Surface defect detection device and method based on polarized optical dark field imaging

Abstract

The invention discloses a surface defect detection device and method based on polarized optical dark field imaging, and belongs to the technical field of industrial defect imaging detection, the device comprises a polarization camera, an imaging lens, a polarization annular LED light source and a sample to be detected which are coaxially arranged; wherein the polarization camera and the imaging lens form a polarization imaging system, the polarization annular LED light source emits linearly polarized light with an inclination angle, and the linearly polarized light is reflected and scattered on a sample to be detected; the linearly polarized light enters the surface of a to-be-detected sample, a small amount of reflection and most of scattered light of the to-be-detected sample are collected by the imaging lens and the polarization camera through a circular hole in the center of the polarization annular LED light source, and finally imaging is achieved. The defect depth can be calculated according to the defect width and the defect inclination angle, so that the defect depth is inversed by the low-frequency linear polarization degree, and the problem that the defect depth information cannot be effectively extracted in the prior art is solved.

Description

Technical Field

[0001] The present invention belongs to the technical field of industrial defect imaging detection, and in particular relates to a surface defect detection device and method based on polarization optical dark field imaging. Background Art

[0002] In the fields of industrial manufacturing and product quality inspection, the detection of sample surface defects (such as scratches and dents) is crucial to ensuring product performance. For materials such as metals, the depth of the defect directly affects key indicators such as mechanical strength and corrosion resistance. Therefore, the depth of the defect is an important basis for determining product qualification. Currently, surface defect detection technologies based on optical imaging (such as traditional dark-field imaging and asymmetric light illumination difference methods) can quickly locate the defect position and produce clear images. However, their imaging mechanism, which relies on light intensity contrast, cannot effectively extract the depth information of the defect, nor can it distinguish between defects of different severity. The resulting image only provides the "planar" distribution of the scratches. The lack of depth information seriously restricts the quantitative analysis and grading of defects. On the other hand, some technologies that can quantify depth (such as light sectioning, white light interferometry and confocal microscopy) obtain the three-dimensional morphology of the sample surface through optical tomography or point-by-point scanning. However, these methods require complex optical system design, precision mechanical control or high-cost instruments and equipment, resulting in large and complex equipment, high maintenance costs, weak system robustness and low detection efficiency, making it difficult to adapt to the high-speed, large-scale and low-cost detection requirements of industrial assembly lines.

[0003] Therefore, how to achieve rapid and highly sensitive analysis of surface defect depth through optical imaging without the need for complex optical components and precision scanning has become a technical challenge that needs to be urgently solved in the field of industrial inspection. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A surface defect detection device based on polarization optical dark field imaging, characterized by comprising: a coaxially arranged polarization camera, an imaging lens, a polarized annular LED light source, and a sample to be detected;

[0006] Among them, the polarization camera and the imaging lens constitute a polarization imaging system, the polarization ring LED light source emits linearly polarized light with a fixed inclination angle, and is obliquely incident on the surface of the sample to be detected in a reflective dark field illumination mode; the linearly polarized light is incident on the surface of the sample to be detected, and a small amount of mirror reflected light and most of the scattered light of the sample to be detected are collected by the imaging lens and the polarization camera through the circular hole in the center of the polarization ring LED light source, forming a polarization image for defect depth inversion; the polarization camera supports single exposure and synchronous acquisition of images in four polarization directions for calculating the low-frequency linear polarization degree map; the surface defect detection device based on polarization optical dark field imaging is used to invert the defect depth based on the low-frequency linear polarization degree map and the defect geometric relationship.

[0007] A surface defect detection method based on polarization optical dark field imaging, used in the surface defect detection device based on polarization optical dark field imaging, comprising:

[0008] Step 1, calculating the linear polarization degree image, specifically comprising: placing the surface defect detection device based on polarization optical dark field imaging coaxially, adjusting the polarized circular LED light source to an appropriate brightness, and synchronously capturing original polarization images in four polarization directions using a polarization camera; calculating the Stokes vector from the original polarization images in the four polarization directions, and generating a low-frequency linear polarization degree image from the Stokes vector calculation;

[0009] Step 2: performing three-dimensional block matching noise reduction filtering on the low-frequency linear polarization degree image obtained in step 1 to obtain a noise-reduced low-frequency linear polarization degree image;

[0010] Step 3, calibrating the linear equation of the defect section inclination angle and the low-frequency linear polarization degree according to the low-frequency linear polarization degree image after noise reduction obtained in step 2;

[0011] Step 4, performing depth inversion on the defect according to the linear equation of the defect section inclination angle and the low-frequency linear polarization degree calculated in step 3;

[0012] Step 5: Perform zero-frequency filtering on the original polarization images of the four polarization directions obtained in step 1 and calculate the high-frequency linear polarization degree image after noise reduction;

[0013] Step 6: Perform weighted fusion on the low-frequency linear polarization degree map obtained after noise reduction in step 2 and the high-frequency linear polarization degree map obtained in step 5.

[0014] The present invention has the following beneficial effects:

[0015] The present invention proposes a method for inverting depth using low-frequency linear polarization. The present invention can calculate the low-frequency linear polarization map of the sample using the original polarization map taken by a polarization camera. Compared with the ordinary intensity map, the low-frequency linear polarization map improves the contrast difference between the defect and the background. In the low-frequency linear polarization map, the width of the defect can be obtained by direct image measurement, and the inclination angle of the section at the defect (such as a scratch) has a strong linear correlation with the low-frequency linear polarization of the defect. By calibrating this linear relationship, the low-frequency linear polarization can be used to solve the inclination angle of the defect section. According to the specific geometric relationship of various defects (such as V-shaped scratches), the present invention can calculate the defect depth from the defect width and the defect inclination, thereby realizing the inversion of the defect depth using low-frequency linear polarization, solving the problem that the existing technology cannot effectively extract defect depth information.

[0016] In addition to directly generating a low-frequency polarization degree map based on the original polarization map, the present invention enhances the contrast of the defect scattering signal by performing zero-frequency filtering on the original polarization map before calculating the linear polarization degree. Furthermore, a three-dimensional block matching noise reduction algorithm is applied to effectively suppress background noise and improve detection sensitivity. During the detection process, the high-frequency linear polarization degree map after zero-frequency filtering is fused with the low-frequency polarization degree map by adjusting a weighting parameter to adjust the visualization level of different defect levels. This method enables flexible adaptation to multiple detection standards.

[0017] The present invention directly connects the imaging lens to the polarization camera, and places them coaxially with the polarized annular LED light source and the sample to be detected in a straight line. The structure is simple, and the problem of complex system structure in traditional methods is solved.

[0018] The polarized circular LED light source used in this invention is an industrial-grade LED, the imaging lens can be an industrial lens, and the polarization camera can be a commercial polarization camera. No consumables or expensive imaging lenses are required, resulting in a low overall device cost. The algorithm module is embedded in the device system, requiring only routine cleaning of the equipment during testing. This invention is a low-cost detection device suitable for industrial use, resolving the high cost, difficult maintenance, and limited industrialization challenges of traditional methods.

[0019] The present invention provides reflective dark-field imaging conditions by illuminating the sample at a large angle through a polarized annular LED light source. The light source inclination angle and polarization direction are fixed and do not need to be adjusted again, thereby realizing a measurement device that does not require optical calibration and is highly robust.

[0020] The imaging lens of the present invention can be a lens with a reduced magnification, which can realize large-field imaging of samples and can be combined with a high-frame-rate polarization camera to achieve high-speed, real-time detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1This is a schematic diagram of the structure of the surface defect detection device based on polarized optical dark field imaging of the present invention, wherein 1-polarized camera, 2-imaging lens, 3-polarized annular LED light source, 4-sample to be detected;

[0022] Figure 2 The calibration and measurement results of the present invention are shown in Figure 1. (a) shows the scatter plot and linear fitting results of the scratch section inclination angle and low-frequency linear polarization degree, (b) shows the distribution histogram of the low-frequency linear polarization degree of scratches with different section inclination angles, and (c) shows the comparison between the inverted scratch depth value and the measured depth value.

[0023] Figure 3 The low-frequency linear polarization maps of the surfaces of good and defective capacitors and the statistical scatter plots of their scratch depth inversion values ​​are shown in Figure 1. (a) shows the low-frequency linear polarization map of the surface of a defective capacitor with a deep scratch, (b) shows the low-frequency linear polarization map of the surface of a good capacitor with a shallow scratch, and (c) shows the statistical scatter plots of the scratch depth inversion values ​​of several good and defective capacitors calculated using the low-frequency linear polarization inversion depth method.

[0024] Figure 4 These are the linear polarization degree diagrams of the same capacitor sample under different filtering conditions; among them, (a) is the low-frequency linear polarization degree diagram of the same capacitor sample, (b) is the linear polarization degree diagram of the high-frequency and low-frequency weighted fusion of the same capacitor sample, and (c) is the high-frequency linear polarization degree diagram of the same capacitor sample. DETAILED DESCRIPTION

[0025] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only intended to illustrate the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.

[0026] The present invention provides a surface defect detection device based on polarization optical dark field imaging, such as Figure 1 As shown, it is used to detect defects on the surface of highly reflective samples and includes: a coaxially arranged polarization camera 1, an imaging lens 2, a polarized circular LED light source 3, and a sample to be inspected 4. The polarization camera 1 and imaging lens 2 form a polarization imaging system. The polarized circular LED light source 3 emits linearly polarized light with a certain inclination angle, which is reflected and scattered by the sample to be inspected 4.

[0027] The principle of the technical solution of the present invention is as follows: a polarized annular LED light source 3 with an inclination emits linearly polarized light, which is incident on the surface of the sample 4 to be tested at a certain angle. A small amount of reflected light and most of the scattered light from the sample 4 to be tested are collected by the imaging lens 2 and the polarization camera 1 through the circular hole in the center of the polarized annular LED light source 3, and finally an image is formed.

[0028] Specifically, the light emitted by the polarized annular LED light source 3 is similar to a cone-shaped light, which can realize dark field illumination at a large angle, and the light is reflected and scattered on the plane to be tested of the sample 4 to be tested.

[0029] The polarized annular LED light source 3 is an LED light source with a certain inclination angle. The light it emits is linearly polarized light and is incident on the surface of the sample to be detected 4 at a certain angle. Its large-angle incidence provides dark field lighting conditions. The incident light is reflected and scattered on the surface of the sample to be detected 4. A small amount of reflected light and most of the scattered light pass through the circular hole in the center of the polarized annular LED light source 3, and are collected by the polarization camera 1 after passing through the imaging lens 2. The inclination angle, light source wavelength and polarization direction of the polarized annular LED light source 3 are not restricted. The tilted white light polarized light used in the present invention when detecting the surface of the capacitor can provide better dark field lighting conditions while ensuring sufficient light intensity. During imaging, the scattered light on the surface of the sample to be detected 4 is mainly collected. If the sample to be detected 4 is of other types, the polarized annular LED light source 3 can be replaced with other inclination angles or wavelengths depending on the specific imaging effect.

[0030] The imaging lens 2 can completely image the sample to be detected 4 on the detector of the polarization camera 1. The imaging lens 2 with different magnifications can be replaced according to the size of the sample to be detected 4 and the detection requirements, so as to achieve complete imaging of the morphology of the sample to be detected 4 within the field of view and avoid edge distortion. By selecting a lens with a suitable magnification, multiple samples to be detected 4 can be detected simultaneously in one image, further improving the detection efficiency.

[0031] The polarization camera 1 has an exposure time of less than 1ms and supports simultaneous capture of polarization images at four polarization directions, such as 0°, 45°, 90°, and 135°, in a single exposure. This eliminates the time difference error caused by traditional mechanically rotating polarizers. If a conventional non-polarization camera is used in place of the polarization camera 1, a polarization module, such as a motor-driven linear polarizer assembly or a liquid crystal polarization modulator, can be added between the imaging lens 2 and the polarized circular LED light source 3 to achieve the same defect detection results.

[0032] The sample 4 to be tested is not limited to the capacitance sample in the embodiment of the present invention, and may also be replaced by other samples with high reflective properties.

[0033] Furthermore, the present invention provides a surface defect detection method based on polarized optical dark field imaging, comprising:

[0034] Step 1, calculating the linear polarization degree image, specifically comprising: the device of the present invention according to Figure 1 Place the polarized annular LED light source 3 coaxially, adjust the brightness to the appropriate level, and use the polarization camera 1 to simultaneously capture the original polarization images in the four polarization directions. Calculate the Stokes vectors from these original polarization images, and then generate a low-frequency linear polarization degree map from the Stokes vectors.

[0035] Step 2: performing three-dimensional block matching noise reduction filtering on the low-frequency linear polarization degree map obtained in step 1. Specifically, the method includes: first, performing block matching on the low-frequency linear polarization degree map obtained in step 1, and stacking similar image blocks into a three-dimensional matrix; then performing collaborative filtering and threshold processing in the transform domain to eliminate noise components; and finally, reconstructing the image through inverse transformation to generate a noise-reduced low-frequency linear polarization degree map.

[0036] Step 3, calibrating the linear equation of the defect section inclination angle and the low-frequency linear polarization degree according to the low-frequency linear polarization degree map after noise reduction obtained in step 2, specifically including: performing a three-dimensional morphology scan (such as a step meter scan) on the actual sample 4 to be tested to obtain the section inclination angle at each defect, and extracting the linear polarization degree value corresponding to the defect position from the low-frequency linear polarization degree map after noise reduction obtained in step 2, using the section inclination angle at the above defect and the low-frequency linear polarization degree value at the defect position to draw a scatter plot, and performing linear fitting on all data points to obtain the linear equation of the defect section inclination angle and the low-frequency linear polarization degree of the sample 4 to be tested.

[0037] Step 4: Defect depth inversion is performed based on the linear equation calculated in step 3. Specifically, the defect width and low-frequency linear polarization degree are extracted from the de-noised low-frequency linear polarization map obtained in step 2. Substituting the low-frequency linear polarization degree into the linear equation in step 3 yields the inclination angle of the defect cross-section. Based on the geometric relationship of the defect cross-section profile (e.g., a V-shaped scratch), the defect depth can be inferred from the known inclination angle and width. This method is known as the low-frequency linear polarization depth inversion method.

[0038] Step 5: Zero-frequency filtering is performed on the original polarization maps for the four polarization directions obtained in step 1, and a de-noised high-frequency linear polarization degree map is calculated. This specifically includes: Fourier transforming the original polarization maps for the four polarization directions obtained in step 1, filtering out the zero-frequency component, and then performing inverse Fourier transforms on each of the four polarization directions. Stokes vectors are then calculated as described in step 1, and the high-frequency linear polarization degree map is calculated. Three-dimensional block matching noise reduction filtering is then performed as described in step 2, ultimately resulting in a de-noised high-frequency linear polarization degree map. This high-frequency linear polarization degree map can significantly enhance the contrast between defects and a uniform background.

[0039] Step 6: Perform weighted fusion on the low-frequency linear polarization degree map obtained after noise reduction in step 2 and the high-frequency linear polarization degree map obtained in step 5. Specifically, the weighted fusion of the low-frequency linear polarization degree map obtained after noise reduction in step 2 and the high-frequency linear polarization degree map obtained in step 5 is adjusted according to the detection requirements. This can adjust the degree of visualization of defects of different levels and achieve flexible adaptation to multiple detection standards.

[0040] The present invention uses low-frequency and high-frequency linear polarization imaging to detect scratches of varying depths on capacitor samples, and uses low-frequency linear polarization to invert scratch depth. Furthermore, the present invention also tested scratches on several good and defective capacitors on a production line, statistically analyzing the scratch depth differences between good and defective capacitors. High- and low-frequency weighted fusion testing was then performed on several samples. This demonstrates that the present invention can effectively extract defect depth information and display defects of varying severity in a graded manner, enabling rapid detection of good and defective products on a production line.

[0041] Reference Figure 2 The calibration and measurement result diagram of the invention is shown. Different scratches on the same sample are imaged and the low-frequency linear polarization degree is calculated using the method of the invention. The geometric profiles of these scratches are detected by a step profiler to obtain the section inclination angle. Figure 2 (a) is a scatter plot of the scratch surface inclination angle and the low-frequency linear polarization degree and the linear fitting result, where the straight line is the fitting line, the points are the scratch sampling points, and the linear correlation R between the scratch surface inclination angle (as shown on the horizontal axis) and the low-frequency linear polarization degree (as shown on the vertical axis) is shown. 2 It is 0.945, that is, there is a good linear relationship between the low-frequency linear polarization degree and the scratch surface inclination angle, and this straight line can be used as the calibration result. Figure 2 The illustration in (a) is the step profiler scanning result of one of the scratches. The horizontal axis is the scanning length and the vertical axis is the scratch depth. The scratch is approximately V-shaped. Figure 2 (b) The histogram of the low-frequency linear polarization distribution of scratches with the inclination angles of the cut plane being 28°, 32°, 36°, 41°, 46°, 50°, 54°, 58°, and 62°, as shown in the figure. Figure 2 As shown in (b). Wherein, the horizontal axis is the low-frequency linear polarization degree, and the vertical axis is the pixel count. It can be obtained that the scratch section inclination angle a that can be distinguished by the method of the present invention is 4-5°. It can be seen that the low-frequency linear polarization degree is linearly related to the scratch section inclination angle, the scratch section is approximately V-shaped, and its section inclination angle is related to the scratch width and scratch depth, wherein the scratch width can be obtained by measuring the pixel length on the low-frequency linear polarization degree image, and the scratch depth can be calculated according to the geometric relationship of the scratch morphology. This is the low-frequency linear polarization degree inversion depth method. The low-frequency linear polarization degree inversion depth method is used to measure six unknown scratches, and the scratch depth value inverted by the low-frequency linear polarization degree is compared with the scratch depth value measured by the step meter, as shown Figure 2 (c) shows the scratch count on the horizontal axis and the scratch depth on the vertical axis. Circles represent measured depths, and squares represent inverted depths. The maximum relative error between the inverted and measured depths is 0.08. This demonstrates that the method of the present invention can quickly determine scratch depths within a certain accuracy range.

[0042] Reference Figure 3 As shown, Figure 3 (a) Figure 3 (b) takes the surface low-frequency linear polarization diagrams of a defective capacitor and a good capacitor as examples, respectively, to demonstrate the detection effect after low-frequency linear polarization calculation and three-dimensional block matching noise reduction in the present invention, wherein the color mapping bar on the right represents the value of linear polarization. It can be seen that scratches may appear on the surface of both good and defective capacitors, but the scratch depth of the good capacitor is lower than the threshold, which will not affect the product performance and cause potential harm. The scratch depths on the surfaces of several good and defective capacitors are calculated and counted by the low-frequency linear polarization inversion depth method, as shown in the figure. Figure 3 Figure (c) shows the scratch count on the horizontal axis and the scratch depth on the vertical axis. Circles represent good products, and squares represent defective products. The scratch depth on defective products is significantly higher than that on good products, with a clear boundary between the two. This result demonstrates that the present invention can effectively detect scratch depth on highly reflective flat surfaces, enabling rapid product quality assessment and classification in a production line environment.

[0043] Reference Figure 4 As shown in FIG, this is a linear polarization diagram of the same capacitor sample under different filtering conditions. Figure 4 (a) Figure 4 (b) Figure 4 (c) are the low-frequency linear polarization degree diagram, the high-frequency and low-frequency weighted fusion linear polarization degree diagram, and the high-frequency linear polarization degree diagram of the same capacitor sample, and these results have all been processed by three-dimensional block matching noise reduction. The color mapping bar on the right represents the value of the linear polarization degree. By comparison Figure 4 Three pictures of Figure 4 (c) can highlight finer defect details, Figure 4 (a) can only reflect the larger and deeper defects on the sample surface. Figure 4 (b) is between the two. By adjusting the proportion of high and low frequencies during fusion, linear polarization diagrams with different levels of detail can be presented to match the detection of different quality control standards.

[0044] The present invention uses the original polarization image captured by the polarization camera 1 to calculate the sample's low-frequency linear polarization map. Compared to a conventional intensity map, this low-frequency linear polarization map improves the contrast difference between the defect and the background. In this low-frequency linear polarization map, the width of the defect can be directly measured from the image. The inclination angle of the section at the defect (such as a scratch) has a strong linear correlation with the defect's low-frequency linear polarization. By calibrating this linear relationship, the low-frequency linear polarization can be used to solve the defect section inclination angle. Based on the specific geometric relationships of various defects (such as V-shaped scratches), the defect depth can be calculated from the defect width and defect inclination, thereby achieving defect depth inversion using low-frequency linear polarization. The present invention solves the problem that existing technologies cannot effectively extract defect depth information.

[0045] In addition to directly generating a low-frequency polarization degree map based on the original polarization map, the present invention enhances the contrast of the defect scattering signal by performing zero-frequency filtering on the original polarization map before calculating the linear polarization degree. Furthermore, a three-dimensional block matching noise reduction algorithm is applied to effectively suppress background noise and improve detection sensitivity. During the detection process, the high-frequency linear polarization degree map after zero-frequency filtering is fused with the low-frequency polarization degree map by adjusting a weighting parameter to adjust the visualization level of different defect levels. This method enables flexible adaptation to multiple detection standards.

[0046] The present invention directly connects the imaging lens 2 to the polarization camera 1, and places them coaxially with the polarization annular LED light source 3 and the sample to be detected 4 in a straight line. The structure is simple, and the problem of complex system structure in traditional methods is solved.

[0047] The polarized annular LED light source 3 used in the present invention is an industrial-grade LED, the imaging lens 2 can be an industrial lens, and the polarization camera 1 can be a commercial polarization camera. Without consumables or expensive imaging lenses, the overall device cost is low. The algorithm module of the present invention is embedded in the device system, requiring only routine cleaning of the equipment during testing. This low-cost detection device is suitable for industrial use, resolving the high cost, maintenance, and industrialization challenges of traditional methods. The present invention provides reflective dark-field imaging conditions by illuminating the sample at a wide angle using a polarized annular LED light source. The light source's inclination angle and polarization direction are fixed, requiring no further adjustment. This results in a robust measurement device that requires no optical calibration.

[0048] The imaging lens 2 of the present invention can be a lens with a reduced magnification, which can realize large-field imaging of the sample, and can be combined with a high-frame-rate polarization camera to achieve high-speed and real-time detection.

[0049] The above descriptions are merely embodiments of the present invention and are not intended to limit the scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied to other related system fields, are also included in the scope of protection of the present invention.

[0050] The contents not described in detail in the specification of the present invention belong to the prior art known to those skilled in the art.

Claims

1. A surface defect detection device based on polarization optical dark field imaging, characterized in that: include: Coaxially arranged polarization camera, imaging lens, polarized circular LED light source and sample to be tested; Among them, the polarization camera and the imaging lens constitute a polarization imaging system, the polarization ring LED light source emits linearly polarized light with a fixed inclination angle, and is obliquely incident on the surface of the sample to be detected in a reflective dark field illumination mode; the linearly polarized light is incident on the surface of the sample to be detected, and a small amount of mirror reflected light and most of the scattered light of the sample to be detected are collected by the imaging lens and the polarization camera through the circular hole in the center of the polarization ring LED light source, forming a polarization image for defect depth inversion; the polarization camera supports single exposure and synchronous acquisition of images in four polarization directions for calculating the low-frequency linear polarization degree map; the surface defect detection device based on polarization optical dark field imaging is used to invert the defect depth based on the low-frequency linear polarization degree map and the defect geometric relationship.

2. The surface defect detection device based on polarization optical dark field imaging according to claim 1, characterized in that: The imaging lens completely images the sample to be detected on the detector of the polarization camera; imaging lenses with different magnifications are replaced according to the size of the sample to be detected and the detection requirements, and the morphology of the sample to be detected is completely imaged within the field of view.

3. The surface defect detection device based on polarization optical dark field imaging according to claim 1, characterized in that: The exposure time of the polarization camera is within 1ms, and it supports synchronously capturing polarization images in four polarization directions with a single exposure.

4. The surface defect detection device based on polarization optical dark field imaging according to claim 1, characterized in that: The polarization camera is replaced by an ordinary non-polarization camera, and a polarization module is added between the imaging lens and the polarized circular LED light source.

5. A surface defect detection method based on polarization optical dark field imaging, used in the surface defect detection device based on polarization optical dark field imaging according to any one of claims 1 to 4, characterized in that: include: Step 1, calculating the linear polarization degree image, specifically comprising: placing the surface defect detection device based on polarization optical dark field imaging coaxially, adjusting the polarized circular LED light source to an appropriate brightness, and synchronously capturing original polarization images in four polarization directions using a polarization camera; calculating the Stokes vector from the original polarization images in the four polarization directions, and generating a low-frequency linear polarization degree image from the Stokes vector calculation; Step 2: performing three-dimensional block matching noise reduction filtering on the low-frequency linear polarization degree image obtained in step 1 to obtain a noise-reduced low-frequency linear polarization degree image; Step 3, calibrating the linear equation of the defect section inclination angle and the low-frequency linear polarization degree according to the low-frequency linear polarization degree image after noise reduction obtained in step 2; Step 4, performing depth inversion on the defect according to the linear equation of the defect section inclination angle and the low-frequency linear polarization degree calculated in step 3; Step 5: Perform zero-frequency filtering on the original polarization images of the four polarization directions obtained in step 1 and calculate the high-frequency linear polarization degree image after noise reduction; Step 6: Perform weighted fusion on the low-frequency linear polarization degree map obtained after noise reduction in step 2 and the high-frequency linear polarization degree map obtained in step 5.

6. The surface defect detection method based on polarization optical dark field imaging according to claim 5, characterized in that: Step 2 includes: first, performing block matching on the low-frequency linear polarization degree map obtained in step 1, stacking similar image blocks into a three-dimensional matrix; then performing collaborative filtering and threshold processing in the transform domain to eliminate noise components; and finally, reconstructing the image through inverse transformation to generate a denoised low-frequency linear polarization degree map.

7. The surface defect detection method based on polarization optical dark field imaging according to claim 5, characterized in that: Step 3 includes: performing a three-dimensional morphological scan on the actual sample to be tested to obtain the section inclination angle at each defect, and extracting the linear polarization value corresponding to the defect position from the noise-reduced low-frequency linear polarization degree map obtained in step 2, using the section inclination angle at the defect and the low-frequency linear polarization degree value at the defect position to draw a scatter plot, and performing linear fitting on all data points to obtain a linear equation of the defect section inclination angle and the low-frequency linear polarization degree of the actual sample to be tested.

8. The surface defect detection method based on polarization optical dark field imaging according to claim 5, characterized in that: Step 4 includes: extracting the width and low-frequency linear polarization degree of the defect based on the denoised low-frequency linear polarization degree map obtained in step 2, substituting the low-frequency linear polarization degree into the linear equation of the defect section inclination angle and the low-frequency linear polarization degree in step 3 to obtain the defect section inclination angle; based on the geometric relationship of the defect section profile, the defect depth is calculated based on the known defect section inclination angle and the defect width, thereby realizing depth inversion of the defect.

9. The surface defect detection method based on polarization optical dark field imaging according to claim 5, characterized in that: Step 5 includes: performing Fourier transform on the original polarization images of the four polarization directions obtained in step 1, filtering out the zero-frequency component, and then performing inverse Fourier transform on each of them, calculating the Stokes vector as described in step 1, generating a high-frequency linear polarization degree map based on the Stokes vector calculation, and performing the three-dimensional block matching noise reduction filtering described in step 2 to finally obtain a noise-reduced high-frequency linear polarization degree map.

10. The surface defect detection method based on polarization optical dark field imaging according to claim 5, characterized in that: Step 6 includes: adjusting the weighted ratio parameter according to the detection requirements to perform weighted fusion on the low-frequency linear polarization degree map after noise reduction obtained in step 2 and the high-frequency linear polarization degree map after noise reduction obtained in step 5.

Images