A method and system for surface defect identification of SRP test samples

Abstract

The application discloses a kind of surface defect identification method and system for SRP test sample, specifically relates to the field of semiconductor sample surface detection, including obtaining the SRP test sample of completion cleaning and drying processing, under the same imaging magnification and the same illumination condition, to the surface to be examined Partial image is collected frame by frame in the way of half field of view transverse movement, half field of view vertical line change, and position alignment and image splicing are carried out according to the coincidence area of adjacent partial image, and dry state image is output;The application continuously tracks and identifies the boundary evolution in the process of local rewetting and redrying of surface abnormal area, to solve the problem that SRP test sample surface cleaning residual anomaly and real surface defect are difficult to reliably distinguish in the absence of pattern wafer defect detection scene.

Description

Technical Field

[0001] This invention relates to the field of semiconductor sample surface inspection technology, and more specifically, to a method and system for identifying surface defects in SRP test samples. Background Technology

[0002] In the surface inspection of SRP test samples related to the detection of defects in patternless wafers, the focus of existing technologies is mainly on screening out surface anomalies as early as possible and reducing the range of false detections. In engineering, after the sample has been sliced, thinned, ground, polished, cleaned and dried, its surface is subjected to bright field, dark field or combined imaging, and then scratches, pits, cracks, spots and adhesion anomalies are identified based on grayscale changes, boundary contours, local textures and connected region features.

[0003] In the actual sample flow process before SRP testing, the sample has often already undergone wet cleaning and entered the inspection state. It is difficult to allow for long-term static or strong secondary processing in the subsequent testing cycle. At this time, in addition to the real defects, the surface may also have residual liquid retraction marks, film-like boundaries left after drying, and local brightness and darkness anomalies caused by loose particles. The same anomaly appears as a complete boundary in the current dry image and does not seem to be significantly different from the actual defects at first glance. However, after slight changes in the surface state, the boundary shrinks, local breaks, position shifts, or even disappears completely. This indicates that the anomaly does not originate from actual damage to the sample surface, but from cleaning residue. Since the existing identification process mainly relies on the appearance information in a single dry image to make a judgment, it lacks the basis for distinguishing whether such anomalies will be reconstructed with changes in the surface state. Therefore, it is easy to continuously identify changeable residual traces as real surface defects.

[0004] Therefore, the difficulty in reliably distinguishing between surface cleaning residue abnormalities and actual surface defects in SRP test samples under patternless wafer defect detection scenarios has become an urgent problem to be solved. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for identifying surface defects in SRP test samples. By continuously tracking and identifying the boundary evolution of abnormal surface regions during local rewetting and redrying processes, the method solves the problem of reliably distinguishing between surface cleaning residue abnormalities and real surface defects in SRP test samples under non-patterned wafer defect detection scenarios.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for identifying surface defects in SRP test samples, comprising:

[0007] S1. Obtain the SRP test sample after cleaning and drying. Under the same imaging magnification and the same lighting conditions, acquire local images of the surface to be tested by moving horizontally in half of the field of view and vertically changing lines in half of the field of view. Then, align the positions and stitch the images according to the overlapping areas of adjacent local images to output the dry image.

[0008] S2. Extract each surface anomaly region based on the dry image. Select a continuous and complete surface outside each surface anomaly region to form a closed band. Establish sampling directions point by point along the closed band toward the interior of the surface anomaly region. Record the first intersection point of each sampling direction with the surface anomaly boundary and output the boundary record.

[0009] S3. Cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and continuously image the corresponding area from the formation of the liquid film to its disappearance, and output the drying sequence.

[0010] S4. For each frame of the image in the drying sequence, search again for the first intersection point of the current surface anomaly boundary along each sampling direction in the boundary record, generate the intersection point displacement record, intersection point disappearance record and region connectivity record corresponding to each sampling direction, and output the evolution record.

[0011] S5. Based on the evolution record, identify the surface abnormal areas in the intersection displacement record that have displacement toward the interior of the surface abnormal area, the surface abnormal areas in the intersection disappearance record that have intersection disappearance, or the surface abnormal areas in the region connectivity record that have changed from a single region to multiple regions as cleaning residual abnormalities. Identify the remaining surface abnormal areas as real surface defects and output the surface defect identification results.

[0012] In a preferred embodiment, S1 includes:

[0013] S1-1. Under the same imaging magnification and the same illumination conditions, the surface to be inspected is acquired frame by frame in the same row by moving horizontally in half of the field of view. The overlapping area of ​​each two adjacent local images is extracted, the horizontal displacement and vertical displacement of the corresponding position in the overlapping area are calculated, and the displacement record in the row is output.

[0014] S1-2. After the local image of the same row is acquired, the local image of the surface to be inspected is acquired in the manner of vertical line wrapping in half the field of view. The overlapping area of ​​the corresponding local images of the two adjacent rows is extracted, the horizontal displacement and vertical displacement of the corresponding position in the overlapping area are calculated, and the inter-row displacement record is output.

[0015] S1-3. Based on the inline displacement record and the interline displacement record, determine the corresponding position of each local image on the surface to be inspected, and perform position alignment and image stitching according to the corresponding position of each local image to output a dry image.

[0016] In a preferred embodiment, S2 includes:

[0017] S2-1. Extract each surface anomaly region based on the dry image, and determine the surface anomaly boundary point by point along the outer edge of each surface anomaly region, and output the surface anomaly boundary.

[0018] S2-2. Starting from the surface anomaly boundary, search for a continuous and complete surface layer by layer outward from the surface anomaly region. Extract the closed outer boundary at a position with a set number of layers between it and the surface anomaly boundary. Determine the annular region between the surface anomaly boundary and the closed outer boundary as the closed zone and output the closed zone.

[0019] In a preferred embodiment, S2 further includes:

[0020] S2-3. Select sampling points along the closed band point by point, and connect the nearest point on the surface anomaly boundary for each sampling point to generate a sampling direction from the closed band to the inside of the surface anomaly area, and output the sampling direction.

[0021] S2-4. Search point by point along each sampling direction from the closed zone toward the interior of the surface anomaly region for the first intersection position with the surface anomaly boundary, and record each first intersection position as a boundary record.

[0022] In a preferred embodiment, S3 includes:

[0023] S3-1. For each surface abnormality area and its corresponding closed zone, determine the local area covering the surface abnormality area and the closed zone, apply rewetting liquid to the local area to form a local rewetting liquid film, and output the local rewetting liquid film.

[0024] S3-2. During the formation of the local rewetting liquid film, images of the local area are acquired frame by frame, the boundary of the local rewetting liquid film in each frame is extracted, and the time of liquid film formation is determined based on the change of the local rewetting liquid film boundary in two adjacent frames.

[0025] In a preferred embodiment, S3 further includes:

[0026] S3-3. Starting from the moment the liquid film forms, continuously acquire images of the local area, extract the local rewetting liquid film boundary in each frame of the image, and output the liquid film boundary sequence.

[0027] S3-4. Based on the liquid film boundary sequence, stop acquiring data when the local rewetting liquid film boundary disappears, and arrange the images acquired between the liquid film formation time and the time when acquisition stops in chronological order to output the drying sequence.

[0028] In a preferred embodiment, S4 includes:

[0029] S4-1. For each frame of image and each sampling direction in the boundary record in the drying sequence, search for boundary change points point by point from the closed zone toward the interior of the surface abnormal area along the sampling direction, and simultaneously extract the corresponding intersection points in the boundary record, the corresponding intersection points in the previous frame image, and the corresponding intersection points in the adjacent sampling directions, and generate candidate points in the current direction, mapping points in the previous frame, and adjacent bounding points, and output candidate records.

[0030] S4-2. For each sampling direction in the candidate record, first calculate the distance between the current candidate point and the previous frame mapping point, and extract the candidate points whose distances are at the lower limit of the previous frame direction. Then, determine whether the candidate points whose distances are at the lower limit of the previous frame direction fall between the adjacent bounding points of the adjacent sampling direction. When the candidate points whose distances are at the lower limit of the current frame direction fall between the adjacent bounding points, the candidate points whose distances are at the lower limit of the previous frame direction are determined as the current intersection point. When the candidate points whose distances are at the lower limit of the current frame direction do not fall between the adjacent bounding points and there are candidate points between the adjacent bounding points, the candidate point closest to the closed band between the adjacent bounding points is determined as the current intersection point. Otherwise, write the sampling direction into the record to be inspected, and output the current intersection point record and the record to be inspected.

[0031] In a preferred embodiment, S4 further includes:

[0032] S4-3. For each sampling direction in the record to be inspected, extract the candidate points of the corresponding sampling direction in the next frame of the current frame image, and map the current intersection point determined in the next frame image back to the current frame image along the corresponding sampling direction to form a return point. When there is a candidate point between the return point and the previous frame mapping point, the candidate point is determined as the current intersection point. When there is no candidate point between the return point and the previous frame mapping point, the sampling direction is written into the intersection disappearance record. When there are multiple candidate points between the return point and the previous frame mapping point, the candidate point that is on the same side as the intersection point of the adjacent sampling direction is determined as the current intersection point. Otherwise, the sampling direction is written into the intersection disappearance record. Output the verification intersection record and the intersection disappearance record.

[0033] S4-4. For each surface anomaly region in each frame of the image, summarize the intersection displacement distance between the current intersection record and the corresponding intersection position in the verification intersection record and the boundary record in the sampling direction order to generate an intersection displacement record. Determine the boundary gap according to the distribution of consecutive adjacent sampling directions in the intersection disappearance record. Then cross-validate the connectivity status of the surface anomaly region in the current frame of the image with the boundary gap. Write a region preservation mark when the boundary gap corresponds to a single connected region, write a region split mark when the boundary gap corresponds to multiple separated connected regions, otherwise write a region merge mark. Merge the region preservation mark, region split mark and region merge mark to generate a region connectivity record. Merge the intersection displacement record, intersection disappearance record and region connectivity record, and output the evolution record.

[0034] In a preferred embodiment, S5 includes:

[0035] S5-1. For each surface anomalous region in the evolution record, according to the time sequence of the drying sequence, count the number of sampling directions of displacement towards the interior of the surface anomalous region in the intersection displacement record of each frame image, the number of intersection disappearance directions in the intersection disappearance record, and the region splitting markers in the region connectivity record, and output the region change record.

[0036] S5-2. For each surface abnormality area, when the number of sampling directions of displacement towards the interior of the surface abnormality area increases continuously between adjacent frame images, and the number of intersection disappearance directions in subsequent frame images is greater than zero or a region splitting mark appears, the surface abnormality area is identified as a cleaning residue abnormality; otherwise, it is identified as a real surface defect.

[0037] S5-3. Summarize the cleaning residue abnormalities and actual surface defects corresponding to each abnormal surface area, and output the surface defect identification results.

[0038] In a preferred embodiment, a surface defect identification system for SRP test samples includes:

[0039] The image acquisition module is used to acquire SRP test samples that have undergone cleaning and drying. Under the same imaging magnification and the same lighting conditions, it acquires local images of the surface under test by moving horizontally in half of the field of view and vertically wrapping in half of the field of view. It then aligns the positions of adjacent local images and stitches them together to output a dry image.

[0040] The boundary construction module extracts various surface anomalous regions based on dry images, selects continuous and complete surfaces outside each surface anomalous region to form a closed band, and establishes sampling directions point by point along the closed band toward the interior of the surface anomalous region. It records the first intersection point of each sampling direction with the surface anomalous boundary and outputs the boundary record.

[0041] The liquid film generation module is used to cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and to continuously image the corresponding area after the liquid film is formed and before it disappears, and output a drying sequence.

[0042] The intersection tracking module re-searches for the first intersection position of the current surface anomaly boundary along each sampling direction in the boundary record for each frame of the image in the drying sequence, generates the intersection displacement record, intersection disappearance record and region connectivity record corresponding to each sampling direction, and outputs the evolution record;

[0043] The result discrimination module identifies surface abnormalities with displacement towards the interior of the abnormal surface region in the intersection displacement record, surface abnormalities with disappearance of intersection points in the intersection disappearance record, or surface abnormalities with the region changing from a single region to multiple regions in the region connectivity record as cleaning residual abnormalities, and identifies the remaining surface abnormalities as real surface defects, and outputs the surface defect identification results.

[0044] The technical effects and advantages of this invention are as follows:

[0045] 1. By continuously tracking the boundary evolution of abnormal surface regions during local rewetting and redrying, and using intersection displacement, intersection disappearance, and changes in regional connectivity as criteria, residual cleaning abnormalities can be distinguished from real surface defects, thereby relatively improving the misjudgment problem in SRP test sample surface inspection in the scenario of no-pattern wafer defect detection.

[0046] 2. By acquiring local images frame by frame through horizontal movement and vertical line breaks in half of the field of view, and combining the position alignment of overlapping areas with image stitching, a dry image covering the surface to be inspected can be obtained, thereby relatively reducing the impact of positional misalignment caused by local field of view dispersion on the subsequent extraction of abnormal areas on the surface.

[0047] 3. By constructing closed zones around each surface anomaly region, establishing sampling directions, and recording the first intersection point with the surface anomaly boundary, a unified reference can be provided for the subsequent search of boundary changes during the drying process, thus giving the intersection point evolution analysis a continuous corresponding basis;

[0048] 4. By covering each abnormal surface area and its corresponding closed zone with a local rewetting liquid film, and continuously collecting data from the moment the liquid film forms until the liquid film boundary disappears, the drying sequence of the corresponding abnormal surface area can be obtained, thereby changing the single-state dry appearance judgment to an identification method based on a continuous change process.

[0049] 5. By re-searching for the current intersection point along each sampling direction for each frame of the dried sequence, and combining the previous frame mapping, adjacent bounding and back projection verification to generate intersection point displacement records, intersection point disappearance records and region connectivity records, the stability of boundary change tracking can be relatively improved.

[0050] 6. By statistically analyzing the number of sampling directions of displacement towards the interior of the abnormal surface region, the number of intersection disappearance directions, and the region splitting markers in chronological order of the drying sequence, and making the final judgment accordingly, the surface defect identification results can be made to correspond with the continuous evolution process of each abnormal surface region. Attached Figure Description

[0051] Figure 1 This is a flowchart of the method steps of the present invention.

[0052] Figure 2 This is a schematic diagram of the system modules of the present invention. Detailed Implementation

[0053] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] Refer to the instruction manual appendix Figure 1-2 The present invention provides a method for identifying surface defects in SRP test samples, comprising:

[0055] S1. Obtain the SRP test sample after cleaning and drying. Under the same imaging magnification and the same lighting conditions, acquire local images of the surface to be tested by moving horizontally in half of the field of view and vertically changing lines in half of the field of view. Then, align the positions and stitch the images according to the overlapping areas of adjacent local images to output the dry image.

[0056] In this embodiment, S1 is used to complete the frame-by-frame acquisition and stitching output of the surface to be inspected, so that the subsequent extraction of abnormal areas on the surface can be directly based on the same dry image. Specifically, the relative height, imaging magnification and illumination conditions between the SRP test sample and the imaging device are fixed first. Then, the outer range of the surface to be inspected is used as the acquisition boundary, and local images are acquired frame by frame in a row scanning manner. After each movement, the overlapping area between adjacent local images is retained. Then, the displacement relationship between adjacent local images in the same row and the displacement relationship between two adjacent rows are calculated respectively. Then, all local images are converted to the corresponding positions on the surface to be inspected to complete the position alignment and image stitching. Through this processing, multiple images under a local field of view can be converted into a single dry image covering the surface to be inspected.

[0057] The implementation process includes the following steps:

[0058] Under the same imaging magnification and illumination conditions, the position corresponding to one corner of the surface under inspection is first determined as the acquisition position of the first local image, and the first local image is acquired. Then, using half the field width of the current local image as the lateral movement distance, the imaging device or SRP test sample is moved along the same row direction, and subsequent local images in the same row are acquired one by one until the lateral boundary of the current local image reaches the lateral edge of the surface under inspection, at which point the acquisition of that row ends. For every two adjacent local images in the same row, the overlapping portion of the two local images in the acquisition sequence is taken as the overlapping area, and the same surface texture points, gray-scale abrupt change points, or other features are extracted within this overlapping area. Boundary turning points are used as corresponding positions. When there are multiple corresponding positions in the overlapping area, the coordinate differences are calculated according to the horizontal and vertical coordinates of the corresponding positions. Then, all coordinate differences are summarized to obtain the horizontal and vertical displacements between the two adjacent local images. Specifically, the horizontal displacement is obtained by subtracting the horizontal coordinate of the corresponding position in the previous local image from the horizontal coordinate of the corresponding position in the later local image, and the vertical displacement is obtained by subtracting the vertical coordinate of the corresponding position in the previous local image from the vertical coordinate of the corresponding position in the later local image. The horizontal and vertical displacements corresponding to each group of adjacent local images in the same row are written into the row displacement record according to the acquisition order.

[0059] After acquiring local images in the same row, the imaging device or SRP test sample is moved longitudinally by half the current local image field of view height to enter the acquisition position of the first local image of the next row, while maintaining an overlap between the first local image of the next row and the corresponding area of ​​the previous row. Then, the local image acquisition of the next row is completed using the same lateral acquisition method as the previous row. For two adjacent rows, the overlapping portion between the local images of the previous row and the corresponding local images of the next row is selected as the overlapping area, and the same surface texture point, gray-scale abrupt change point, or boundary turning point is extracted within this overlapping area as the corresponding position. When there are multiple corresponding positions within the overlapping area, the coordinate difference is calculated according to the lateral and lateral coordinates of the corresponding positions, and all coordinate differences are summarized to obtain the lateral and lateral displacement between the corresponding local images of the two adjacent rows, and written into the row displacement record in row order. Here, the corresponding local image is preferentially selected based on the intersecting lateral coverage areas. When a local image in the next row overlaps with two or more local images in the previous row, the local image of the previous row with the larger overlapping area is selected as the corresponding local image.

[0060] Based on the in-row and inter-row displacement records, the position of the first local image on the surface to be inspected is first determined as the origin. Then, the horizontal and vertical displacements within the same row are accumulated one by one according to the acquisition sequence to determine the corresponding positions of each local image in that row on the surface to be inspected. At the line break, the corresponding position of the first local image in the next row relative to the corresponding local image in the previous row is determined based on the inter-row displacement record. The in-row and vertical displacements of that row are then accumulated until all local images have obtained their corresponding positions. After obtaining the corresponding positions of each local image, each local image is translated to its corresponding position for alignment, and the overlapping areas of adjacent local images are merged. When there are two sets of pixel values ​​at the same position in the overlapping area, the average of the two sets of pixel values ​​is taken as the pixel value after stitching. When there is boundary misalignment in the overlapping area, the corresponding position of the next local image is corrected according to the in-row and inter-row displacement records before stitching. After all local images have been aligned and stitched, a dry image covering the entire surface to be inspected is obtained for subsequent extraction of abnormal areas on the surface.

[0061] Through the above implementation process, each local image of the surface to be inspected can be converted into a complete dry image, avoiding the impact of positional offset, boundary misalignment or incomplete coverage on the extraction results of surface anomalies during subsequent extraction. At the same time, the intra-row displacement record and inter-row displacement record correspond to the displacement relationship within the same row and between two adjacent rows, respectively. After all local images are stitched together, the surface positional relationship in the dry image remains continuous, and the surface anomaly area can be directly extracted based on this dry image.

[0062] In practical applications: When the surface to be inspected is an SRP test sample surface that has been cleaned and dried, the upper left corner of the surface to be inspected can be used as the acquisition position of the first local image. Then, the first row of acquisition is completed by moving horizontally according to half the field of view. After the first row is completed, the second row of acquisition is entered by moving vertically according to half the field of view, and all rows are acquired in sequence. Then, the horizontal and vertical displacement between adjacent local images in the same row and between corresponding local images in two adjacent rows are calculated respectively. Then, all local images are stitched together into a dry image, thus providing an image basis for subsequent extraction of surface abnormal areas based on the dry image.

[0063] S2. Extract each surface anomaly region based on the dry image. Select a continuous and complete surface outside each surface anomaly region to form a closed band. Establish sampling directions point by point along the closed band toward the interior of the surface anomaly region. Record the first intersection point of each sampling direction with the surface anomaly boundary and output the boundary record.

[0064] In this embodiment, S2 is used to determine the surface anomalous boundary, closed zone, sampling direction, and boundary record corresponding to each surface anomalous region in the dry image, so that subsequent local rewetting film coverage, drying sequence acquisition, and intersection evolution analysis are all carried out around the same set of surface anomalous regions. In specific execution, the normal surface region and the surface anomalous region are first distinguished in the dry image, and then the surface anomalous boundary is determined along the outer edge of each surface anomalous region. Then, starting from the surface anomalous boundary, the continuous complete surface is searched layer by layer outward. The closed outer boundary is extracted at the position that satisfies the surrounding relationship, and the annular region between the surface anomalous boundary and the closed outer boundary is taken as the closed zone. On this basis, sampling points are selected point by point along the closed zone, and each sampling point is connected with the corresponding boundary point on the surface anomalous boundary to form a sampling direction from the closed zone to the inside of the surface anomalous region. Then, the first intersection position with the surface anomalous boundary is searched along each sampling direction and written into the boundary record.

[0065] The implementation process includes the following steps:

[0066] When extracting surface anomalous regions from dry images, the process first involves extracting the connected surface regions of the entire image and identifying normal surface regions as those with continuous grayscale, continuous texture, and no interrupted closed boundaries. Then, regions that do not belong to normal surface regions but form independent connected regions in the image are identified as surface anomalous regions. For each surface anomalous region, starting from the boundary between the anomalous and normal surface regions, boundary points are extracted point by point along the boundary contour, and closed boundary lines are generated according to the connection order of these boundary points in the image, serving as the surface anomalous boundary. When there is no normal surface interval between two surface anomalous regions, these two anomalous regions are merged into one before extracting the surface anomalous boundary. When a hole exists within a connected region, the outer closed boundary line is identified as the surface anomalous boundary of the current surface anomalous region; the hole boundary is not output as a separate surface anomalous boundary. After processing all surface anomalous regions, the surface anomalous boundaries corresponding to each surface anomalous region are output.

[0067] When searching for a continuous, complete surface layer by layer from the surface anomaly boundary outwards from the surface anomaly region, the first ring of pixels adjacent to the outer edge of the surface anomaly boundary is defined as the first layer, and the next ring of pixels adjacent to the outer edge of the first layer is defined as the second layer. This process is repeated layer by layer. Within each layer, it is first checked whether the entire layer falls within the normal surface area, and then it is checked whether the layer continuously surrounds the corresponding surface anomaly region. If both conditions are met—being entirely within the normal surface area and continuously surrounding the corresponding surface anomaly region—the layer is retained as a usable layer. As the search continues outwards to a position a set number of layers away from the surface anomaly boundary, the outer closed boundary at that position is extracted as the closed outer boundary, and the surface anomaly boundary is... The annular region between the boundary and the closed outer boundary is defined as the closed zone. A fixed number of layers is used here, specifically three, four, or five layers. In one embodiment, four layers are used to maintain a stable interval between the closed zone and the surface anomaly boundary. If, during the layer-by-layer search process, a layer intersects with other surface anomaly regions, or a layer touches the boundary of the surface to be inspected, or a layer cannot continuously encircle the corresponding surface anomaly region, then the outward expansion stops, and the outer closed boundary of the previous layer is taken as the closed outer boundary. If the first layer cannot meet the continuous encirclement condition, the current surface anomaly region is written into the area to be reviewed, and no closed zone is generated in the current processing round. After completion, the closed zones corresponding to each surface anomaly region are output.

[0068] When selecting sampling points along the closed zone, firstly, all boundary points on the closed zone are read sequentially according to the boundary point order, and each boundary point of the closed zone is treated as a sampling point. Then, for each sampling point, the straight-line distance from the sampling point to each boundary point on the surface anomaly boundary is calculated. The boundary point whose distance is at the lower limit is determined as the nearest point, and the current sampling point is connected to the nearest point to generate a sampling direction from the closed zone to the interior of the surface anomaly region. If the same sampling point corresponds to multiple boundary points with the same distance, the angle between the line connecting the sampling point to each boundary point and the local outward normal of the closed zone is further calculated, and the boundary point whose angle is at the lower limit is determined as the nearest point. If the angles are still the same, the boundary point that appears earlier in the boundary point order of the surface anomaly boundary is selected as the nearest point. After all sampling points are processed, the corresponding sampling directions are output in the order of the sampling points. Each sampling direction inherits the region number of the surface anomaly region to which it belongs and the direction number within that region, so that subsequent boundary records, intersection displacement records, and intersection disappearance records can continue to reference them with the same number.

[0069] When searching for the first intersection position with the surface anomaly boundary point by point along each sampling direction from the closed zone into the surface anomaly region, the starting point of the sampling direction on the closed zone is taken as the search starting point. Image positions are read point by point along the sampling direction into the surface anomaly region. When the current read position first falls on the surface anomaly boundary, this position is determined as the first intersection position between the current sampling direction and the surface anomaly boundary, and this first intersection position is written into the boundary record according to the region number and direction number. When the sampling direction reaches the search endpoint inside the surface anomaly region but still does not intersect with the surface anomaly boundary, the sampling direction is written into the empty intersection point record, and the empty intersection point record is also merged into the boundary record for subsequent intersection point disappearance and detection direction processing. Here, the search endpoint is taken as the extension position of the theoretical line length between the current sampling direction and the surface anomaly boundary, specifically, it can be taken as the position of twice the distance between the nearest point and the sampling point, so as to ensure that the sampling direction can pass through the surface anomaly boundary and enter the surface anomaly region under normal circumstances. After completing the processing of all sampling directions, the boundary record corresponding to each surface anomaly region is output.

[0070] Through the above implementation process, surface anomalous regions can be identified first in the dry image. Then, a closed band, sampling direction, and boundary record corresponding to each surface anomalous region can be constructed around it. This allows for direct determination of local regions based on the surface anomalous region and its corresponding closed band during subsequent local rewetting film coverage. The intersection search in the drying sequence can also be performed directly along the sampling directions in the boundary record. At the same time, backtracking or writing rules are given for cases such as failure to generate closed bands, juxtaposition of closest points, and missing first intersection positions. This ensures that each surface anomalous region corresponds to a closed band, sampling direction, and boundary record. Subsequently, local rewetting film coverage, drying sequence acquisition, and intersection position analysis can be performed directly around the surface anomalous region.

[0071] In practical applications: When there is a scratch-like surface abnormality region and a film-like surface abnormality region in a dry image, the connected range of the two surface abnormality regions can be extracted first, and then the surface abnormality boundaries can be extracted along their respective outer edges. Subsequently, normal surface regions are searched layer by layer from their respective surface abnormality boundaries. At the fourth layer position that satisfies the continuous encirclement condition, the closed outer boundary is extracted, and the corresponding closed band is formed. Then, sampling points are selected point by point along their respective closed bands. The nearest point is determined according to the distance of each sampling point to each boundary point on the surface abnormality boundary, and the sampling direction is generated. Finally, the first intersection position with the surface abnormality boundary is searched along each sampling direction and written into the boundary record. This provides a unified input for subsequent local rewetting liquid film coverage and drying sequence analysis for the two surface abnormality regions.

[0072] S3. Cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and continuously image the corresponding area from the formation of the liquid film to its disappearance, and output the drying sequence.

[0073] In this embodiment, S3 is used to form a local rewetting liquid film around each surface anomalous region and its corresponding closed zone, and continuously acquire images of the local region from the start of liquid film formation until the boundary of the local rewetting liquid film disappears, thereby obtaining a drying sequence corresponding to the surface anomalous region. Specifically, the local region is first determined based on the surface anomalous region and the closed zone, and then rewetting liquid is applied to the local region so that both the surface anomalous region and the closed zone fall within the liquid film coverage area. Subsequently, images are acquired frame by frame during the liquid film formation process, and the liquid film formation time is determined based on the change of the local rewetting liquid film boundary in adjacent frames. Then, the local rewetting liquid film boundary in each frame is continuously extracted from the liquid film formation time until the local rewetting liquid film boundary disappears, and the images acquired from the liquid film formation time to the time when acquisition stops are arranged in chronological order as a drying sequence.

[0074] The implementation process includes the following steps:

[0075] For each surface anomalous region and its corresponding closed band, the surface anomalous boundary of the surface anomalous region and the outer boundary of the closed band are first read. Then, the outer boundary of the closed band is used as the outer boundary of the local region, and the outer edge of the surface anomalous region is used as the inner coverage area of ​​the local region, thereby determining the local region that simultaneously covers the surface anomalous region and the closed band. When the same local region contains multiple surface anomalous regions, the local region is re-divided according to their respective corresponding closed bands, and multiple surface anomalous regions are not merged into the same local region. After the local region is determined, a rewetting liquid is applied to the local region. Specifically, it can be applied by dripping, spraying, or contact transfer. In one embodiment, a micro-drip application method is used. After the rewetting liquid is dripped to the center of the local region, the liquid is spread outward along the surface until the liquid boundary continuously surrounds the surface anomalous region and the closed band, and then the application is stopped. If the liquid boundary does not cover the entire range of the closed band, the rewetting liquid is added again until the closed band is completely covered by the liquid. After the application is completed, a local rewetting liquid film is output, wherein each local rewetting liquid film inherits the region number of the corresponding surface anomalous region.

[0076] During the formation of a local rewetting film, when acquiring images of the local area frame by frame, the first frame after the application of the rewetting liquid is used as the starting frame, and images of the local area are continuously acquired under the same imaging magnification and illumination conditions. For each frame, the brightness or reflection boundary between the liquid-covered and uncovered areas is first identified, and then the closed contour is extracted point by point along this boundary as the boundary of the local rewetting film in the current frame. Subsequently, the boundary of the local rewetting film in the current frame is compared with the boundary of the local rewetting film in the previous frame, and point by point, the boundary is calculated. Calculate the coordinate difference between corresponding boundary points; when the coordinate difference of all corresponding boundary points in two consecutive frames is zero, the corresponding time in the next frame is determined as the liquid film formation time; if the boundary of the local rewetting liquid film in the current frame and the boundary of the local rewetting liquid film in the previous frame still have boundary expansion, boundary contraction or boundary break closure, then continue to acquire images frame by frame until the boundary of the local rewetting liquid film in two consecutive frames remains unchanged; after this processing, the boundary of the local rewetting liquid film corresponding to the liquid film formation time is used as the starting boundary of the subsequent drying process;

[0077] From the moment the liquid film forms, when continuously acquiring images of a local area, the imaging magnification, illumination conditions, and local area position remain unchanged. Images are acquired frame by frame at the same acquisition rhythm as the liquid film formation process. For each frame, the local rewetting liquid film boundary is extracted according to the brightness or reflection boundary between the liquid-covered and uncovered areas, and this local rewetting liquid film boundary is written into the liquid film boundary sequence in the order of acquisition time. If a local break occurs in the local rewetting liquid film boundary in a certain frame, it is first checked whether the two ends of the break are still located on the outer edge of the same liquid-covered area. When the two ends of the break still enclose the same liquid-covered area, the shortest connection between the two ends of the break is written as the local rewetting liquid film boundary in the current frame. When the two ends of the break enclose multiple separate liquid-covered areas, the outer boundary of each separate liquid-covered area is extracted and summarized according to the coverage area as the local rewetting liquid film boundary in the current frame. After processing all frames, the liquid film boundary sequence of the corresponding local area is output.

[0078] When acquisition stops based on the liquid film boundary sequence, the local rewetting liquid film boundaries in each frame after the liquid film formation time are checked in chronological order. If there is no longer a continuous local rewetting liquid film boundary surrounding the liquid coverage area in the current frame image, and no continuous local rewetting liquid film boundary surrounding the liquid coverage area in the next frame image, the corresponding time of the current frame is determined as the stop acquisition time, and image acquisition stops. If no local rewetting liquid film boundary is extracted in the current frame image, but a continuous boundary reappears in the next frame image, acquisition continues, and the corresponding time of the current frame is not determined as the stop acquisition time. After acquisition stops, all images acquired between the liquid film formation time and the stop acquisition time are arranged in chronological order to form a drying sequence corresponding to the current surface abnormal area. When there are multiple surface abnormal areas in the same surface to be inspected, the above processing is performed on the local areas corresponding to each surface abnormal area, and the corresponding drying sequences are output respectively.

[0079] Through the above implementation process, a local rewetting liquid film can be formed in the local area corresponding to the surface anomaly area and the closed zone. Then, taking the moment of liquid film formation as a unified starting point, the entire process of the local rewetting liquid film from formation to disappearance is continuously recorded, thus providing a continuous image basis for subsequent searching of the first intersection position of the current surface anomaly boundary along the sampling direction. At the same time, by setting clear determination methods for the liquid film formation moment and the stop acquisition moment, invalid images that have not yet spread the liquid or have disappeared can be avoided from being included in the drying sequence.

[0080] In practical applications: Once a closed band corresponding to a certain surface anomalous area has been formed, the local area can be determined first by the outer boundary of the closed band. Then, a rewetting liquid is dropped onto the center of this local area, allowing the liquid to spread and continuously cover the surface anomalous area and the closed band. Subsequently, images of the local area are acquired frame by frame, and the moment of liquid film formation is determined when the boundary of the local rewetting liquid film remains unchanged in two consecutive frames. Images are then continuously acquired from this moment of liquid film formation until no continuous local rewetting liquid film boundary appears in two consecutive frames. Finally, all images from the moment of liquid film formation to the moment of stopping acquisition are arranged in chronological order to obtain the drying sequence corresponding to the surface anomalous area.

[0081] S4. For each frame of the image in the drying sequence, search again for the first intersection point of the current surface anomaly boundary along each sampling direction in the boundary record, generate the intersection point displacement record, intersection point disappearance record and region connectivity record corresponding to each sampling direction, and output the evolution record.

[0082] In this embodiment, S4 is used to determine the current intersection point of each sampling direction in the current frame frame by frame based on the dry sequence and boundary records, and further form intersection point displacement records, intersection point disappearance records, and region connectivity records to obtain evolution records that can be directly used for subsequent discrimination. In specific execution, for each frame image in the dry sequence, the boundary change points in the current frame are searched along each sampling direction in the boundary records, and candidate records are generated by combining the corresponding intersection point positions in the boundary records, the corresponding intersection point positions in the previous frame image, and the corresponding intersection points in adjacent sampling directions; then, the current frame is processed based on the candidate records. The current intersection point of the frame is initially determined, and then the sampling direction that has not completed the initial determination is re-reviewed in the next frame for verification. The verification results are added to the current intersection point record or the intersection point disappearance record. On this basis, the intersection point changes and connectivity changes of each surface anomaly region in the current frame are statistically analyzed in the order of sampling direction to form an evolution record. In order to ensure that the current intersection point in the next frame image has a clear source, this implementation method first forms a current intersection point record and a record to be inspected for each frame image. Then, after the direction candidate point screening and current intersection point determination are completed in the next frame image, the record to be inspected in the previous frame image is re-reviewed for verification. Finally, the evolution record is output frame by frame.

[0083] The implementation process includes the following steps:

[0084] For each frame of the drying sequence and each sampling direction in the boundary record, first read the corresponding closed band, sampling direction, and intersection point in the boundary record corresponding to the current surface anomaly region. Then, using the starting point of the sampling direction on the closed band as the search starting point, read the image position in the current frame image point by point along the sampling direction towards the interior of the surface anomaly region. When a grayscale jump, edge transition, or connectivity switch occurs at the current reading position, indicating a transition from a continuous, intact surface to a surface anomaly region, this position is recorded as a boundary change point. For the same sampling direction, if multiple boundary change points appear along the search path, all boundary change points are retained as candidate points for the current direction. Then, read the corresponding intersection point in the boundary record, and use the current intersection point of the same surface anomaly region and the same sampling direction in the previous frame image as the previous frame mapping point. If the current frame image is in the drying sequence... In the first frame image, the corresponding intersection point position in the aforementioned boundary record replaces the mapping point of the previous frame; then, the corresponding intersection point positions of the adjacent sampling directions on both sides of the current sampling direction in the boundary record or the corresponding intersection point positions in the previous frame image are read, and these two positions are respectively used as the left adjacent direction enclosing point and the right adjacent direction enclosing point, and the left adjacent direction enclosing point and the right adjacent direction enclosing point are merged into the adjacent direction enclosing point; after all sampling directions are processed, the current direction candidate point, the previous frame mapping point and the adjacent direction enclosing point are written according to the surface anomaly area number and the direction number to form a candidate record; if a certain sampling direction is located at the beginning or end of the sampling direction sequence, and there is only one side of adjacent sampling direction, then the corresponding intersection point position of the existing side of adjacent sampling direction and the corresponding intersection point position of the current sampling direction in the boundary record are used together to generate the adjacent direction enclosing point, so as to keep the subsequent determination process of the sampling direction uninterrupted;

[0085] For each sampling direction in the candidate record, first read all current direction candidate points and previous frame mapping points corresponding to that sampling direction, and calculate the previous frame direction distance from each current direction candidate point to the previous frame mapping point along the same sampling direction. If the previous frame direction distance is zero, it indicates that the current direction candidate point coincides with the previous frame mapping point, and the current direction candidate point is directly retained as a candidate point whose previous frame direction distance is at the lower limit. If there are multiple current direction candidate points, the current direction candidate point whose previous frame direction distance value is at the lower limit is taken as the candidate point to be judged. Among them, the candidate point whose previous frame direction distance value is at the lower limit is the direction candidate point with the smallest previous frame direction distance value. Then, read the adjacent direction enclosing points, and form an adjacent direction enclosing area by the left and right adjacent direction enclosing points and the corresponding left and right adjacent sampling points on the closed band, and then determine whether the candidate point to be judged falls into the adjacent direction enclosing area. When the candidate point to be judged falls into the adjacent direction enclosing area, the candidate point to be judged is determined as the current intersection point, and written into the current intersection point record according to the surface anomaly area number and direction number. If the candidate point to be judged does not fall into the adjacent direction enclosing area, the candidate point to be judged is determined as the current intersection point, and written into the current intersection point record according to the surface anomaly area number and direction number. If a point is found within the adjacent enclosing region, the system continues to check if there are other candidate points for the current direction within the adjacent enclosing region. If there are other candidate points for the current direction within the adjacent enclosing region, the system calculates the distance from each other candidate point to the closed zone direction, and identifies the candidate point whose closed zone direction distance is below the lower limit as the current intersection point, writing it into the current intersection point record. If the candidate point to be judged does not fall within the adjacent enclosing region and there are no other candidate points for the current direction within the adjacent enclosing region, the sampling direction is written into the inspection record. If a sampling direction has multiple candidate points for the current direction with the same closed zone direction distance and all below the lower limit, the system first checks if only one candidate point falls within the adjacent enclosing region. If only one candidate point falls within the adjacent enclosing region, the candidate point is identified as the current intersection point. If multiple candidate points fall within the adjacent enclosing region, the candidate point closer to the closed zone is selected as the current intersection point. If multiple candidate points do not fall within the adjacent enclosing region, the sampling direction is written into the inspection record. After completing the initial judgment of all sampling directions, the current intersection point record and the inspection record are output.

[0086] For each sampling direction in the record to be inspected, after completing the selection of candidate points and determining the current intersection point in the next frame image, all candidate points corresponding to the same surface anomaly area and the same sampling direction in the next frame image are read, and the determined current intersection point in the next frame image is also read. Then, the current intersection point in the next frame image is back-mapped to the current frame image along the corresponding sampling direction to form a back-projection point. The back-projection adopts the same-direction projection method in the sampling direction, projecting the coordinates of the current intersection point in the next frame image onto the same sampling direction coordinate line in the current frame image. Subsequently, the line segment interval between the back-projection point in the current frame image and the mapping point in the previous frame is taken, and the number of candidate points in the current direction within the line segment interval is checked. When there is a candidate point in the current direction within the line segment interval, the candidate point in the current direction is determined as the current intersection point and written into the verification intersection point record. When there is no candidate point in the current direction within the line segment interval, the sampling direction is written into the intersection point disappearance record. When there are multiple candidate points in the line segment interval, the current direction is not recorded. When processing multiple candidate points in the current direction, the current intersection points of the adjacent sampling directions on both sides of the current sampling direction are read, and a line connecting these two current intersection points is formed. Then, the multiple candidate points in the current direction are checked on which side of the line connecting the adjacent intersection points. If only one of the multiple candidate points in the current direction is on the same side of the line connecting the adjacent intersection points as the mapping point of the previous frame, then that candidate point is determined as the current intersection point and written into the verification intersection point record. If all the candidate points in the current direction are on the same side of the line connecting the adjacent intersection points as the mapping point of the previous frame, then the candidate point closer to the closed band is taken as the current intersection point and written into the verification intersection point record. If none of the multiple candidate points in the current direction are on the same side of the line connecting the adjacent intersection points as the mapping point of the previous frame, then the sampling direction is written into the intersection disappearance record. For the last frame image in the dried sequence, since there is no subsequent frame image, the reverse mapping verification is not performed, and the sampling direction in the record to be inspected is directly written into the intersection disappearance record. After processing all the directions to be inspected, the verification intersection point record and the intersection disappearance record are output.

[0087] For each surface anomaly region in each frame of the image, the corresponding intersection point positions in the current intersection point record, the verification intersection point record, and the boundary record are read sequentially according to the sampling direction. For the same sampling direction, the current intersection point in the current intersection point record is prioritized. If the current intersection point record does not contain a current intersection point for that sampling direction, but the verification intersection point record does, the current intersection point in the verification intersection point record is used. Then, the intersection point displacement distance from each current intersection point to its corresponding intersection point position in the boundary record along the same sampling direction is calculated and written into the intersection point displacement record. When the current intersection point is located between its corresponding intersection point position in the boundary record and the closed zone, it is recorded as a displacement towards the interior of the surface anomaly region. Next, the intersection point disappearance record is read sequentially according to the sampling direction, and the sampling direction segments with consecutive direction numbers in the intersection point disappearance record are identified as boundary gaps. After the boundary gaps are determined, the connectivity range of the corresponding surface anomaly region is extracted in the current frame image, and the connections to the boundary gaps are counted. The number of connected regions is calculated, and the boundary gap is cross-validated with the number of connected regions. When the boundary gap corresponds to only one connected region, a region preservation flag is written. When the boundary gap corresponds to two or more separate connected regions, a region split flag is written. When the originally separate regions on both sides of the boundary gap are reconnected into a whole in the current frame image, a region merging flag is written. When there is no boundary gap in the current frame image, the connected range of the surface anomaly region in the current frame image is compared with the connected range in the previous frame image. If the number of connected ranges remains unchanged, a region preservation flag is written. If the number of connected ranges increases, a region split flag is written. If the number of connected ranges decreases, a region merging flag is written. After completion, the region preservation flag, region split flag, and region merging flag are merged into a region connectivity record. Then, the intersection displacement record, intersection disappearance record, and region connectivity record are merged according to the surface anomaly region number and frame order to output the evolution record.

[0088] Through the above implementation process, each frame of the drying sequence can be transformed into the intersection change result and the region change result corresponding one-to-one with the sampling direction. This allows the subsequent distinction between cleaning residual anomalies and real surface defects to be based on the intersection displacement, intersection disappearance and region connectivity changes of the same surface anomaly region in consecutive frames, rather than on the appearance difference of a single frame.

[0089] In practical applications: When a certain surface anomalous region gradually shrinks in the drying sequence, multiple boundary change points can be searched along each sampling direction in the current frame image. Candidate records are formed by combining the current intersection point position in the previous frame image with the corresponding intersection point position in the adjacent sampling direction. Then, the current intersection point of most sampling directions is determined according to the distance of the previous frame direction and the adjacent enclosing area. For the few undetermined sampling directions, the current intersection point of the next frame image is called back for verification. Subsequently, the current intersection point of all sampling directions in the current frame image is compared with the corresponding intersection point position in the boundary record to obtain the intersection point displacement record. Combined with the intersection point disappearance record and the connectivity range in the current frame image, a region connectivity record is generated. Finally, the evolution record corresponding to the current frame image is formed for subsequent judgment on whether the surface anomalous region belongs to the cleaning residue anomalous region.

[0090] S5. Based on the evolution record, identify the surface abnormal areas in the intersection displacement record that have displacement toward the interior of the surface abnormal area, the surface abnormal areas in the intersection disappearance record that have intersection disappearance, or the surface abnormal areas in the region connectivity record that have changed from a single region to multiple regions as cleaning residual abnormalities, identify the remaining surface abnormal areas as real surface defects, and output the surface defect identification results.

[0091] In this embodiment, S5 is used to make a final judgment on each surface anomalous region based on the evolution record, distinguishing the surface anomalous regions that meet the abnormal change characteristics of cleaning residue from the surface anomalous regions that maintain the characteristics of the entity boundary, and outputting the surface defect identification result. Specifically, in the order of the drying sequence, the intersection displacement record, intersection disappearance record, and region connectivity record of each surface anomalous region in each frame image are read first. Then, the number of sampling directions of displacement towards the interior of the surface anomalous region, the number of intersection disappearance directions, and the occurrence of region splitting markers are counted respectively to form the region change record corresponding to the surface anomalous region. Then, based on the region change record, it is determined whether the surface anomalous region has a change chain of boundary inward contraction and further intersection disappearance or region splitting during the continuous drying process. If it has, it is identified as a cleaning residue anomalous region; if it does not have, it is identified as a real surface defect. Finally, the judgment results of all surface anomalous regions are summarized and the surface defect identification result is output.

[0092] The implementation process includes the following steps:

[0093] For each surface anomalous region in the evolution record, firstly, according to the region number of the surface anomalous region, read the intersection displacement record, intersection disappearance record, and region connectivity record corresponding to each frame of the drying sequence. Then, count the three fields frame by frame in the time sequence of the drying sequence. When counting the intersection displacement record, read the intersection displacement distance corresponding to each sampling direction in the current frame image one by one, and record the sampling direction between the corresponding intersection position in the boundary record and the closed zone as the sampling direction of the displacement towards the interior of the surface anomalous region. Then, count each of these sampling directions to obtain the number of sampling directions of displacement towards the interior of the surface anomalous region in the current frame image. When counting the intersection disappearance record, read the intersection disappearance records in the current frame image one by one. The direction numbers are counted, and the number of intersection disappearance directions in the current frame image is obtained. When counting the region connectivity record, the region preservation marker, region split marker, and region merger marker in the current frame image are read. When there is a region split marker in the current frame image, the frame is recorded as a split frame. When there is no region split marker, the frame is recorded as a non-split frame. After the statistics of all frame images are completed, the number of sampling directions, the number of intersection disappearance directions, and the region split marker corresponding to the displacement into the surface abnormal region of each frame image are written into the same region change record in chronological order. Each time position in the region change record corresponds to a frame image in the drying sequence, and the order of the time positions is consistent with the order of the images in the drying sequence.

[0094] For each surface anomaly region's regional change record, firstly, check in chronological order whether the number of sampling directions of displacement towards the interior of the surface anomaly region between adjacent frames increases frame by frame. In this embodiment, a continuous increase is defined as the corresponding number increasing sequentially in three consecutive frames. If the number only increases between two frames and then falls back or remains unchanged in the next frame, it is not considered a continuous increase. After determining a continuous increase, check subsequent frames starting from the termination frame of the continuous increase. The check range for subsequent frames is all remaining frames after the termination frame of the continuous increase. When the number of intersection disappearance directions in any subsequent frame is greater than zero, or when any frame has a regional splitting marker, the surface anomaly region is identified as a cleaning residue anomaly. If the regional change record does not show the corresponding direction of displacement towards the interior of the surface anomaly region in three consecutive frames, then the continuous increase is considered a continuous increase. If the number of sampling directions of displacement within a surface anomaly region increases sequentially, or if there is a continuous increase but the number of all intersection disappearance directions in subsequent frame images is zero and there are no region splitting markers in any frame, then the surface anomaly region is identified as a real surface defect. After this processing, the cleaning residue anomaly corresponds to a change chain that first occurs inward contraction of the boundary, and then continues to occur with intersection disappearance or region splitting; the real surface defect corresponds to a surface anomaly region that did not form this change chain during the drying process. For surface anomaly regions with fewer than three frames in the drying sequence, since it is not possible to form a frame-by-frame comparison of three consecutive frames, the judgment is directly based on whether there is a number of intersection disappearance directions greater than zero or a region splitting marker in subsequent frame images. If they exist, they are identified as cleaning residue anomalies; if they do not exist, they are identified as real surface defects.

[0095] After identifying all surface anomaly regions, the identification results for each surface anomaly region are read one by one and written into the cleaning residue anomaly and the actual surface defect according to the region number. When a surface anomaly region corresponding to a certain region number is identified as a cleaning residue anomaly, the region number, the location range of the surface anomaly region in the dry image, and the corresponding cleaning residue anomaly label are written into the result table. When a surface anomaly region corresponding to a certain region number is identified as an actual surface defect, the region number, the location range of the surface anomaly region in the dry image, and the corresponding actual surface defect label are written into the result table. After all surface anomaly regions are written, all region numbers, location ranges, and label information in the result table are summarized, and the surface defect identification results are output. When outputting, the results can be arranged in the order of region number or in the order of the spatial location of the surface anomaly regions in the dry image. In one embodiment, the results are output in the order of region number so as to correspond to the aforementioned boundary record, drying sequence, and evolution record item by item.

[0096] Through the above implementation process, the intersection displacement records, intersection disappearance records and region connectivity records corresponding to each frame of the evolution record can be further merged into the overall change chain of each surface abnormal region. Then, based on the boundary inward movement, intersection disappearance and region splitting relationship of the same surface abnormal region in the continuous drying process, the final judgment is completed, so that the distinction between cleaning residual abnormalities and real surface defects is based on the continuous change process, rather than on the appearance difference of a single frame.

[0097] In practical applications: when a certain surface anomalous region has two, four, and six sampling directions for displacement toward its interior in the first few frames of the drying sequence, and one or more intersection disappearance directions appear in subsequent frames, or a region splitting mark is written in the current frame, then the surface anomalous region is identified as a cleaning residue anomalous region; when another surface anomalous region has changes in the intersection displacement distance of individual sampling directions in the drying sequence, but the number of sampling directions for displacement toward its interior does not increase sequentially in three consecutive frames, and no intersection disappearance direction or region splitting mark appears in subsequent frames, then the surface anomalous region is identified as a real surface defect. Finally, the identification results of all surface anomalous regions are summarized into a unified surface defect identification result.

[0098] Furthermore, a surface defect identification system for SRP test samples includes:

[0099] The image acquisition module is used to acquire SRP test samples that have undergone cleaning and drying. Under the same imaging magnification and the same lighting conditions, it acquires local images of the surface under test by moving horizontally in half of the field of view and vertically wrapping in half of the field of view. It then aligns the positions of adjacent local images and stitches them together to output a dry image.

[0100] The boundary construction module extracts various surface anomalous regions based on dry images, selects continuous and complete surfaces outside each surface anomalous region to form a closed band, and establishes sampling directions point by point along the closed band toward the interior of the surface anomalous region. It records the first intersection point of each sampling direction with the surface anomalous boundary and outputs the boundary record.

[0101] The liquid film generation module is used to cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and to continuously image the corresponding area after the liquid film is formed and before it disappears, and output a drying sequence.

[0102] The intersection tracking module re-searches for the first intersection position of the current surface anomaly boundary along each sampling direction in the boundary record for each frame of the image in the drying sequence, generates the intersection displacement record, intersection disappearance record and region connectivity record corresponding to each sampling direction, and outputs the evolution record;

[0103] The result discrimination module identifies surface abnormalities with displacement towards the interior of the abnormal surface region in the intersection displacement record, surface abnormalities with disappearance of intersection points in the intersection disappearance record, or surface abnormalities with the region changing from a single region to multiple regions in the region connectivity record as cleaning residual abnormalities, and identifies the remaining surface abnormalities as real surface defects, and outputs the surface defect identification results.

[0104] Working Principle: This solution addresses surface inspection of SRP test samples in patternless wafer defect detection. First, local images of the surface to be inspected are acquired frame by frame and stitched together to form a dry image. Then, surface anomalous regions are extracted from the dry image. A closed band is formed outward from each anomalous region, and a sampling direction is established from the closed band towards the interior of the anomalous region. The initial intersection points of each sampling direction with the surface anomalous boundary are recorded. Based on this, a local rewetting liquid film is applied to each surface anomalous region and its corresponding closed band. Images are continuously acquired from the formation of the liquid film until the liquid film boundary disappears, resulting in a drying sequence. Subsequently, images are processed within the drying sequence... The process involves re-searching for the current intersection point frame by frame along each sampling direction, generating records of intersection point displacement, intersection point disappearance, and region connectivity. These records are then used to determine whether the abnormal surface region continues to shrink inward during the drying process, whether intersection points disappear, and whether region splitting occurs. This distinguishes the cleaning residue anomalies from the actual surface defects. The core of the entire process is to first fix the initial boundary position of each abnormal surface region in the dry image, then track how this boundary changes during rewetting and redrying, and finally use the change process itself to complete the identification, rather than just looking at the appearance of a single image.

[0105] For example, in a scenario of detecting defects in patternless wafers, a cleaned and dried SRP test sample may have both a real scratch and a film spot left after cleaning. Both may appear as anomalous areas with complete boundaries in the dry image, making them difficult to distinguish based on appearance alone. According to this solution, the entire sample surface is first stitched together for imaging. Then, closed zones and sampling directions are established around the two surface anomalous areas. Subsequently, rewetting liquid is applied to their respective local areas, and images of the drying process are continuously acquired. As the liquid gradually recedes, the surface anomalous area corresponding to the film spot will show inward boundary retreat, disappearance of some sampling direction intersections, and even region splitting. In contrast, the surface anomalous area corresponding to the real scratch maintains a relatively stable boundary and does not continuously reconstruct along the drying process. Therefore, the system will ultimately identify the former as a cleaning residue anomalous area and the latter as a real surface defect. In this way, in actual detection, easily misjudged residual traces can be separated from the defects that truly need attention.

[0106] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for identifying surface defects in SRP test samples, characterized in that, include: S1. Obtain the SRP test sample after cleaning and drying. Under the same imaging magnification and the same lighting conditions, acquire local images of the surface to be tested by moving horizontally in half of the field of view and vertically changing lines in half of the field of view. Then, align the positions and stitch the images according to the overlapping areas of adjacent local images to output the dry image. S2. Extract each surface anomaly region based on the dry image. Select a continuous and complete surface outside each surface anomaly region to form a closed band. Establish sampling directions point by point along the closed band toward the interior of the surface anomaly region. Record the first intersection point of each sampling direction with the surface anomaly boundary and output the boundary record. S3. Cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and continuously image the corresponding area from the formation of the liquid film to its disappearance, and output the drying sequence. S4. For each frame of the image in the drying sequence, search again for the first intersection point of the current surface anomaly boundary along each sampling direction in the boundary record, generate the intersection point displacement record, intersection point disappearance record and region connectivity record corresponding to each sampling direction, and output the evolution record. S5. Based on the evolution record, identify the surface abnormal areas in the intersection displacement record that have displacement toward the interior of the surface abnormal area, the surface abnormal areas in the intersection disappearance record that have intersection disappearance, or the surface abnormal areas in the region connectivity record that have changed from a single region to multiple regions as cleaning residual abnormalities. Identify the remaining surface abnormal areas as real surface defects and output the surface defect identification results.

2. The method for surface defect identification of SRP test samples according to claim 1, characterized in that: S1 includes: S1-1. Under the same imaging magnification and the same illumination conditions, the surface to be inspected is acquired frame by frame in the same row by moving horizontally in half of the field of view. The overlapping area of ​​each two adjacent local images is extracted, the horizontal displacement and vertical displacement of the corresponding position in the overlapping area are calculated, and the displacement record in the row is output. S1-2. After the local image of the same row is acquired, the local image of the surface to be inspected is acquired in the manner of vertical line wrapping in half the field of view. The overlapping area of ​​the corresponding local images of the two adjacent rows is extracted, the horizontal displacement and vertical displacement of the corresponding position in the overlapping area are calculated, and the inter-row displacement record is output. S1-3. Based on the inline displacement record and the interline displacement record, determine the corresponding position of each local image on the surface to be inspected, and perform position alignment and image stitching according to the corresponding position of each local image to output a dry image.

3. The method for surface defect identification of SRP test samples according to claim 2, characterized in that: S2 includes: S2-1. Extract each surface anomaly region based on the dry image, and determine the surface anomaly boundary point by point along the outer edge of each surface anomaly region, and output the surface anomaly boundary. S2-2. Starting from the surface anomaly boundary, search for a continuous and complete surface layer by layer outward from the surface anomaly region. Extract the closed outer boundary at a position with a set number of layers between it and the surface anomaly boundary. Determine the annular region between the surface anomaly boundary and the closed outer boundary as the closed zone and output the closed zone.

4. The method for surface defect identification of SRP test samples according to claim 3, characterized in that: S2 further includes: S2-3. Select sampling points along the closed band point by point, and connect the nearest point on the surface anomaly boundary for each sampling point to generate a sampling direction from the closed band to the inside of the surface anomaly area, and output the sampling direction. S2-4. Search point by point along each sampling direction from the closed zone toward the interior of the surface anomaly region for the first intersection position with the surface anomaly boundary, and record each first intersection position as a boundary record.

5. The method for surface defect identification of SRP test samples according to claim 4, characterized in that: S3 includes: S3-1. For each surface abnormality area and its corresponding closed zone, determine the local area covering the surface abnormality area and the closed zone, apply rewetting liquid to the local area to form a local rewetting liquid film, and output the local rewetting liquid film. S3-2. During the formation of the local rewetting liquid film, images of the local area are acquired frame by frame, the boundary of the local rewetting liquid film in each frame is extracted, and the time of liquid film formation is determined based on the change of the local rewetting liquid film boundary in two adjacent frames.

6. The method for surface defect identification of SRP test samples according to claim 5, characterized in that: S3 further includes: S3-3. Starting from the moment the liquid film forms, continuously acquire images of the local area, extract the local rewetting liquid film boundary in each frame of the image, and output the liquid film boundary sequence. S3-4. Based on the liquid film boundary sequence, stop acquiring data when the local rewetting liquid film boundary disappears, and arrange the images acquired between the liquid film formation time and the time when acquisition stops in chronological order to output the drying sequence.

7. A method for identifying surface defects in SRP test samples according to claim 6, characterized in that: S4 includes: S4-1. For each frame of image and each sampling direction in the boundary record in the drying sequence, search for boundary change points point by point from the closed zone toward the interior of the surface abnormal area along the sampling direction, and simultaneously extract the corresponding intersection points in the boundary record, the corresponding intersection points in the previous frame image, and the corresponding intersection points in the adjacent sampling directions, and generate candidate points in the current direction, mapping points in the previous frame, and adjacent bounding points, and output candidate records. S4-2. For each sampling direction in the candidate record, first calculate the distance between the current candidate point and the previous frame mapping point, and extract the candidate points whose distances are at the lower limit of the previous frame direction. Then, determine whether the candidate points whose distances are at the lower limit of the previous frame direction fall between the adjacent bounding points of the adjacent sampling direction. When the candidate points whose distances are at the lower limit of the current frame direction fall between the adjacent bounding points, the candidate points whose distances are at the lower limit of the previous frame direction are determined as the current intersection point. When the candidate points whose distances are at the lower limit of the current frame direction do not fall between the adjacent bounding points and there are candidate points between the adjacent bounding points, the candidate point closest to the closed band between the adjacent bounding points is determined as the current intersection point. Otherwise, write the sampling direction into the record to be inspected, and output the current intersection point record and the record to be inspected.

8. The method for surface defect identification of SRP test samples according to claim 7, characterized in that: S4 further includes: S4-3. For each sampling direction in the record to be inspected, extract the candidate points of the corresponding sampling direction in the next frame of the current frame image, and map the current intersection point determined in the next frame image back to the current frame image along the corresponding sampling direction to form a return point. When there is a candidate point between the return point and the previous frame mapping point, the candidate point is determined as the current intersection point. When there is no candidate point between the return point and the previous frame mapping point, the sampling direction is written into the intersection disappearance record. When there are multiple candidate points between the return point and the previous frame mapping point, the candidate point that is on the same side as the intersection point of the adjacent sampling direction is determined as the current intersection point. Otherwise, the sampling direction is written into the intersection disappearance record. Output the verification intersection record and the intersection disappearance record. S4-4. For each surface anomaly region in each frame of the image, summarize the intersection displacement distance between the current intersection record and the corresponding intersection position in the verification intersection record and the boundary record in the sampling direction order to generate an intersection displacement record. Determine the boundary gap according to the distribution of consecutive adjacent sampling directions in the intersection disappearance record. Then cross-validate the connectivity status of the surface anomaly region in the current frame of the image with the boundary gap. Write a region preservation mark when the boundary gap corresponds to a single connected region, write a region split mark when the boundary gap corresponds to multiple separated connected regions, otherwise write a region merge mark. Merge the region preservation mark, region split mark and region merge mark to generate a region connectivity record. Merge the intersection displacement record, intersection disappearance record and region connectivity record, and output the evolution record.

9. A method for identifying surface defects in SRP test samples according to claim 8, characterized in that: S5 includes: S5-1. For each surface anomalous region in the evolution record, according to the time sequence of the drying sequence, count the number of sampling directions of displacement towards the interior of the surface anomalous region in the intersection displacement record of each frame image, the number of intersection disappearance directions in the intersection disappearance record, and the region splitting markers in the region connectivity record, and output the region change record. S5-2. For each surface abnormality area, when the number of sampling directions of displacement towards the interior of the surface abnormality area increases continuously between adjacent frame images, and the number of intersection disappearance directions in subsequent frame images is greater than zero or a region splitting mark appears, the surface abnormality area is identified as a cleaning residue abnormality; otherwise, it is identified as a real surface defect. S5-3. Summarize the cleaning residue abnormalities and actual surface defects corresponding to each abnormal surface area, and output the surface defect identification results.

10. A surface defect identification system for SRP test samples, characterized in that, include: The image acquisition module is used to acquire SRP test samples that have undergone cleaning and drying. Under the same imaging magnification and the same lighting conditions, it acquires local images of the surface under test by moving horizontally in half of the field of view and vertically wrapping in half of the field of view. It then aligns the positions of adjacent local images and stitches them together to output a dry image. The boundary construction module extracts various surface anomalous regions based on dry images, selects continuous and complete surfaces outside each surface anomalous region to form a closed band, and establishes sampling directions point by point along the closed band toward the interior of the surface anomalous region. It records the first intersection point of each sampling direction with the surface anomalous boundary and outputs the boundary record. The liquid film generation module is used to cover each surface anomalous area and its corresponding closed zone with a local rewetting liquid film, and to continuously image the corresponding area after the liquid film is formed and before it disappears, and output a drying sequence. The intersection tracking module re-searches for the first intersection position of the current surface anomaly boundary along each sampling direction in the boundary record for each frame of the image in the drying sequence, generates the intersection displacement record, intersection disappearance record and region connectivity record corresponding to each sampling direction, and outputs the evolution record; The result discrimination module identifies surface abnormalities with displacement towards the interior of the abnormal surface region in the intersection displacement record, surface abnormalities with disappearance of intersection points in the intersection disappearance record, or surface abnormalities with the region changing from a single region to multiple regions in the region connectivity record as cleaning residual abnormalities, and identifies the remaining surface abnormalities as real surface defects, and outputs the surface defect identification results.

Images