Multi-mode optical imaging fusion defect enhancement detection system and detection method

Abstract

The invention discloses a multi-mode optical imaging fusion defect enhancement detection system and detection method. The system comprises a high-angle red light mode emission light path, a low-angle yellow-green light mode emission light path, a detection mask, a two-dimensional displacement platform, a coaxial receiving light path and a control and processing module, the red light source irradiates to the collimating and dodging module, the collimating and dodging module collimates and dozes the red light source and then irradiates to the polarizer to generate linearly polarized light, and the linearly polarized light after collimating and dodging passes through the annular diaphragm and irradiates to the surface of the detection mask; the coaxial receiving light path receives reflected light on the detection mask, the band-pass optical filter and the polarization analyzer are switched through the control and processing module to form linear light spots in the same polarization state, the linear light spots are imaged to the target surface of the CCD camera through the telecentric light path to form high-contrast linear stripes, and dynamic adaptation is achieved by switching the polarization state of the polarization analyzer in a time-sharing mode. Therefore, pinholes and scratches are distinguished.

Description

Technical Field

[0001] This invention relates to the field of optical inspection technology, and more specifically, to a defect enhancement detection system and method based on multi-mode optical imaging fusion. Background Technology

[0002] Defects in semiconductor photomasks are a core concern in semiconductor manufacturing. Semiconductor photomask manufacturing serves as a bridge between the theoretical design of circuit patterns and wafer fabrication. Defects in semiconductor photomasks are directly transmitted to the wafer and amplified, directly impacting the precision of the processed chips. Furthermore, chips produced using photomasks cover a wide range of fields, from basic electronics to high-end intelligence; therefore, defects in semiconductor photomasks have a significant impact on chip applications. Thus, studying semiconductor photomask defects is crucial. The final quality inspection after photomask etching is the core of defect control, directly affecting photolithography yield. Final quality inspection primarily uses a combination of multi-angle manual visual inspection and automated scanning to ensure that the photomask meets the "zero fatal defects" standard, avoiding mass scrapping during wafer photolithography. However, it lacks the ability to effectively distinguish the morphological differences of submicron-level pinholes and scratches, increasing the difficulty of human judgment and leading to a higher misjudgment rate. Current technological bottlenecks are concentrated in two aspects: First, a single optical mode cannot simultaneously highlight the characteristic differences between pinholes and scratches. While dark-field illumination can enhance particle scattering signals, it is insufficient in responding to pinhole-penetrating light; bright-field detection can identify pinholes, but it is easily affected by surface reflections, leading to missed scratches. Second, the robustness of defect feature fusion algorithms is insufficient. Existing deep learning models suffer from decreased classification accuracy when fusing cross-modal data due to a lack of dynamic context awareness. For example, patent CN113834818A uses alternating transmitted and reflected light to simultaneously acquire defect information of the chromium layer and transparent substrate in a single frame scan, doubling the detection efficiency. However, this system still suffers from insufficient resolution in submicron-level defect detection and cannot be used to improve the resolution of distinguishing between pinholes and scratches. Patent CN120352354A proposes a multispectral collaborative detection system that uses a liquid crystal adjustable filter to switch bands and dynamically update the defect type-spectral mapping table, reducing the identification time of defect types and improving work efficiency, but it is limited by the light source switching speed. According to industry standards such as GB / T 16880-1997, the resolution for mask defect detection must reach 0.1μm, and the pass rate for zero fatal defects must be ≥99.99%. Traditional methods are no longer able to meet the requirements of the new generation of manufacturing processes. Therefore, there is an urgent need for an innovative detection method to solve the problem of identifying pinhole and scratch defects using the same system, while also taking into account mass production efficiency. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this invention proposes a multi-mode optical imaging fusion defect enhancement detection system and method.

[0004] The first aspect of the present invention provides a defect enhancement detection system based on multi-mode optical imaging fusion, comprising: a high-angle red light mode emission optical path, a low-angle yellow-green light mode emission optical path, a detection mask, a two-dimensional displacement platform, a coaxial receiving optical path, and a control and processing module;

[0005] The high-angle red light mode emission optical path includes a red light source, a collimation and homogenization module, a polarizer, and an annular aperture.

[0006] The red light source illuminates the collimation and homogenization module, which then collimates and homogenizes the red light source before illuminating the polarizer to generate linearly polarized light. The collimated and homogenized linearly polarized light passes through the annular aperture and illuminates the surface of the detection mask.

[0007] The low-angle yellow-green light mode emission optical path includes a yellow-green light source, a collimating and homogenizing module, a polarizer and a cylindrical mirror arranged in sequence.

[0008] The coaxial receiving optical path includes a bandpass filter, an analyzer, and a telecentric optical path. The coaxial receiving optical path receives reflected light from the detection mask. The control and processing module switches the bandpass filter and analyzer to form a linear light spot with the same polarization state. The linear light spot is imaged onto the target surface of the CCD camera through the telecentric optical path to form high-contrast linear stripes.

[0009] In a preferred embodiment of the present invention, the collimation and homogenization module includes a collimating lens group and a DOE homogenization module. The collimating lens group converts Gaussian light into flat-top ring light. The Gaussian energy distribution of the laser is corrected according to the phase modulation of the DOE homogenization module, so that the illumination intensity is uniform, thereby realizing laser collimation and outputting a uniform light source to uniformly illuminate the detection mask.

[0010] In a preferred embodiment of the present invention, the telecentric optical path includes three aspherical lenses and two spherical lenses arranged sequentially, wherein the three aspherical lenses are respectively denoted as aspherical lens L1, aspherical lens L2 and aspherical lens L3, and the two spherical lenses are respectively denoted as spherical lens L4 and spherical lens L5.

[0011] In a preferred embodiment of the present invention, the aspherical lens L1 has a conic coefficient of -1.2, which is used to effectively correct spherical aberration; the aspherical lens L2 has a conic coefficient of 0.8, and the higher-order coefficients are used to correct coma and astigmatism; the aspherical lens L3 has a conic coefficient of -0.5, and the higher-order coefficients are used to correct field curvature and distortion.

[0012] In a preferred embodiment of the present invention, the spherical lens L4 has a radius of 12 mm and a thickness of 3 mm, and is made of BK7 optical glass material with a refractive index of 1.52; the spherical lens L5 has a radius of -8 mm and a thickness of 2.5 mm, and is made of ZF6 optical glass material with a refractive index of 1.61.

[0013] In a preferred embodiment of the present invention, the CCD camera has a pixel size of 6.5μm×6.5μm and a pixel count of 2048×2048.

[0014] A second aspect of the present invention provides a defect enhancement detection method based on multi-mode optical imaging fusion, applicable to a defect enhancement detection system based on multi-mode optical imaging fusion, comprising the following steps:

[0015] S1: System preparation: Check the hardware status of red / yellow-green light source, filter, electromagnetic switching device of analyzer, two-dimensional displacement platform, and CCD camera. Initial settings: filter is set to 633nm red light, analyzer is set to orthogonal state, and two-dimensional displacement platform is set to zero.

[0016] S2: Pinhole detection: After the red light is collimated and homogenized, it forms a ring spot through the ring aperture and illuminates the detection mask at 45°. The two-dimensional displacement platform drives the detection mask to scan the whole world, and the CCD camera simultaneously acquires images.

[0017] S3: Scratch detection: Switch the filter to 550nm yellow-green light, set the analyzer to the same polarization state, and after the yellow-green light is collimated and homogenized, it forms a linear light spot through the cylindrical mirror. Illuminate the detection mask at 30°, scan the same path with the two-dimensional displacement platform, and acquire the image with the CCD camera. In this case, the filter filters red light, and the same polarization state enhances the linear stripes. The scratches in the image appear as high-contrast linear stripes.

[0018] S4: Image Processing and Output: Coarse registration is performed using two-dimensional displacement platform coordinates, and the algorithm refines and aligns the two images. The algorithm distinguishes defects based on shape features to obtain the detection results.

[0019] S5: End Reset: The light source is turned off, the filter and analyzer are reset, and the two-dimensional displacement platform is returned to zero.

[0020] In a preferred embodiment of the present invention, in step S2, the bandpass filter filters out stray light, and the orthogonal polarization enhances the contrast, so that the pinhole in the image appears as a circular bright spot.

[0021] In a preferred embodiment of the present invention, in step S4, the circular bright spot is a pinhole and the linear stripe is a scratch. By fusing the images and marking the pinholes in red and the scratches in blue, the detection results of the quantity, position and size are output.

[0022] The technical solution of the present invention has the following advantages compared with the prior art:

[0023] 1. This application employs a dual-modal detection optical path. The red light detection optical path for detecting pinholes utilizes the mirror reflection from the chromium plane of the mask to separate it from the field of view, while the diffuse reflection from the rough edge of the pinhole covers the field of view. This eliminates the need for complex filtering components, and geometric optics naturally separates the diffuse reflection signal of the pinhole from the reflected light from the chromium background of the mask. The yellow-green light detection optical path for detecting scratches utilizes the uniform reflection from the flat chromium surface of the mask and the brightness difference of the inclined surface at the scratch location to transform the linear characteristics of the scratch into identifiable stripes, physically distinguishing it from the circular characteristics of the pinhole.

[0024] 2. This application forms a closed-loop calibration system by integrating dual-emission optical paths and coaxial receiving optical paths. It outputs images of different modes by measuring the same area of ​​the mask through time-division detection, and compares the detection positions of pinholes and scratches through algorithms.

[0025] 3. This application uses defect detection algorithms and intelligent classification to accurately identify defects. It establishes a defect detection database based on the morphology of pinholes and scratches, matches defects in different states using deep forest and rule engine, and uses incremental learning algorithms to dynamically adapt to defect morphology iterations, thereby improving the robustness of the model. Attached Figure Description

[0026] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, some of the drawings in the following description are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0027] Figure 1 This is a structural diagram of the defect enhancement detection system based on multi-mode optical imaging fusion according to an embodiment of the present invention;

[0028] Figure 2 This is a high-angle (45°) red light mode emission optical path diagram according to an embodiment of the present invention;

[0029] Figure 3 This is a low-angle (30°) yellow-green light mode emission optical path diagram according to an embodiment of the present invention;

[0030] Figure 4 This is a schematic diagram of the coaxial optical receiving path according to an embodiment of the present invention;

[0031] Figure 5 This is a schematic diagram of the telecentric optical path imaging module according to an embodiment of the present invention;

[0032] Figure 6 This is a schematic diagram of the analysis process of the control and processing module in an embodiment of the present invention;

[0033] Figure 7This is a flowchart of the defect detection algorithm processing according to an embodiment of the present invention.

[0034] In the diagram: 1. High-angle red light mode emission optical path, 2. Low-angle yellow-green light mode emission optical path, 3. Detection mask, 4. Two-dimensional displacement platform, 5. Coaxial receiving optical path, 6. Control and processing module. Detailed Implementation

[0035] To better understand the above-mentioned objectives, features, and advantages of the present invention, the present invention will be further described in detail below with reference to specific embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in these embodiments can be combined with each other.

[0036] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and therefore the scope of protection of the invention is not limited to the specific embodiments disclosed below.

[0037] Example 1

[0038] See Figure 1 As shown, this invention proposes a multi-mode optical imaging fusion-based system for enhancing the detection of minute defects, such as... Figure 1As shown, this invention includes a high-angle (45°) red light mode emission optical path 1, a low-angle (30°) yellow-green light mode emission optical path 2, a detection mask 3, a two-dimensional displacement platform 4, a coaxial receiving optical path 5, and a control and processing module 6. The imaging module 1 (high-angle (45°) red light mode emission optical path) includes a red light source a that illuminates a collimation and homogenization module b. In module b, a parabolic mirror collimates and a DOE homogenizes the light, which then illuminates a 0° polarizer c to generate linearly polarized light. The collimated and homogenized linearly polarized light passes through an annular aperture d and illuminates the surface of the mask 3. Then, it passes through the two-dimensional displacement platform 4 for global defect imaging. The coaxial receiving optical path 5 receives diffusely reflected light from the mask, filters out the influence of yellow-green light through a bandpass filter i (633nm ± 5nm), and the light passes through an orthogonally placed analyzer j and a telecentric optical path k at the rear end to image onto the target surface of CCD 1. Secondly, the imaging process for detecting scratches involves module 2 (low-angle (30°) yellow-green light mode emission path). The yellow-green light source e illuminates the collimating and homogenizing module f. The parabolic mirror in f collimates the light, and the DOE homogenizes it, illuminating the 0° polarizer g to generate linearly polarized light. The collimated and homogenized linearly polarized light passes through the cylindrical mirror h, forming a linear spot along the scratch direction, which illuminates the surface of the mask 3. Then, the two-dimensional displacement platform 4 moves the detection mask 3, enabling a global defect scan of the detection mask 3. The coaxial receiving optical path 5 receives the reflected light from the mask. The control and processing module 6 switches the bandpass filter i to a yellow-green bandpass filter (550nm±5nm) to filter out the influence of red light. Simultaneously, module 6 switches the analyzer j to form a linear spot with the same polarization state. The image is then projected onto the target surface of the CCD 1 via the telecentric optical path k, forming high-contrast linear stripes that are physically distinguished from the pinhole circular bright spot.

[0039] The high-angle (45°) red light emitting optical path, such as Figure 2 As shown, the 660nm red laser a achieves global illumination of submicron pinholes.

[0040] Furthermore, the collimation and uniform light module b includes a collimating lens group and a DOE uniform light module, which converts Gaussian light into flat-top ring light. The phase modulation of the DOE corrects the Gaussian energy distribution of the laser, making the illumination intensity uniform, thereby achieving laser collimation and outputting a uniform light source to uniformly illuminate the mask. This avoids the state of pinhole illumination where the center is bright and the edges are dark, which could be misjudged as scratches, and reduces the impact of the light source on the pinhole detection accuracy.

[0041] Furthermore, the electrically tunable polarization controller c achieves precise control of the 0° linear polarization of the incident red light. Combined with the time-division switching of the analyzer in the receiving optical path, it improves the extinction ratio of pinhole detection and suppresses light leakage from the chromium layer background.

[0042] Furthermore, the annular aperture d (inner diameter 5mm, outer diameter 15mm) cuts the circular parallel red light into an annular beam, achieving a strict control of the angle between the edge fiber and the principal optical axis at 45°±0.5°. This causes the light reflected by the chromium-coated mirror on the mask (45° symmetrical) to deviate from the telecentric objective lens's field of view (±30°), achieving natural extinction. The central ray passing through the annular aperture is coaxial with the principal optical axis, ensuring that the principal optical axis component of the pinhole diffuse reflection is collected by the receiving light path, forming a "bright ring against a dark background."

[0043] The low-angle (30°) yellow-green light emitting optical path, such as Figure 3 As shown, the 575nm yellow-green laser e is sensitive to micron-level scratches on the surface and can more clearly capture edge details such as the sidewall tilt angle and bottom roughness of the scratch.

[0044] Furthermore, the collimation and uniform light module f includes a collimating lens group and a DOE uniform light module, which converts Gaussian light into flat-top ring light. The phase modulation of the DOE corrects the Gaussian energy distribution of the laser, making the illumination intensity uniform, thereby achieving laser collimation and outputting a uniform light source. This uniformly illuminates the mask, avoids the existence of bright and dark states in the scratch illumination, prevents the missed detection of minute scratches, and reduces the impact of the light source on the scratch detection accuracy.

[0045] Furthermore, the electrically tunable polarization controller g enables precise control of the 0° linear polarization of the yellow-green incident light. Combined with the time-division switching of the analyzer in the receiving optical path, it improves the stripe contrast of scratch detection and enhances the recognition of fine scratches.

[0046] Furthermore, the cylindrical mirror h (focal length 20mm, aspect ratio 4:1) compresses the circular light spot into a linear light spot (2mm long and 0.5mm wide along the direction of the scratch), concentrating the illumination light energy along the direction of the scratch. This results in bright stripes formed by strong reflected light on the light-facing side (where the scratch and linear illumination are in the same direction), and dark stripes formed by weak reflected light on the backlight side (where the scratch and linear illumination are in the opposite direction). This process physically distinguishes the two defects—the high-contrast linear stripes and the circular bright spot of the pinhole—from each other.

[0047] Receiving optical path as Figure 4 As shown, the bandpass filter i has two specifications. It can switch between a red bandpass filter (633nm±5nm) and a yellow-green bandpass filter (550nm±5nm) according to the time-division detection of different modes of light source. It only allows the wavelength of the current mode to pass through and filters out crosstalk light of other wavelengths to ensure the spectral purity of the image.

[0048] Furthermore, the analyzer comes in two specifications, allowing for time-division switching between a red light 90° orthogonal angle analyzer and a yellow-green light 0° same-angle analyzer, based on the time-division detection of different light sources. The 90° orthogonal polarization state of the red light mode filters the specularly reflected light from the chromium layer, allowing only the randomly polarized diffuse reflection light from the pinhole to pass through. The 0° same-polarization state of the yellow-green mode enhances the specular reflection light on the light-facing side of the scratch, improving stripe contrast.

[0049] Furthermore, to mitigate measurement errors caused by the motion of the two-dimensional displacement platform, a telecentric optical path is used for mask imaging, such as... Figure 5 As shown. A telecentric optical path requires a larger field of view (NA) to improve the collection efficiency of weak diffuse reflection light from pinholes. Setting the field of view of the telecentric optical path to ±30° allows for the collection of scattered light within the principal optical axis range, while reflecting light at 45° will not be received by the telecentric optical path, thus improving the signal-to-noise ratio. The corresponding telecentric optical path has a field of view of ±30°, NA = 1.0, and a working distance of 50mm. Simultaneously, the object-side telecentric design ensures consistent magnification for pinholes of different depths, avoiding misjudgment of size. The field of view of the telecentric optical path can also receive light reflected from scratches, improving detection accuracy.

[0050] Figure 5 The telecentric optical system k described herein has a telecentricity of <0.01°. The magnification of the telecentric system is adjusted according to the size requirements of the mask, and the distortion is controlled within the range of <0.1%. The change in distortion is simply calculated by using a 100lp / mm standard resolution plate to acquire images and then calculating it using formula (1).

[0051]

[0052] In the formula: D is the amount of distortion change; d max d min These represent the maximum and minimum pixel sizes, respectively; d avg This represents the average pixel size.

[0053] further, Figure 5The telecentric optical path system comprises, but is not limited to, the optical system configuration described below. The telecentric optical path integrates three aspherical lenses (L1-L3) and two spherical lenses (L4, L5) to form an apochromatic optical system. The three aspherical lenses have radii of curvature of -5mm, 8mm, and -3mm, respectively. The aspherical lenses effectively correct aberrations such as spherical aberration and coma present in traditional spherical lenses, enabling more precise focusing of light. Image quality is further improved by optimizing the aspherical coefficients. Specifically, the first aspherical lens has a conic coefficient of -1.2 to effectively correct spherical aberration; the second aspherical lens has a conic coefficient of 0.8, with higher-order coefficients used to correct coma and astigmatism; and the third aspherical lens has a conic coefficient of -0.5, with higher-order coefficients used to correct field curvature and distortion, ensuring image sharpness and accuracy across the entire field of view.

[0054] Two spherical lenses serve as auxiliary imaging elements to further optimize the optical path. The first spherical lens has a radius of 12mm and a thickness of 3mm. It is made of BK7 optical glass with a refractive index of 1.52, and its main component is borosilicate crown glass. It is used to initially converge light and balance some aberrations. The second spherical lens has a radius of -8mm and a thickness of 2.5mm. It is made of ZF6 optical glass with a refractive index of 1.61. Its main component is silicate glass with a certain proportion of lead oxide added. It is used to further correct aberrations and control the imaging position and magnification.

[0055] Furthermore, the telecentric optical path imaging module eliminates the parallax problem caused by changes in the object distance of the mask, ensuring that the imaging ratio of each point on the mask remains consistent. For pinholes and scratches at the same location on the mask, images under different light sources can be precisely aligned, facilitating subsequent comprehensive analysis and thus more accurately determining the characteristics and location of pinholes and scratches.

[0056] Furthermore, the CCD camera 1 receives time-division captured images of mask defects, forming multimodal image thumbnails. The CCD camera 1 has a pixel size of 6.5μm × 6.5μm and a pixel count of 2048 × 2048. Then, a defect detection algorithm is used to detect pinholes and scratches.

[0057] The control and processing module needs to complete four control and processing procedures, and the corresponding data processing flow is as follows: Figure 6 As shown.

[0058] 1. The electromagnetic actuators of the filter (633nm→550nm) and analyzer (orthogonal→same polarization) in the receiving optical path are controlled by a synchronous trigger signal. This process adopts a master-slave triggering mechanism, with the light source switching as the "master," and the filter and analyzer switching synchronously. Finally, the correctness of the imaging system is confirmed through dual-path optical path structure.

[0059] 2. The control module also needs to control the movement of the two-dimensional displacement platform, load the preset scanning path, and input parameters such as coordinates, speed, step size = CCD field width. It needs to ensure that the scanning path is consistent in both modes, so as to achieve pixel-level alignment of the image.

[0060] 3. The control and processing module needs to output a TTL trigger signal for each frame of the field of view moved by the two-dimensional displacement platform, and synchronously trigger CCD acquisition.

[0061] 4. After the measurement is completed, the control and processing module needs to be powered off and the positions of the filter and analyzer need to be reset electromagnetically to prepare for the next test.

[0062] like Figure 7 As shown, the entire process of the algorithm processing module proceeds step by step through "spatial alignment → difference detection → fusion labeling → result output".

[0063] First, the absolute coordinates (accuracy 0.1μm) of the encoder of the two-dimensional displacement platform are used to achieve coarse registration of the pinhole (red light mode) and scratch (yellow-green light mode) images. If there is a sub-pixel deviation (<0.5 pixels), the phase correlation algorithm is used to further refine the registration to ensure pixel-level alignment of the two modal images.

[0064] Then, for pinhole defects (red light mode), Otsu adaptive threshold segmentation was used to extract the bright spot region. Noise was filtered out using morphological opening operations, and the roundness of the region was analyzed using least squares circular fitting. The roundness (Y) calculation formula is as follows:

[0065]

[0066] In the formula: A is the area of ​​the pinhole bright spot region; L is the perimeter of the bright spot region.

[0067] Initially, roundness > 0.8 is defined as a pinhole, and the defect diameter and center coordinates are calculated. For scratch defects (yellow-green light mode), the Sobel edge enhancement operator is first used, with the enhancement direction perpendicular to the scratch to highlight linear texture. Then, Hough transform is used to extract straight line features, selecting stripes with a length > 10 pixels and an angle consistent with the cylindrical mirror's optical path direction. Finally, Gaussian fitting is used to analyze the stripe width. The Gaussian function G(x) formula is:

[0068]

[0069] In the formula: A is the amplitude; μ is the center position of the stripe; σ is the standard deviation (related to the width); B is the background gray value.

[0070] Stripe widths are extracted from the fitting results. Initially, a width less than 5 pixels is defined as a scratch, achieving dual verification of texture and shape.

[0071] Subsequently, under a unified two-dimensional displacement platform mechanical coordinate system, pinholes are marked as red circles and scratches as blue linear stripes, and a multimodal fusion image is generated through pseudo-color overlay. Finally, the fusion image, defect heat map, and parameters such as statistical quantity, size, and grayscale are visualized in real time, and the original data and defect parameters are stored in an SQL database, realizing the visualization, traceability, and standardized output of the inspection results.

[0072] The entire process utilizes synchronous parallel computation to accelerate algorithms such as threshold segmentation and Hough transform. It naturally distinguishes between two types of defects based on the morphological differences between circular and linear shapes, ensuring both processing efficiency and improving the accuracy of identifying minute defects. The two-dimensional displacement platform coordinates of mode 1 and mode 2 are extracted (e.g., *(x1,y1) corresponds to (x1,y1)*), and the images are directly matched according to the coordinates.

[0073] The second aspect of this invention provides a high-precision detection device and method for detecting mask defects in photolithography, comprising the following steps:

[0074] S1: System Preparation: Check the status of hardware such as the red / yellow-green light source, filter, electromagnetic switching device of the analyzer, 2D displacement platform, and CCD. Initial settings: filter is set to 633nm (red light), analyzer is set to orthogonal state, and 2D displacement platform is set to zero.

[0075] S2: Pinhole Detection (Red Light 45° Mode): After collimation and homogenization, red light passes through an annular aperture to form an annular spot, illuminating the mask at a 45° angle. A two-dimensional displacement platform drives a global scan of the mask, while a CCD simultaneously acquires images. A bandpass filter filters stray light, and orthogonal polarization enhances contrast. Pinholes appear as circular bright spots in the image.

[0076] S3: Scratch Detection (Yellow-Green Light 30° Mode): Switch the filter to 550nm (yellow-green light), and set the analyzer to the same polarization state. After collimation and homogenization, the yellow-green light forms a linear spot through a cylindrical mirror, illuminating the mask at a 30° angle. A two-dimensional displacement platform scans along the same path, and a CCD acquires the image. The filter filters out red light, and the same polarization state enhances the linear fringes. The scratches in the image appear as high-contrast linear fringes.

[0077] S4: Image Processing and Output: Coarse registration is performed using a 2D displacement platform coordinate system, followed by fine-tuning and alignment of the two images using an algorithm. The algorithm distinguishes defects based on shape features: circular bright spots represent pinholes, and linear stripes represent scratches. Finally, by fusing the images and labeling pinholes (red) and scratches (blue), the algorithm outputs detection results including quantity, location, and size.

[0078] S5: End Reset: The light source is turned off, the filter and analyzer are reset, and the two-dimensional displacement platform is returned to zero.

[0079] In summary, this application has the following beneficial effects:

[0080] 1) Time-division polarization: The presence of a pinhole exposes the quartz substrate of the mask, generating scattered light with randomly distributed polarization states. Scratches rely on the reflection of the chromium layer, producing reflected light with the same polarization state. Dynamic adaptation is achieved by switching the polarization state of the analyzer in a time-division manner, thereby distinguishing between pinholes and scratches;

[0081] 2) Wavelength Enhancement: Submicron-level pinholes can be identified based on diffraction effects, with 660nm red light exhibiting a significant diffraction effect. The core characteristic of scratches is a linear edge gradient, with 575nm yellow-green light being sensitive to micron-level surface scratches. Further improvements in the distinguishability between pinholes and scratches can be achieved by using different wavelength modes.

[0082] 3) Multi-angle incidence: To improve the diffuse reflection effect of pinholes, a 45° incidence angle is used to deflect the light from the chromium layer surface away from the principal optical axis of the receiving light path, increasing the diffuse reflection component containing the principal optical axis. To improve the recognition of scratches, a 30° incidence angle is used, where the angle between the light-facing sidewall and the incident light is small, resulting in strong reflected light and forming bright stripes, while the angle between the light-reflecting sidewall and the incident light is large, resulting in weak reflected light and forming dark stripes, thus improving recognition. By using different lighting angles, the distinction between pinholes and scratches is further improved.

[0083] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to the above embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0085] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A defect enhancement detection system based on multi-mode optical imaging fusion, comprising: The system comprises a high-angle red light mode emission optical path, a low-angle yellow-green light mode emission optical path, a detection mask, a two-dimensional displacement platform, a coaxial receiving optical path, and a control and processing module; its key feature is that... The high-angle red light mode emission optical path includes a red light source, a collimation and homogenization module, a polarizer, and an annular aperture. The red light source illuminates the collimation and homogenization module, which then collimates and homogenizes the red light source before illuminating the polarizer to generate linearly polarized light. The collimated and homogenized linearly polarized light passes through the annular aperture and illuminates the surface of the detection mask. The low-angle yellow-green light mode emission optical path includes a yellow-green light source, a collimation and homogenization module, a polarizer and a cylindrical mirror arranged in sequence. The coaxial receiving optical path includes a bandpass filter, an analyzer, and a telecentric optical path. The coaxial receiving optical path receives reflected light from the detection mask. The control and processing module switches the bandpass filter and analyzer to form a linear light spot with the same polarization state. The linear light spot is imaged onto the target surface of the CCD camera through the telecentric optical path to form high-contrast linear stripes.

2. The defect enhancement detection system based on multi-mode optical imaging fusion according to claim 1, characterized in that, The collimation and homogenization module includes a collimating lens group and a DOE homogenization module. The collimating lens group converts Gaussian light into flat-top ring light. The Gaussian energy distribution of the laser is corrected according to the phase modulation of the DOE homogenization module, so that the illumination intensity is uniform, realizing laser collimation and outputting a uniform light source to uniformly illuminate the detection mask.

3. The defect enhancement detection system based on multi-mode optical imaging fusion according to claim 2, characterized in that, The telecentric optical path includes three aspherical lenses and two spherical lenses arranged in sequence. The three aspherical lenses are denoted as aspherical lens L1, aspherical lens L2 and aspherical lens L3, and the two spherical lenses are denoted as spherical lens L4 and spherical lens L5.

4. The defect enhancement detection system based on multi-mode optical imaging fusion according to claim 1, characterized in that, Aspherical lens L1 has a conic coefficient of -1.2, which is used to effectively correct spherical aberration; aspherical lens L2 has a conic coefficient of 0.8mm, and higher-order coefficients are used to correct coma and astigmatism; aspherical lens L3 has a conic coefficient of -0.5mm, and higher-order coefficients are used to correct field curvature and distortion.

5. The defect enhancement detection system based on multi-mode optical imaging fusion according to claim 3, characterized in that, The spherical lens L4 has a radius of 12mm and a thickness of 3mm, and is made of BK7 optical glass with a refractive index of 1.52; the spherical lens L5 has a radius of -8mm and a thickness of 2.5mm, and is made of ZF6 optical glass with a refractive index of 1.

61.

6. The defect enhancement detection system based on multi-mode optical imaging fusion according to claim 1, characterized in that, The CCD camera has a pixel size of 6.5μm×6.5μm and a pixel count of 2048×2048.

7. A defect enhancement detection method based on multi-mode optical imaging fusion, applied to the defect enhancement detection system based on multi-mode optical imaging fusion as described in any one of claims 1-6, characterized in that, Including the following steps: S1: System preparation: Check the hardware status of red / yellow-green light source, filter, electromagnetic switching device of analyzer, two-dimensional displacement platform, and CCD camera. Initial settings: filter is set to 633nm red light, analyzer is set to orthogonal state, and two-dimensional displacement platform is set to zero. S2: Pinhole detection: After the red light is collimated and homogenized, it forms a ring spot through the ring aperture and illuminates the detection mask at 45°. The two-dimensional displacement platform drives the detection mask to scan the whole world, and the CCD camera simultaneously acquires images. S3: Scratch detection: Switch the filter to 550nm yellow-green light, set the analyzer to the same polarization state, and after the yellow-green light is collimated and homogenized, it forms a linear light spot through the cylindrical mirror. Illuminate the detection mask at 30°, scan the same path with the two-dimensional displacement platform, and acquire the image with the CCD camera. In this case, the filter filters red light, and the same polarization state enhances the linear stripes. The scratches in the image appear as high-contrast linear stripes. S4: Image Processing and Output: Coarse registration is performed using two-dimensional displacement platform coordinates, and the algorithm refines and aligns the two images. The algorithm distinguishes defects based on shape features to obtain the detection results. S5: End Reset: The light source is turned off, the filter and analyzer are reset, and the two-dimensional displacement platform is returned to zero.

8. The defect enhancement detection method based on multi-mode optical imaging fusion according to claim 7, characterized in that, In step S2, the bandpass filter filters out stray light, and the orthogonal polarization enhances the contrast, making the pinhole appear as a circular bright spot in the image.

9. The defect enhancement detection method based on multi-mode optical imaging fusion according to claim 8, characterized in that, In step S4, the circular bright spots are pinholes and the linear stripes are scratches. By fusing the images and marking the pinholes in red and the scratches in blue, the detection results of the quantity, location, and size are output.

Images