferably, in step S20, the step of inputting the reflection enhancement fusion data S into an i-th circular structural element SE i with a radius of r i and a category number of n to perform closed operation to obtain the optimized enhancement image D, specifically comprises: D i = B i -S, wherein B i is a background reference image of the i-th circular structural element SE i at a scale; and D i is the i-th circular structural element SEi a difference enhanced image at a scale of T = μ denotes a max operation over all scales; and denotes a morphological dilation operation for expanding high gray level regions of the image.

[0014] Preferably, in step S30, T = μ N +k·σ N wherein k is a sensitivity factor, and N is set according to the local particle distribution characteristics of the clean board; and

[0015] Preferably, in step S40, the step of performing the morphological opening operation on the binary preliminary mask to obtain the final mask, comprises: wherein A is an area and P is a perimeter; and the preset particle size rule is A > A min , C > 0.6, and R < 2, wherein A min is a preset minimum effective particle area threshold.

[0016] Preferably, in step S10, the step of performing guided filtering and local statistical enhancement based transform decomposition processing on the original gray image to obtain the background illumination component and the weak reflection detail component of the clean board surface, comprises:

[0017] First, performing local blur processing on the original gray image to generate a guide image, the guide image being used to simulate the smooth background illumination distribution of the clean board surface;

[0018] Then, within a preset local window, according to the local mean, variance and covariance relationship between the original gray image and the guide image, calculating a local linear fitting coefficient, and forming a first local illumination fitting result and a second local illumination fitting result;

[0019] Next, performing a first difference operation according to the first local illumination fitting result and the second local illumination fitting result to obtain a difference intermediate image that suppresses illumination changes, and performing local normalization enhancement processing on the difference intermediate image to amplify the particle defect signal, thereby obtaining the background illumination component and the weak reflection detail component of the clean board surface.

[0020] Preferably, in step S10, the step of performing component fusion processing on the background illumination component and the weak reflection detail component by using exponential inverse transform and difference enhancement method to obtain the reflection enhanced fusion data S, comprises: first performing exponential inverse transform processing on the weak reflection detail component, performing a second difference operation on the exponential inverse transformed image and the background illumination component, and then performing local dynamic adjustment on the result of the second difference operation according to a preset enhancement scale factor to generate the reflection enhanced fusion data S.

[0021] Preferably, in step S50, the generated corresponding thermal distribution map is provided for quality rating or local repair of the clean board by an automatic mechanical arm.

[0022] The application also provides a clean board surface defect detection system comprising:

[0023] An image preprocessing and enhancement module is configured to acquire an original gray image of the clean board surface, perform guided filter and local statistical enhancement based transform decomposition processing on the original gray image to obtain a background illumination component and a weak reflection detail component of the clean board surface, and perform component fusion processing on the background illumination component and the weak reflection detail component by using exponential inverse transform and differential enhancement method to obtain reflection enhancement fusion data S.

[0024] A multi-scale morphological optimization module is configured to perform pixel-by-pixel closed operation processing on the reflection enhancement fusion data S input to an i-th circular structural element SE with a radius of r i and a category number of n to obtain an optimized enhancement image D. i

[0025] A gray scale stretching and adaptive segmentation module is configured to perform linear gray scale stretching on the optimized enhancement image D to obtain a gray scale stretched image D', calculate a mean value μ N and a standard deviation σ N in a local window with a size of N×N, generate an adaptive threshold T based on the mean value μ N and the standard deviation σ N , and generate a binary preliminary mask M including particle signal information according to the gray scale stretched image D' and the adaptive threshold T.

[0026] A morphological screening and geometric feature extraction module is configured to take the binary preliminary mask M as input, remove isolated noise by using a morphological opening operation method, extract area, perimeter, major axis and minor axis of connected domains in the binary preliminary mask, calculate circularity C and aspect ratio R, and retain regions satisfying a preset particle-like dirt size rule to obtain a binary optimized mask M f .

[0027] A defect statistics and thermal analysis module is configured to take the binary optimized mask M f as a final particle defect mask, calculate the center of gravity coordinates, area, gray scale intensity, circularity and aspect ratio of each particle defect, and generate a corresponding thermal distribution map.

[0028] The application also provides a clean board surface defect detection device, which comprises a memory, a processor, and a clean board surface defect detection program stored in the memory and capable of running on the processor, and the clean board surface defect detection program is implemented when the processor is executed to realize the clean board surface defect detection method.

[0029] ​The application further provides a computer program product comprising a clean plate surface defect detection program, which, when executed by a processor, implements the clean plate surface defect detection method.

[0030] The application has the beneficial effect that by introducing multi-stage dynamic processing based on guided filtering and local statistical enhancement, the application effectively suppresses the interference of complex textures and uneven illumination on the clean plate surface on image detection, realizes high-sensitivity extraction of tiny particle defects, and significantly improves image detection accuracy and stability.

[0031] The application combines multi-scale morphological operation and geometric feature joint screening, can accurately distinguish real particle defects from background artifacts during image detection, reduces false detection rate and missed detection rate, and meets the stringent requirements of clean room, operating room and other high-standard environments for clean plate surface image detection. BRIEF DESCRIPTION OF DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0033] Figure 1 The flowchart of the first embodiment of the clean plate surface defect detection method of the application.

[0034] Figure 2 The corresponding thermal distribution diagram of the first embodiment of the clean plate surface defect detection method of the application.

[0035] Figure 3 The device diagram of the clean plate surface defect detection method of the application. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the application will be described clearly and completely below with reference to the drawings in the embodiments of the application. Obviously, the described embodiments are only some embodiments of the application, not all embodiments. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the application.

[0037] Embodiment one: as shown in the flowchart of the first embodiment of the clean plate surface defect detection method of the application, the first embodiment of the clean plate surface defect detection method of the application is proposed. Figure 1

[0038] ​In the first embodiment, the clean plate surface defect detection method comprises:

[0039] Step S10: Obtain the original gray image of the clean plate surface, perform guided filtering and local statistical enhancement based transform decomposition processing on the original gray image to obtain the background light component and the weak reflection detail component of the clean plate surface; and perform component fusion processing on the background light component and the weak reflection detail component by using exponential inverse transform and differential enhancement method to obtain reflection enhancement fusion data S.

[0040] It should be noted that the guided filtering and local statistical enhancement based transform decomposition processing refers to that the light change of the clean plate surface is dynamically fitted by taking the fuzzy guide map as a reference in the local window, the weak and small particle defect signal is enhanced by combining the local gray mean value, standard deviation and covariance information, and the whole includes two steps of guided filtering smoothing processing and local statistical contrast enhancement; the background light component of the clean plate surface refers to the smooth background gray image obtained by local fitting, which is used to represent the overall light intensity distribution of the clean plate under complex light conditions, and mainly includes the comprehensive light information of the surface reflection strong light area, the diffuse reflection area and the texture area; the weak reflection detail component refers to the detail information obtained after the background light component is removed from the original gray image, and mainly includes high-frequency weak signals such as particle defects, microscopic surface concave-convex or local contaminants; the exponential inverse transform and differential enhancement method refers to that after the logarithmic domain enhancement of the weak reflection detail component is completed, the exponential function is used to restore it to the linear gray domain, and the restored image and the background light component are again subjected to differential calculation to highlight the particle defect signal twice, and the whole includes the exponential inverse transform step and the secondary differential strengthening step; the reflection enhancement fusion data S refers to the image finally obtained after the above multiple enhancement and fusion, which is used to accurately represent the gray difference between the particle defects and the complex background on the clean plate surface, and mainly includes the particle defect significant area and the clear background information after removing most of the background texture interference.

[0041] It can be understood that, compared with the traditional fixed threshold or global filtering method, the guided filtering and local statistical enhancement processing in the present application can not only finely estimate the complex light components, but also adaptively adjust the contrast in each local window, reduce the false detection and missed detection phenomenon caused by background texture fluctuation, so as to ensure that the particle defects are extracted with high fidelity in image detection. The exponential inverse transform and differential enhancement method can effectively amplify the weak reflection detail signal through dynamic switching between the multiple gray domains and the logarithmic domain, and further improve the gray separation degree of the particle defects relative to the background. This method is suitable for the clean plate surface where multiple light sources, strong reflection and complex texture exist at the same time, and significantly improves the overall image detection accuracy and robustness.

[0042] Step S20: input the reflection enhancement fusion data S into a plurality of n kinds of filters with a radius of r ithe i-th circular structural element SE i respectively, to obtain the optimized enhanced image D;

[0043] It should be noted that the reflection enhancement fusion data S is input to the n kinds of circular structural elements with a radius of r i the i-th circular structural element SE i respectively, to obtain the optimized enhanced image D.

[0044] It can be understood that the traditional method usually only uses a single fixed scale morphological operation, which cannot simultaneously adapt to different size particle defects and complex texture backgrounds of various scales on the clean plate surface; and the present application can simultaneously retain multi-size particle defect signals through the design of multi-scale (n different radii) closing operation, reduce the false detection rate caused by the complexity of the surface texture, and improve the integrity of the overall particle detection. The maximum difference strategy in multi-scale processing can fuse the multiple difference images obtained under different structural elements into an optimized enhanced image D to cover more real particle defect information and eliminate texture noise to the greatest extent, so that the subsequent binarization and morphological screening steps are more stable and reliable.

[0045] For example, in a clean plate sample, there are particle defects with diameters of 0.3 mm and 0.8 mm, and there are wire drawing texture (width about 1 mm) and weak embossing texture (width about 2 mm) on the surface; small particles can be detected but large particle signals are suppressed by texture, while the present application can use multi-scale circular structural elements with radii of 5px and 12px respectively, and after maximum difference fusion, the signal of small particle defects is completely retained, and the outline of large particle defects is also strengthened, so that the overall detection rate of particle defects is improved, and the image detection accuracy is significantly improved.

[0046] Step S30: performing linear gray stretch on the optimized enhanced image D to obtain a gray stretched image D', calculating the mean μ N and the standard deviation σ N in a local window size N×N, and generating an adaptive threshold T based on the mean μ N and the standard deviation σ N , and generating a binary preliminary mask M including particle signal information according to the gray stretched image D' and the adaptive threshold T;

[0047] It should be noted that in step S30, T = μ N +k·σN wherein k is a sensitivity factor, and N is set corresponding to the local particle distribution characteristics of the cleaned board; a binary preliminary mask

[0048] It should be understood that, compared with the traditional fixed threshold or global Otsu algorithm, the adaptive threshold in step S30 combines local statistical characteristics, which can automatically adjust the threshold according to the gray fluctuation of each small area, improve the detection sensitivity of weak particle defects in different brightness areas and complex texture areas, and significantly reduce false detection and missed detection caused by uneven illumination or background texture. Compared with the traditional fixed threshold or global Otsu algorithm, the adaptive threshold combines local statistical characteristics, which can automatically adjust the threshold according to the gray fluctuation of each small area, improve the detection sensitivity of weak particle defects in different brightness areas and complex texture areas, and significantly reduce false detection and missed detection caused by uneven illumination or background texture.

[0049] Step S40: Taking the binary preliminary mask M as input, first using morphological opening operation to remove isolated noise, then extracting the area, perimeter, major axis and minor axis of the connected domain in the binary preliminary mask, calculating the roundness C and aspect ratio R, and retaining the area that meets the preset particle size rule of the dirt, to obtain the binary optimized mask M f ;

[0050] It should be noted that taking the binary preliminary mask as input, first using morphological opening operation to remove isolated noise, means that first using a small size structure element to perform corrosion operation on the mask to remove random isolated points, and then through inflation operation to restore the actual particle shape, so as to reduce the false detection area caused by accidental light spots, dust, mirror highlights and other factors; then extracting the area, perimeter, major axis and minor axis information of each connected domain in the binary preliminary mask, and calculating the roundness and aspect ratio, minor axis major axis and other geometric characteristics, retaining the area that meets the preset particle size and shape rule of the dirt, and finally generating the binary optimized mask, to ensure that only the real particle defect area is retained.

[0051] It can be understood that, compared with the traditional method of only relying on area filtering or fixed threshold morphological processing, step S40 comprehensively considers the multi-dimensional geometric feature information such as area, perimeter, roundness and aspect ratio, so that the algorithm can more accurately separate the particle defects and surface texture artifacts, especially suitable for stable detection of particle defects in complex texture background.

[0052] It should be understood that the threshold values of the geometric feature parameters (area, circularity, aspect ratio) can be flexibly set according to the actual defect morphology characteristics of the clean plate; for example, the circularity C can be set to be greater than 0.6 to exclude non-circular texture artifacts, the area threshold value can be set according to the minimum physical size of the particles (such as ≥0.2 mm2), and the aspect ratio R is controlled within a certain range to exclude strip-shaped background interference. This multi-feature comprehensive discrimination strategy improves the accuracy and robustness of particle defect image detection.

[0053] Step S50: obtaining a binary optimization mask M f As the final particle defect mask, the center of gravity coordinates, area, gray intensity, circularity and aspect ratio of each particle defect are counted, and a corresponding heat distribution map is generated.

[0054] It should be noted that taking the binary optimization mask as the final particle defect mask, counting the center of gravity coordinates, area, gray intensity, circularity and aspect ratio of each particle defect means that the defect regions obtained after multiple optimization and screening are fully quantified in terms of geometric and gray features, so as to finely represent the distribution characteristics and shape features of the particle defects. The generated heat distribution map is used for visual display of the defect position and distribution density, helping to quickly locate the key defect region in the subsequent clean plate quality control or local repair process. The generation of the corresponding heat distribution map adopts a kernel density estimation method based on defect center of gravity coordinate density weighting, taking the center of gravity position of each particle defect as the weighted kernel center, mapping the color heat according to the area or gray intensity, and finally forming a two-dimensional or pseudo-color distribution map, which directly reflects the overall defect hot area and local high-density aggregation.

[0055] It can be understood that the traditional detection method generally only outputs a simple binary image or a mark image, and lacks intuitive presentation of the particle defect distribution density, spatial aggregation and severity. The heat distribution map generated by the method of the present application not only helps the quality inspection personnel to quickly judge the overall cleanliness, but also assists in finding local high-risk areas in the production or use process, facilitating subsequent targeted maintenance and traceability analysis. By introducing the area, gray intensity and other parameters of the particle defects into the generation process of the kernel density estimation heat distribution map, the present application realizes the weight differentiation of the defect severity, not only reflects the spatial distribution of the position, but also reflects the comprehensive degree of the influence of the defect on the cleanliness, so that the image detection result has more engineering application value and diagnostic significance.

[0056] For example, in a clean plate test sample with a surface area of 0.5 m 2 , a total of 120 particle defects are detected, and the traditional binary image can only show the position; while in the heat distribution map generated by the present application, the center region is displayed as a red high-density aggregation hot area, and the average area of the corresponding defects is 0.4 mm 2, the peripheral area is displayed as a green low-density area, corresponding to a defect area less than 0.2mm2, and the gray intensity is lower than 140. The thermal distribution map helps the quality inspection personnel to quickly identify the process pollution problem in the center processing area, accurately determine the subsequent cleaning or repair range, and significantly improve the detection and processing efficiency.

[0057] Embodiment two: In addition, the present application provides a clean plate surface defect detection system, which adopts the clean plate surface defect detection method in the above embodiment, and can solve the technical problem of clean plate surface defect detection. Compared with the prior art, the beneficial effects of the clean plate surface defect detection system provided by the present application are the same as those of the clean plate surface defect detection method provided by the above embodiment, and the other technical features of the clean plate surface defect detection system are the same as those disclosed in the above embodiment method, which will not be repeated here.

[0058] Embodiment three: The present application provides a clean plate surface defect detection device, please refer to Figure 3A clean board surface defect detection device includes at least one processor and a memory connected to the at least one processor in communication. The memory stores instructions executable by the at least one processor to enable the at least one processor to perform a clean board surface defect detection method according to an embodiment. The clean board surface defect detection device according to an embodiment can include, but is not limited to, a mobile terminal such as a mobile phone, a notebook, a digital broadcasting receiver, a PDA (Personal Digital Assistant), a PAD (Portable Application Description), a PMP (Portable Media Player), a car terminal (e.g., a car navigation terminal), and the like, and a stationary terminal such as a digital TV, a desktop computer, and the like. The clean board surface defect detection device is only an example and should not impose any limitation on the function and use range of the clean board surface defect detection device according to an embodiment. The clean board surface defect detection device can include a processing device 1001 that can perform various appropriate actions and processes according to programs stored in a read-only memory 1002 or loaded into a random access memory 1004 from a storage device 1003. The random access memory 1004 also stores various programs and data required for the operation of the clean board surface defect detection device. The processing device 1001, the read-only memory 1002, and the random access memory 1004 are connected to each other through a bus 1005. An I / O interface 1006 is also connected to the bus. In general, the following systems can be connected to the I / O interface 1006: input devices 1007 including, for example, a touch screen, a touch pad, a keyboard, a mouse, an image sensor, a microphone, an accelerometer, a gyroscope, and the like; output devices 1008 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, and the like; the storage device 1003; and a communication device 1009. The communication device 1009 can allow the clean board surface defect detection device to communicate with other devices wirelessly or wired to exchange data. Although the clean board surface defect detection device having various systems is illustrated in the drawing, it is understood that all of the illustrated systems are not required to be implemented or provided. More or less systems can be alternatively implemented or provided.

[0059] Embodiment four: the application further provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of a clean plate surface defect detection method as described above. The computer program product provided by the application can solve the technical problem of clean plate surface defect detection. Compared with the prior art, the beneficial effects of the computer program product provided by the application are the same as those of the clean plate surface defect detection method provided by the above-mentioned embodiment, and are not described here.

[0060] In particular, according to the embodiments disclosed by the application, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, the embodiments disclosed by the application include a computer program product comprising a computer program carried on a computer readable medium, the computer program comprising program code for executing the method shown in the flowchart. In such embodiments, the computer program can be downloaded and installed from a network through a communication device, or installed from a storage device 1003, or installed from a ROM 1002. When the computer program is executed by a processing device 1001, the above-mentioned functions defined in the method of the embodiments disclosed by the application are executed.

[0061] It should be understood that various parts of the application disclosed can be realized with hardware, software, firmware or a combination thereof. In the description of the above-mentioned embodiments, specific features, structures, materials or characteristics can be combined in any one or more embodiments or examples in a suitable manner.

[0062] Obviously, those skilled in the art can make various modifications and variations to the application without departing from the spirit and scope of the application. Thus, if these modifications and variations of the application fall within the scope of the claims of the application and their equivalent technologies, the application also intends to include these modifications and variations.

Claims

1. A method of detecting surface defects of a clean wafer, characterized by, The method comprises: Step S10: obtaining an original gray image of the clean plate surface, performing guided filter and local statistical enhancement based transform decomposition processing on the original gray image to obtain a background light component and a weak reflection detail component of the clean plate surface; An exponential inverse transform and differential enhancement method is used to perform component fusion processing on the background light component and the weak reflection detail component to obtain reflection enhancement fusion data S. The background light component is comprehensive illumination information including a surface reflection strong light region, a diffuse reflection region and a texture region, and is used to represent the overall light intensity distribution of the clean plate under complex illumination conditions. The weak reflection detail component is detail information obtained after the background light component is removed from the original gray image, including a particle defect signal, a microscopic surface concave-convex signal or a local contaminant signal. Step S20: Input the reflection enhancement fusion data S into a region with n species and a radius of... The i-th circular structuring element Perform pixel-by-pixel closing operations to obtain the optimized and enhanced image D; Step S30: Perform linear grayscale stretching on the optimized and enhanced image D to obtain a grayscale stretched image. In local window size Internal calculation mean and standard deviation and based on the mean and standard deviation Generate adaptive threshold Stretching the image based on grayscale and adaptive threshold Generate a binary preliminary mask including particle signal information. ; Step S40: using a binary preliminary mask As input, first remove isolated noise using a morphological opening operation, then extract the area, perimeter, major axis and minor axis of the connected domain in the binary preliminary mask, calculate the circularity and aspect ratio , and retain the area that meets the preset size rule of granular dirt to obtain a binary optimized mask ; Step S50: binarizing the optimized mask As the final particle defect mask, the barycentric coordinates, area, gray intensity, circularity and aspect ratio of each particle defect are counted and the corresponding thermal distribution map is generated.

2. A method of detecting surface defects on a clean plate as recited in claim 1, wherein, In step S20, the reflection-enhanced fusion data S is input to the i-th circular structural element with radius of n kinds respectively to obtain the steps of the optimized enhancement image D, specifically including: , , ; wherein, is the background reference image at the scale of the i-th circular structural element ; is the differential enhancement image at the scale of the i-th circular structural element , used to represent the particle signal suppressed by the background; represents the maximum value operation in all scale ranges; is a morphological dilation operation, used to expand the high gray level area of the image; is a morphological erosion operation, used to reduce the high gray level area of the image.

3. The method for detecting surface defects of a cleanroom plate as described in claim 1, characterized in that, In step S30, wherein, is a sensitivity factor, The settings of correspond to the local particle distribution characteristics of the clean board; the binary preliminary mask .

4. The method for detecting surface defects of a cleanroom panel as described in claim 1, characterized in that, In step S40, , wherein A is the area and P is the perimeter; the preset size rule of the granular dirt is: wherein, is the preset minimum effective particle area threshold.

5. The method for detecting surface defects of a cleanroom panel as described in claim 1, characterized in that, In step S10, the exponential inverse transform and differential enhancement method is used to perform component fusion processing on the background light component and the weak reflection detail component to obtain the reflection enhancement fusion data S, which specifically comprises: first performing exponential inverse transform processing on the weak reflection detail component, performing a second differential operation on the exponentially inverse transformed image and the background light component, and then performing local dynamic adjustment on the result of the second differential operation according to a preset enhancement scale factor to generate the reflection enhancement fusion data S.

6. The method for detecting surface defects of a cleanroom panel as described in claim 1, characterized in that, In step S50, the generated corresponding thermal distribution map is provided for quality rating or clean plate local repair automatic mechanical arm.

7. A clean sheet surface defect detection system for use in a clean sheet surface defect detection method according to any one of claims 1 to 6, characterized by, The clean plate surface defect detection system comprises: An image preprocessing and enhancement module is configured to obtain an original gray image of the clean plate surface, perform guided filter and local statistical enhancement based transform decomposition processing on the original gray image to obtain a background light component and a weak reflection detail component of the clean plate surface, and perform component fusion processing on the background light component and the weak reflection detail component using an exponential inverse transform and differential enhancement method to obtain reflection enhancement fusion data S. The background light component is comprehensive illumination information including a surface reflection strong light region, a diffuse reflection region and a texture region, and is used to represent the overall light intensity distribution of the clean plate under complex illumination conditions. The weak reflection detail component is detail information obtained after the background light component is removed from the original gray image, including a particle defect signal, a microscopic surface concave-convex signal or a local contaminant signal. a multi-scale morphological optimization module, configured to input the reflection enhancement fusion data S into an i-th circular structural element with a radius of and a category number of n respectively to perform a pixel-by-pixel close operation processing to obtain an optimized enhancement image D; The gray scale stretching and adaptive segmentation module is configured to perform linear gray scale stretching on the optimized enhanced image D to obtain a gray scale stretched image , in a local window size , to calculate a mean value and a standard deviation , and generate an adaptive threshold value based on the mean value and the standard deviation , to generate a binary preliminary mask including particle signal information from the gray scale stretched image and the adaptive threshold value ; The morphological screening and geometric feature extraction module is used to perform initial binary masking. As input, a morphological open-ended algorithm is first used to remove isolated noise. Then, the area, perimeter, major axis, and minor axis of the connected components in the binary preliminary mask are extracted, and the roundness is calculated. and aspect ratio And retain the area that meets the preset particulate dirt size rules to obtain a binary optimized mask. ; A defect statistics and thermal analysis module for binary optimization masks As the final particle defect mask, the barycenter coordinates, area, gray intensity, circularity and aspect ratio of each particle defect are counted and the corresponding thermal distribution map is generated.

8. A clean board surface defect detection apparatus characterized by comprising: The clean plate surface defect detection device comprises a memory, a processor and a clean plate surface defect detection program stored on the memory and executable on the processor, and the clean plate surface defect detection program implements the clean plate surface defect detection method of any one of claims 1 to 6 when executed by the processor.

9. A computer program product, characterised in that, The computer program product comprises a clean plate surface defect detection program, and the clean plate surface defect detection program implements the clean plate surface defect detection method of any one of claims 1 to 6 when executed by the processor.

Images