A detection system for a silicon photonics chip

Abstract

The present invention relates to the field of optical testing technology, specifically a detection system for a silicon photonic chip, comprising a structural perspective control module, a reflection anomaly capture module, a direction dislocation verification module, a response distribution balance module, and a structural state calibration module. In the present invention, by extracting grayscale differences and continuous brightness change points in an image, combined with sudden jump boundary recognition and tangent plane construction, accurate positioning of microstructure mutation areas and extraction of incident directions are achieved, the acquisition direction is rotated to suppress grayscale fluctuations, the optical stability of the edge alignment layer is improved, the brightness expansion trajectory is collected and the fracture position is identified, and reflection anomaly points are screened in combination with diffusion amplitude differences, effectively enhancing the recognition ability of inhomogeneous reflections, arranging brightness trends and calculating angles, extracting directional continuous offset trajectories, realizing dynamic modeling of directional deviations, calculating the illumination difference inside and outside the structure and comparing contour paths, and improving the accuracy of distinguishing boundary response evolution and abnormal contour distribution.

Description

Technical Field

[0001] The present invention relates to the field of optical testing technology, and in particular to a detection system for a silicon photonic chip. Background Art

[0002] Optical testing technology includes test methods and detection processes used to evaluate and verify optical device systems and their performance. The core content is to quantitatively measure the optical characteristic parameters of optical components through specific means, including luminous flux distribution, spectral response, phase difference, reflectivity, transmittance and optical coupling efficiency. Optical testing methods are widely used in the fields of lasers, optical fiber sensors, optical communication modules, silicon photonic chips, etc., and rely on precision testing instruments and devices, such as interferometers, spectrometers, detection array devices, etc., for high-precision data acquisition and analysis. The technical system in this field covers multiple aspects such as light source control, optical path design, detection method selection, and test process optimization. It aims to accurately describe the structure and functional status of optical components to ensure their applicability and performance stability in complex systems.

[0003] Existing silicon photonics chip inspection processes primarily rely on static single-frame image grayscale contrast to detect edge structures. This makes it impossible to dynamically track the continuous evolution of diffraction features over time or direction during structural changes, often resulting in unrecognizable edge distortions. During reflection behavior analysis, a lack of a fracture identification standard based on brightness trajectory variation patterns makes it difficult to capture localized irregular reflection anomalies arising from complex reflective surfaces, resulting in the inability to classify some areas of weak reflection distortion. Brightness trend extraction utilizes only a directional edge algorithm, lacking the ability to identify structural continuity in the angular dimension, making it easy for small directional dislocations to be misidentified as boundary textures. Analysis methods for changes in illumination intensity response fail to establish a chain of differential evolution between the inside and outside of the structure, resulting in delayed detection of boundary energy conduction trends and impacting local stability assessments. Static contour path comparison methods are used during structural morphology verification, but they struggle to identify subtle contour changes due to stress migration or boundary drift. This results in low granularity for identifying structural anomalies and can lead to misidentification or omission of some contour blocks. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the existing technology and propose a silicon photonic chip detection system to improve the accuracy of distinguishing boundary response evolution and abnormal contour distribution.

[0005] In order to achieve the above object, the present invention provides a detection system for a silicon photonic chip, comprising:

[0006] The structural perspective control module is used to obtain the edge diffraction image area, extract the horizontal and vertical grayscale differences and three brightness change points, calculate the trend difference, determine the jump boundary, construct the cutting plane and the incident direction, record the grayscale fluctuations of consecutive frames, and rotate the acquisition direction until the fluctuation is lower than the reference value, thereby generating the edge alignment image layer;

[0007] A reflection anomaly capture module, based on the edge alignment image layer, records illumination uniformity, collects reflection brightness traces, identifies and marks breaks, filters areas with more intense diffusion than neighboring areas, and generates a distribution map of abnormal reflection points;

[0008] The direction misalignment verification module extracts the brightness trend based on the abnormal reflection point set distribution map, compares the angle difference with the neighboring direction, selects the segments with consistent direction to form a trajectory, and generates a continuous direction offset morphology map;

[0009] A response distribution balance module, based on the continuous directional offset morphology map, extracts brightness fluctuations in each area of the offset morphology map, calculates the difference in light intensity inside and outside the structure, determines whether the boundary difference continues to expand outward, and generates a regional response offset layer;

[0010] The structural state calibration module compares the boundary drift area in the response layer with the initial contour path based on the regional response offset layer, identifies the tensile and compression differences, and generates a local anomaly map of the chip structure.

[0011] As a further solution of the present invention, the edge alignment image layer includes the grayscale jump point position, the incident direction parameter, the image stable fluctuation area, and the brightness difference structure information; the abnormal reflection point set distribution map includes the brightness trajectory break position, the abnormal reflection intensity point, and the neighborhood reflection abnormal diffusion area; the continuous direction offset morphology map includes the brightness trend path, the angle excess fragment, and the continuous offset direction line segment; the regional response offset layer includes the boundary response extension area, the illumination intensity difference area, and the brightness fluctuation amplitude map; the chip structure local abnormality map includes the stretching abnormality area, the compression abnormality area, and the boundary contour drift area.

[0012] As a further solution of the present invention, the structural perspective control module includes:

[0013] The edge extraction submodule is used to obtain the image area where the edge diffraction structure on the silicon photonic chip is located, extract the position difference of the horizontal and vertical image grayscale values, calculate the position points in the grayscale difference grid that are greater than the brightness jump boundary threshold, and construct the edge path line to obtain the jump boundary difference coefficient;

[0014] The trend recognition submodule extracts three consecutive grayscale change points based on the jump boundary difference coefficient, calculates the grayscale change direction index, selects the point group with the same direction to construct the brightness change trend path, and obtains the grayscale change trend value;

[0015] The direction adjustment submodule constructs a structural cutting plane and extracts the incident direction according to the grayscale change trend value, records the regional grayscale fluctuation range in continuous frames, gradually rotates the acquisition direction and compares the grayscale fluctuation with the set reference fluctuation range, selects the acquisition angle with the smallest grayscale fluctuation and establishes an image layer to obtain an edge-aligned image layer.

[0016] As a further solution of the present invention, the grayscale change direction index calculation formula is specifically:

[0017] ;

[0018] in, Represents the grayscale change direction indicator, represents the sudden boundary strength coefficient, Representative The original grayscale value of each point, , represents the position weight distribution factor, Defined as when Time Otherwise, take , represents the direction consistency weight, represents the grayscale distribution normalization factor, Represents the denominator offset constant.

[0019] As a further solution of the present invention, the reflection anomaly capture module includes:

[0020] An illumination recording submodule records uniform illumination of the illuminated area based on the edge alignment image layer, collects surface reflection intensity values within the illuminated area, records the brightness value of each point in the plane layer and arranges them in order according to the layer coordinates, constructs a brightness extension trajectory sequence based on the grayscale values of adjacent points, and generates a brightness extension trajectory value;

[0021] The track break submodule calls the brightness extension track value to determine whether there is a grayscale mutation position in the track layer, detects whether the grayscale difference between the points before and after the mutation is greater than the brightness continuity threshold, records the coordinates of the break point, marks the break position to form a track gap layer, and obtains the brightness track break point set;

[0022] The anomaly screening submodule screens the reflection area around the breakpoint according to the brightness trajectory breakpoint set, compares the brightness diffusion amplitude of the breakpoint with the average diffusion amplitude in the neighborhood, screens the breakpoints whose brightness diffusion amplitude is greater than the diffusion intensity baseline value, and marks the layer position coordinates to generate a distribution map of the abnormal reflection point set.

[0023] As a further solution of the present invention, the direction misalignment verification module includes:

[0024] The direction extraction submodule extracts the brightness direction according to the layer arrangement order of the abnormal reflection point set distribution map, records the direction of change of the brightness value of adjacent pixels, constructs a brightness trend sequence based on the point sequence, and eliminates segments with brightness value fluctuation amplitudes less than the brightness trend fluctuation threshold to obtain the brightness trend sequence value;

[0025] An angle judgment submodule, which calls the brightness trend sequence value to obtain the brightness extension direction of the adjacent area and performs angle measurement, calculates the angle value between the adjacent trend and the extension direction, compares the angle difference with the set offset reference angle threshold, marks the trend segment position exceeding the threshold, and generates the direction angle deviation;

[0026] The trajectory construction submodule selects the direction segments whose angle differences are continuously less than the offset reference angle threshold according to the direction angle deviation, extracts the layer coordinates as connecting nodes, connects the continuous segments to form a trajectory structure line, and superimposes it on the original layer to obtain a continuous direction offset morphology map.

[0027] As a further solution of the present invention, the calculation formula for the angle between the adjacent directions and the extension direction is specifically:

[0028] ;

[0029] in, Representative The direction of the paragraph The angle between adjacent strikes and extension directions, Representative Section 1 The direction angle of the position point, Representative The sum of the brightness values of all points in the segment, Representative The strike angle of the midpoint of the segment, Representative The direction angles of adjacent regions, Representative The total number of position points involved in the calculation in the segment, Representative The average brightness value of adjacent areas.

[0030] As a further solution of the present invention, the response distribution balancing module includes:

[0031] The fluctuation extraction submodule extracts the time series layer of the brightness value in the region based on the continuous directional offset morphology map, records the brightness variation range of the pixels in the structure region and the background region, eliminates the data segments whose variation range is lower than the brightness fluctuation threshold, constructs the brightness fluctuation region distribution map, and generates the regional brightness fluctuation interval value;

[0032] The intensity difference submodule uses the regional brightness fluctuation interval value to extract the illumination intensity range of the two regions inside and outside the structure boundary, calculates the difference in the grayscale values of the two regions, and samples the brightness difference sequence outward from the boundary to obtain the grayscale expansion relationship of consecutive pixel positions to obtain the illumination difference coefficient of the structure boundary;

[0033] The boundary diffusion submodule determines the illumination difference change trend in the boundary direction based on the structural boundary illumination difference coefficient, makes a consistency judgment on the difference growth direction at multiple adjacent boundary points, screens the area that continues to expand outward, and overlays the original layer to obtain the regional response offset layer.

[0034] As a further solution of the present invention, the structural state calibration module includes:

[0035] The boundary extraction submodule extracts the coordinate sequence in the regional boundary layer based on the boundary drift area in the regional response offset layer and constructs the current contour path sequence, simultaneously extracts the boundary path of the corresponding area in the initial contour image, records the point mapping relationship between the two paths in the layer coordinate system, and obtains the regional boundary path control group;

[0036] The contour comparison submodule calls the regional boundary path control group, calculates the displacement direction and amplitude indicators between the current path and the initial path, determines whether there is an outward stretching and inward compression trend in the path direction, and selects the boundary segments whose direction difference is greater than the contour deformation judgment threshold to obtain the contour difference offset coefficient;

[0037] The anomaly marking submodule locates the regional segments where the contour trend differences are concentrated according to the contour difference offset coefficient, screens the contour blocks whose continuous trend differences exceed the offset judgment interval, marks and overlays the corresponding layer areas, and obtains the local anomaly map of the chip structure.

[0038] As a further solution of the present invention, the calculation formula for the displacement direction and amplitude index between the current path and the initial path is specifically:

[0039] ;

[0040] in, Represents the displacement direction and amplitude index between the current path and the initial path, Represents the current path Points, Represents the initial path Points, Representative The direction difference of the points, The total number of representative points.

[0041] Compared with the prior art, the advantages and positive effects of the present invention are:

[0042] In the present invention, by extracting grayscale differences and continuous brightness change points in the image, combined with jump boundary recognition and cutting plane construction, accurate positioning of microstructure mutation areas and extraction of incident directions are achieved, the acquisition direction is rotated to suppress grayscale fluctuations, the optical stability of the edge alignment layer is improved, the brightness expansion trajectory is collected and the fracture position is identified, and the reflection anomaly points are screened in combination with the diffusion amplitude difference, which effectively enhances the recognition ability of inhomogeneous reflections, arranges the brightness direction and calculates the angle, extracts the directional continuous offset trajectory, realizes dynamic modeling of directional deviation, calculates the illumination difference inside and outside the structure and compares the contour path, and improves the discrimination accuracy of boundary response evolution and abnormal contour distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0044] Figure 1 It is a system flow chart of the present invention.

[0045] Figure 2 It is a submodule flow chart of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0047] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations, or explanations. Any embodiment or design described as an "exemplary" in the present invention should not be interpreted as being preferred or advantageous over other embodiments or designs. Rather, the use of the word "exemplary" is intended to present concepts in a concrete manner. Furthermore, in the embodiments of the present invention, "and / or" can mean both or either of the two.

[0048] In the embodiments of the present invention, the terms "image" and "picture" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same. The terms "of," "corresponding," and "corresponding" may sometimes be used interchangeably. It should be noted that, when the distinction is not emphasized, the meanings they convey are the same.

[0049] In the embodiments of the present invention, sometimes a subscript such as W1 may be written as a non-subscript such as W1. When the difference is not emphasized, the meanings to be expressed are the same.

[0050] In order to make the technical problems, technical solutions and advantages to be solved by the present invention clearer, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0051] See also Figure 1 and Figure 2 , a silicon photonic chip detection system, comprising:

[0052] The structural perspective control module is used to obtain the image area where the edge diffraction structure on the silicon photonic chip is located, extract the position difference of the horizontal and vertical image grayscale values, and extract three consecutive brightness change points, calculate the difference in the change trend, determine whether there is a sharp turning area based on the set brightness jump boundary, construct the structural cutting plane and extract the position incident direction, record the grayscale fluctuation range of the area in consecutive frames, and gradually rotate the acquisition direction until the fluctuation is lower than the reference fluctuation value to generate the edge alignment image layer;

[0053] The reflection anomaly capture module records uniform illumination of the illuminated area based on edge alignment image layers, collects the brightness expansion trajectory of the surface reflection intensity within the plane, determines whether the brightness trajectory is broken in the layer and marks the broken parts, selects those with more drastic changes in brightness diffusion amplitude compared to the neighboring reflection areas, and generates a distribution map of abnormal reflection points.

[0054] The directional misalignment verification module extracts the brightness direction based on the distribution map of the abnormal reflection point set, calls the brightness extension direction of the adjacent area for angle comparison, calculates the included angle, and determines whether the angle difference exceeds the offset reference value. It selects segments with consistent directions to form an offset trajectory and generates a continuous directional offset morphology map.

[0055] The response distribution balance module extracts the brightness fluctuation regional distribution of each area based on the continuous direction offset morphology map, extracts the range of light intensity variation inside and outside the structure, performs difference calculation processing, determines whether the difference continues to expand outward in the boundary area, and generates a regional response offset layer;

[0056] The structural state calibration module compares the boundary drift area in the regional response offset layer to see whether the current regional contour direction is consistent with the corresponding boundary path of the initial contour image, screens contour blocks with differences in tension and compression directions, and generates a local anomaly map of the chip structure.

[0057] The edge alignment image layer includes the grayscale jump point location, incident direction parameters, image stable fluctuation area, and brightness difference structure information. The abnormal reflection point set distribution map includes the brightness trajectory break location, abnormal reflection intensity point, and neighborhood reflection abnormal diffusion area. The continuous direction offset morphology map includes the brightness trend path, angle limit fragment, and continuous offset direction line segment. The regional response offset layer includes the boundary response extension area, illumination intensity difference area, and brightness fluctuation amplitude map. The chip structure local anomaly map includes the stretching anomaly area, the compression anomaly area, and the boundary contour drift area.

[0058] See also Figure 1 and Figure 2 , the structural perspective control module includes:

[0059] The edge extraction submodule is used to obtain the image area where the edge diffraction structure on the silicon photonic chip is located, extract the position difference of the horizontal and vertical image grayscale values, calculate the position points in the grayscale difference grid that are greater than the brightness jump boundary threshold, and construct the edge path line to obtain the jump boundary difference coefficient.

[0060] First, perform image denoising operations, such as using a 3×3 median filter to remove scattered noise, and screen out image blocks with an average grayscale value of less than 50 in the image grayscale range of 0 to 255, and determine them as areas where edge diffraction structures may exist. For example, in an image, there is an area with an average grayscale value of 38, then add this area to the candidate set, and then use a 3×3 window to traverse the candidate area horizontally and vertically, perform differential calculations on the grayscale values of adjacent pixels, and extract the horizontal and vertical image grayscale value position differences. For example, the difference between point (i, j) and (i+1, j) is 12, and the difference with (i, j+1) is 35, forming a grayscale difference set in two directions. The above set is expanded into an image grid, and then a brightness jump boundary threshold is set, which is based on the image noise level. For example, when the image standard deviation is less than 10, the threshold is set to 25; when the image standard deviation is greater than 10 and less than 20, it is set to 30; when it is greater than 20, it is set to 35. In this example, the image standard deviation is 14, and the threshold is set to 30. Based on this value, pixels with grayscale difference greater than 30 are selected as jump points. The jump points are connected to edge paths according to the distance less than 2 pixels between each other. The connection between paths is determined by judging the coordinate distance point by point. For example, if the distance between points (i, j) and (i+1, j+1) is 1.4 and less than 2, they are connected to form multiple edge paths. Finally, the one with the largest grayscale jump intensity is selected in the path. For example, the difference on the path is 32, 35, 40, 45, and 38, a total of 7 points. The average difference is 38.1, which is the jump boundary difference coefficient.

[0061] The trend recognition submodule extracts three consecutive grayscale change points based on the jump boundary difference coefficient, calculates the grayscale change direction index, filters the point group with consistent direction to construct the brightness change trend path, and obtains the grayscale change trend value.

[0062] The calculation formula of the grayscale change direction index is as follows:

[0063] ;

[0064] in, Represents the grayscale change direction indicator, represents the sudden boundary strength coefficient, Representative The original grayscale value of each point, , represents the position weight distribution factor, Defined as when Time Otherwise, take ( is the initial reference grayscale), represents the direction consistency weight, represents the grayscale distribution normalization factor, Represents the denominator offset constant.

[0065] Parameter definition and value source:

[0066] (Jump boundary strength coefficient): It is set by monitoring the output amplitude of the image edge detection filter and is taken as This coefficient is based on the ratio of the high threshold to the low threshold in Canny edge detection, and the reasonable range is , refer to actual image processing standards.

[0067] (Point gray value): directly collected by the image sensor, the value is , , (8-bit grayscale range ).

[0068] (Position weight allocation factor): Based on the statistical analysis of the impact of spatial location on trends, the weight is , , The weight value is based on the demand for suppressing outliers, and the middle point has a higher weight. The total is .

[0069] (Direction symbol): When Time output , otherwise output Initial reference grayscale , calculated from the grayscale mean of the previous frame image.

[0070] (Directional consistency weight): The statistical significance of the product of directional signs is set as , the balance between the symbol product and the noise margin in the reference direction consistency detection, the reasonable range .

[0071] (Normalization factor): Based on the image grayscale dynamic range adjustment, take , calculated by the gray value square sum and equipment sensitivity calibration data, the reasonable range .

[0072] (Denominator offset constant): A fixed constant introduced to prevent the denominator from being zero. , which meets the requirement of minimizing the denominator.

[0073] Calculation examples and calculation process, molecular calculation:

[0074] ;

[0075] Directional signed product:

[0076] ;

[0077] Direction item: ;

[0078] Numerator sum: ;

[0079] Denominator calculation:

[0080] ;

[0081] Sum of denominators: ;

[0082] Final result:

[0083] .

[0084] Result interpretation:

[0085] The results show that the directional consistency index , the value reflects the consistency strength of the grayscale change direction. When , the directional consistency of the three points is determined to meet the threshold condition, and the brightness change trend path construction phase begins. The product of the directional signs in the numerator is negative, indicating that the grayscale of the third point decreases relative to the previous point, but the weighted grayscale accumulation term dominates the result, so it still passes the screening.

[0086] The direction adjustment submodule constructs a structural cutting plane and extracts the incident direction based on the grayscale change trend value, records the regional grayscale fluctuation range in continuous frames, gradually rotates the acquisition direction and compares the grayscale fluctuation with the set reference fluctuation range, selects the acquisition angle with the smallest grayscale fluctuation and establishes an image layer to obtain the edge-aligned image layer.

[0087] According to the grayscale change trend value, the structural tangent plane is constructed and the incident direction is extracted. First, the starting point and end point coordinates of the trend path are taken to establish a direction vector. For example, the starting point is (20, 30) and the end point is (100, 60). The main direction of the path is determined to be from the upper left to the lower right. Based on this direction, a set of incident directions perpendicular to it is constructed. A direction is taken every 5° from 0° to 180°, and a total of 36 incident angles are obtained. Then, the grayscale average value of the area in 10 consecutive frames of images is collected in each direction. For example, in the 60° direction, the grayscales of the 10 frames of images are 70, 71, 69, 73, 68, 70, 72, 70, 69, and 71, respectively. The grayscale fluctuation range is the maximum. The maximum value minus the minimum value, that is, 73 minus 68 equals 5, and the grayscale fluctuation value in this direction is recorded. Then the grayscale fluctuation reference interval is set to [5,15]. This setting is given according to the stability tolerance allowed by the system. When the image contrast is stable, it is set to 5, and expanded to 15 in dynamic contrast scenarios. In this setting, if the grayscale fluctuation of the direction is within this interval, it is recorded as a candidate direction. Finally, the angle with the smallest grayscale fluctuation is selected from all candidate directions as the optimal acquisition direction. For example, the fluctuation in the 70° direction is 4.8, which is not within the set interval and is excluded. The fluctuation in the 60° direction is 5, which is accepted and the minimum value. In this way, this direction is selected to establish an image layer, forming an edge-aligned image layer.

[0088] See also Figure 1 and Figure 2 , the reflection exception capture module includes:

[0089] The illumination recording submodule records the uniform illumination of the illuminated area based on the edge alignment image layer, collects the surface reflection intensity value within the illuminated area, records the brightness value of each point in the plane layer and arranges them in order according to the layer coordinates, constructs a brightness extension trajectory sequence based on the grayscale values of adjacent points, and generates a brightness extension trajectory value.

[0090] First, determine the target area in the layer where illumination measurement needs to be performed. The area is set as a fixed rectangular area by aligning the edge with the outline bounding box of the structure jump area in the layer. For example, the area is set to 60 pixels × 80 pixels. Illumination is collected point by point within this area. Each pixel receives a uniform light source from a standard angle of incidence, and the brightness value after reflection is collected. The brightness value ranges from 0 to 255 grayscale levels. After recording the brightness value of each point, all brightness values are arranged into a one-dimensional sequence from left to right and from top to bottom in the layer. The numbering method is P(i,j), where i is the row number and j is the column number. For example, the brightness of point (10,15) is 142, then the brightness of point (10,15) is 142. At the 595th position in the sequence, the brightness expansion trajectory is constructed. That is, the brightness difference between the current point and its four adjacent points (up, down, left, and right) is used as a reference, and the brightness expansion direction is recorded in sequence. In an actual example, if the brightness of point (10,15) is 142, the brightness of the point to its right is 145, the point below is 140, the left is 138, and the point above is 143, then it is expanded to the right and upward in the direction of the difference to form a brightness expansion trajectory. The trajectory prioritizes the direction according to the order of the maximum brightness difference. In this way, a continuous brightness jump path is generated, and each jump step value is recorded in sequence to form a brightness expansion trajectory value sequence, and finally the brightness expansion trajectory value of the area is obtained.

[0091] The trajectory break submodule calls the brightness extension trajectory value to determine whether there is a grayscale mutation position in the trajectory layer. It detects whether the grayscale difference between the points before and after the mutation is greater than the brightness continuity threshold, records the coordinates of the break point, marks the break position to form a trajectory gap layer, and obtains the brightness trajectory break point set.

[0092] Call the brightness extension trajectory value to determine whether there is a grayscale mutation position in the trajectory layer. In the brightness extension trajectory value sequence obtained in the previous step, calculate the brightness difference between each two adjacent points in turn, and set the brightness continuity threshold T_cont to determine the mutation position. The threshold is set according to the overall grayscale variation range of the image. When the image grayscale standard deviation is less than 20, set T_cont to 12, when the standard deviation is between 20 and 35, set it to 18, and when it exceeds 35, set it to 22. In an example, the image area standard deviation is 28, then set T_cont to 18. When the brightness difference between any two points in the trajectory is greater than this value, it is considered a grayscale mutation point. For example, if there are continuous bright points in the trajectory, The degrees are 142, 145, 147, and 173. The difference from 145 to 147 is 2, which is not a mutation. The difference from 147 to 173 is 26, which is greater than the threshold of 18 and is determined to be a mutation point. The layer coordinate position of this point is recorded. For example, this mutation point is located at (11, 17), which is marked as a break point. The break point is separated from the trajectory to generate a break gap. After performing the same operation on all trajectories in the entire layer, the coordinate set of all break points is obtained, and each break point position is marked on the layer using an image marker point, for example, marking the point with a red pixel value. Finally, all the marked points are summarized to form a trajectory gap layer, and then the brightness trajectory break point set is obtained.

[0093] The anomaly screening submodule screens the reflection area around the breakpoint according to the brightness trajectory breakpoint set, compares the brightness diffusion amplitude of the breakpoint with the average diffusion amplitude in the neighborhood, screens the breakpoints whose brightness diffusion amplitude is greater than the diffusion intensity baseline value, and marks the layer position coordinates to generate a distribution map of the abnormal reflection point set.

[0094] According to the brightness trajectory breakpoint bit set, the reflection area around the breakpoint is screened. First, each breakpoint is selected in the trajectory breakpoint bit set, and a neighborhood window with a radius of 3 pixels is established with it as the center, which contains 7×7 pixels and a total of 49 pixels. The brightness values of all pixels in the neighborhood are recorded, and the average diffusion amplitude of the area is calculated, that is, the difference between the maximum brightness and the minimum brightness in the neighborhood. For example, the breakpoint is located at (12,18), the maximum brightness in its neighborhood is 190, and the minimum brightness is 130, so the average diffusion amplitude is 60. At the same time, the difference between the two values of the brightness mutation before and after the breakpoint itself in the trajectory is extracted as the brightness diffusion amplitude of the point, such as before and after the breakpoint The brightness values are 150 and 185 respectively, so the diffusion amplitude is 35. Then compare the brightness diffusion amplitude of the breakpoint with the average diffusion amplitude of the neighborhood. If the value is greater than the set diffusion intensity reference value, the point is marked as an abnormal reflection point. The diffusion intensity reference value is set according to the system allowable error and is set to 30 in most applications. When the diffusion amplitude of the breakpoint is greater than 30, it is judged to be abnormal. For example, the diffusion amplitude of the aforementioned point is 35, which is greater than 30. Record its coordinates (12, 18) and mark the corresponding position in the layer. Then draw a layer distribution map for all abnormal reflection points in coordinate order, and uniformly represent their points with green highlighted pixels. Finally, generate the distribution map of the abnormal reflection point set.

[0095] See also Figure 1 and Figure 2 , the direction misalignment verification module includes:

[0096] The direction extraction submodule extracts the brightness trend according to the layer arrangement order based on the distribution map of the abnormal reflection point set, records the direction of change of the brightness values of adjacent pixels, constructs a brightness trend sequence based on the point sequence, and eliminates segments with brightness value fluctuations less than the brightness trend fluctuation threshold to obtain the brightness trend sequence value.

[0097] First, in the layer, the abnormal reflection points are traversed in the order of coordinates from left to right horizontally and from top to bottom vertically. The initial brightness change path is constructed by recording the pixel coordinate difference and the corresponding brightness value difference between each pair of adjacent points. Between each two consecutive points, a unidirectional trend segment is formed according to the direction of increase or decrease of the brightness value (increase is recorded as "+", decrease is recorded as "-", and unchanged is recorded as "0"), and the direction of the coordinate vector is recorded as the trend direction. For example, the brightness of point A is 145, the brightness of point B is 139, the brightness difference is -6, and the direction vector is from (20,30) to (21,31). Then the trend is recorded as the lower right direction and the brightness decreases. All such directions and brightness relationships are continuously recorded in the layer to obtain the initial sequence of brightness trends. Subsequently, a brightness trend fluctuation threshold is set to filter out segments with small changes. The threshold is set according to the overall brightness change of the layer. When the brightness standard deviation is less than 20, it is set to 5; when it is between 20 and 35, it is set to 8; when it is greater than 35, it is set to 10. In this example, if the brightness standard deviation is 24, the fluctuation threshold is set to 8, and all segments with a brightness change of less than 8 are eliminated. For example, if the brightness of a segment changes from 152 to 146, and the change is only 6, which is less than the threshold 8, then this segment is eliminated and other segments with larger changes are retained. Finally, the brightness trend sequence value is sorted out.

[0098] The angle judgment submodule calls the brightness direction sequence value to obtain the brightness extension direction of the adjacent area and performs angle measurement, calculates the angle value between the adjacent direction and the extension direction, compares the angle difference with the set offset reference angle threshold, marks the direction segment position that exceeds the threshold, and generates the direction angle deviation.

[0099] The calculation formula for the angle between adjacent strikes and extension directions is:

[0100] ;

[0101] in, Representative No. The direction of the paragraph The angle between adjacent strikes and extension directions, Representative Section 1 The direction angle of the position point, Representative The sum of the brightness values of all points in the segment, Representative The strike angle of the midpoint of the segment, Representative The direction angles of adjacent regions, Representative The total number of position points involved in the calculation in the segment, Representative The average brightness value of adjacent areas.

[0102] Parameter acquisition and value setting:

[0103] : No. Section 1 The angle of travel at a point. This is calculated by extracting the coordinates of the two endpoints of a line segment using image processing techniques such as edge detection and Hough transform. Angle values typically range from 0° to 180°.

[0104] : No. The sum of the brightness values of all points in the segment. The brightness value (grayscale value) of all pixels in the segment is accumulated. The brightness value range is 0 to 255.

[0105] : No. The strike angle of the midpoint of the segment. The center point of the segment is obtained by calculating its strike angle.

[0106] : No. The direction angle between adjacent regions. This is obtained by analyzing the structural characteristics of the adjacent regions and calculating their main directions. The angle value usually ranges from 0° to 180°.

[0107] : No. The total number of position points involved in the calculation in the segment. The number of pixels in the segment is obtained.

[0108] : No. The average value of the brightness value in the adjacent area. The brightness value of all pixels in the adjacent area is calculated by averaging their brightness values. The brightness value range is 0 to 255.

[0109] Specific numerical settings:

[0110] : [30°,35°,40°,45°,50°];

[0111] : set to 1000;

[0112] : 45°;

[0113] : 40°;

[0114] :5;

[0115] :200.

[0116] Formula calculation process:

[0117] calculate :

[0118] ;

[0119] calculate ;

[0120] calculate :

[0121] ;

[0122] Calculate the numerator:

[0123] ;

[0124] Calculate the denominator:

[0125] ;

[0126] Final calculation :

[0127] .

[0128] This result shows that the The direction of the paragraph The angle between the extension directions of adjacent regions is approximately 5.23. This value is used to determine whether the angle difference exceeds the set offset reference angle threshold, and then mark the position of the direction segment that exceeds the threshold and generate the direction angle deviation.

[0129] The trajectory construction submodule selects the strike segments whose angle difference is continuously less than the offset reference angle threshold according to the direction angle deviation, extracts the layer coordinates as connecting nodes, connects the continuous segments to form a trajectory structure line and superimposes it on the original layer to obtain a continuous direction offset morphology map.

[0130] According to the direction angle deviation, the trend segments with angle differences continuously less than the offset reference angle threshold are screened. First, the segments with angle values exceeding the reference threshold are screened out from all the trend segments marked in the previous stage, and the segments with continuous angle differences less than the threshold are retained as candidate trajectory segments. In the screening process, each segment without a marked deviation is used as the starting point, and the next segment is continuously searched backward to see if it is also unmarked. If three or more consecutive segments all meet the angle value lower than the threshold, they are constructed as a connectable trajectory segment. For example, the angles of segments a, b, and c are 12 degrees, 14 degrees, and 17 degrees respectively, which are all less than the threshold of 20 degrees. It is selected as the basis for trajectory connection, and then the coordinate position of each segment in the layer is extracted from its end point. For example, the end point of segment a is (30,45) and the starting point of segment b is (30,45). They are directly connected to construct a trajectory connection node sequence. The segments are connected in the order of layer coordinates to form a trajectory structure line. The trajectory structure line is superimposed on the original layer in the form of a path with a single pixel width to form a complete continuous direction trajectory. In the layer image, a specific grayscale value such as 200 is assigned to the connection path to distinguish the grayscale information of the original structure layer, thereby realizing visual superposition between layers and finally obtaining a continuous direction offset morphology map.

[0131] See also Figure 1 and Figure 2 , the response distribution balancing module includes:

[0132] The fluctuation extraction submodule extracts the time series layer of the brightness value in the area based on the continuous directional offset morphology map, records the brightness change range of the pixels in the structure area and the background area, eliminates the data segments whose change amplitude is lower than the brightness fluctuation threshold, constructs the brightness fluctuation area distribution map, and generates the regional brightness fluctuation interval value.

[0133] According to the continuous direction offset morphology map, the time series layer of the brightness value in the area is extracted. First, the sampling range is set for the structural area and the background area in the layer respectively. The structural area is composed of pixel points in a closed figure connected by continuous edges, and the background area is a ring area extending 5 pixels outward from the structural boundary. The system records the brightness values of all pixels in the two types of areas at each time sampling point (such as each frame of image) and constructs a time series grayscale vector for each point. In the actual scene, if the image period is 10 frames, the grayscale of point P in 10 frames is [152,154,156,153,155,158,160,162,159,157]. The difference between the maximum and minimum values of the sequence is calculated to obtain the brightness change amplitude. The value is 160-152=8. Then the brightness fluctuation threshold is set to filter out points with insufficient fluctuation. The threshold is set according to the background noise range. If the average noise amplitude of the image is less than 4, the threshold is set to 6; the noise amplitude is between 4 and 8, the threshold is set to 9; and the value is higher than 8, the threshold is set to 12. In the current sample, the noise amplitude is 6, and the fluctuation threshold is set to 9. The brightness fluctuation range of each pixel is compared. If it is less than 9, the point data segment is eliminated. For example, the fluctuation of the above point P is 8, and it is removed if it is lower than 9. After performing the same judgment on all pixels, the remaining high fluctuation points are located and marked, and then restored to the layer according to their spatial position to form a brightness fluctuation area distribution map. The maximum, minimum and average fluctuation amplitudes in each area are calculated and summarized as the regional brightness fluctuation interval value.

[0134] The intensity difference submodule calls the regional brightness fluctuation interval value to extract the illumination intensity range of the two areas inside and outside the structure boundary, calculates the difference in the grayscale values of the two areas, and samples the brightness difference sequence outward from the boundary to obtain the grayscale expansion relationship of consecutive pixel positions to obtain the illumination difference coefficient of the structure boundary.

[0135] Call the regional brightness fluctuation interval value to extract the illumination intensity range of the two areas inside and outside the structure boundary. First, identify the structure boundary line in the layer and divide the pixels into inside the boundary (inside the structure) and outside the boundary (inside the background band). In each type of area, the maximum and minimum brightness values of all retained points are counted to form the illumination intensity range. For example, the brightness of the points inside the structure is between [145,168], and the brightness outside the structure is between [110,132]. Then, the brightness difference between the areas is calculated as the mean inside the structure minus the mean outside the structure. If the mean inside the structure is 156 and the mean outside the structure is 121, the difference is 35. Then, On the structural boundary, sampling is performed at every other boundary point of 1 pixel, and the brightness values of consecutive pixels outside the boundary are extended 3 pixels outward. The difference between the brightness values and the structural boundary points is recorded to construct a brightness difference sequence. For example, if the brightness of the boundary point is 155, and the brightness of the three outward points is 140, 130, and 125, respectively, then the difference sequence is 15, 25, and 30. This sequence is used to reflect the expansion trend of the grayscale change at the boundary. Finally, the maximum value of the continuous brightness difference and its relative position are recorded, and the average difference on each boundary line is combined to form the structural boundary illumination difference coefficient. This coefficient is the regional response intensity reflection of the continuous grayscale difference.

[0136] The boundary diffusion submodule determines the illumination difference change trend in the boundary direction based on the structural boundary illumination difference coefficient, makes a consistency judgment on the difference growth direction at multiple adjacent boundary points, screens the area that continues to expand outward, and overlays the original layer to obtain the regional response offset layer.

[0137] The illumination difference change trend in the boundary direction is determined according to the illumination difference coefficient of the structure boundary. The continuous difference increase and decrease change direction is extracted from each point on the structure boundary and its outward-extending brightness difference sequence. For example, if the brightness difference of one group of points is 12, 17, and 22, it is recorded as a continuous increase. If the brightness difference of another group is 22, 19, and 15, it is recorded as a continuous decrease. The difference directions of all boundary points are compared, and the consistency judgment range is set to at least 3 consecutive points with the same direction. That is, if the boundary shows continuous growth or decrease from point A to point C, it is considered consistent. All point segments that meet the consistency conditions are selected as candidate expansion areas. Subsequently, the outward expansion boundary coordinates corresponding to the candidate area are marked and the layer pixel structure is reconstructed. The area identifier is superimposed on the original layer, and the original pixel grayscale is uniformly covered with a highlight grayscale value such as 200 or a specified label value to form a new layer, and finally the regional response offset layer is obtained.

[0138] See also Figure 1 and Figure 2 , the structural state calibration module includes:

[0139] The boundary extraction submodule extracts the coordinate sequence in the regional boundary layer and constructs the current contour path sequence based on the boundary drift area in the regional response offset layer. It also simultaneously extracts the boundary path of the corresponding area in the initial contour image, records the point mapping relationship between the two paths in the layer coordinate system, and obtains the regional boundary path control group.

[0140] Based on the boundary drift area in the regional response offset layer, the boundary area in the layer that is judged to be continuously expanding with the brightness difference change is first extracted. According to the spatial coordinate arrangement order of each pixel point in the layer, the boundary points are extracted from left to right and from top to bottom to construct a continuous closed boundary path. In a typical example, the drift area is an approximately elliptical structure, and its boundary point sequence is extracted as: (45,30), (45,31), (46,32)...(60,50). This boundary point sequence is recorded as the current contour path sequence, and the boundary path of the corresponding area of the chip in the initial state image is extracted at the same time. By comparing the coordinates of the center point of the area with the area label number, alignment recognition is performed to find the same area in the original state. The boundary point sequence of the initial layer is, for example: (45,30), (45,31), (46,31)…(59,49). Each point in the current contour path is mapped one-to-one with the initial contour path. A point pairing relationship is established between each pair of points through relative sequence correspondence. If a region has 40 boundary points extracted in the current layer and 38 points in the initial layer, the shorter path is interpolated and completed. The average position point is inserted to align the number of points to ensure the complete pairing relationship. A control group for the regional boundary path is formed, and a mapping relationship table for each point is recorded. For example, the 10th point in the current layer corresponds to the 10th point in the initial layer. These mapping relationships are used for subsequent path displacement and deformation analysis.

[0141] The contour comparison submodule calls the regional boundary path control group to calculate the displacement direction and amplitude indicators between the current path and the initial path, determine whether there is an outward stretching and inward compression trend in the path direction, and screen the boundary segments whose direction difference is greater than the contour deformation judgment threshold to obtain the contour difference offset coefficient.

[0142] The calculation formula for the displacement direction and amplitude index between the current path and the initial path is as follows:

[0143] ;

[0144] in, Represents the displacement direction and amplitude index between the current path and the initial path, Represents the current path Points, Represents the initial path Points, Representative The direction difference of the points, The total number of representative points.

[0145] The path point coordinate data is obtained in real time by a high-precision image contour extraction module. The image resolution is 1920×1080. The path contour boundary is obtained by a binary edge detection algorithm. The path points are sampled every 10 pixels. 100 points in a path are selected for comparison, that is, .

[0146] Parameter description and quantification basis:

[0147] :Current path The coordinate value of the point, in pixels, is obtained by collecting the position numbered i in the contour boundary image;

[0148] :Initial path The coordinate value of each point, in pixels, is obtained through the boundary point sequence recorded by the standard initial path set by the system;

[0149] : Direction difference, which represents the absolute value of the angle between the current point's direction and the initial point's direction, in degrees. It is obtained by calculating the tangent direction of the line connecting two consecutive points and comparing them. The direction angle is quantified clockwise with north as 0 degrees.

[0150] Direction By formula:

[0151] ;

[0152] in, is the current point, is its next point. The direction difference is calculated as:

[0153] ;

[0154] The direction value change is based on ±15 degrees. Boundary segments with an error greater than this value will be considered deformed.

[0155] The actual numerical example calculation process samples 3 groups of example path points for demonstration:

[0156] Point 1: Current point , initial point ;

[0157] Point 2: Current point , initial point ;

[0158] Point 3: Current point , initial point .

[0159] Calculate the displacement modulus (Euclidean distance):

[0160] ;

[0161] ;

[0162] .

[0163] Direction angle calculation, assuming the subsequent points are as follows:

[0164] Follow-up to point 1:

[0165] , ;

[0166] Initial point follow-up:

[0167] , , direction difference ;

[0168] Similar calculations yield: , .

[0169] Substitute the data into the formula:

[0170] .

[0171] The results show that the average directional deviation coefficient between the current path and the initial path in the paragraph is 70.7, indicating that there is an obvious path deviation trend in the selected point samples. The product of the deviation degree and the direction difference is enhanced, which comprehensively reflects the directional deviation intensity of the deformation boundary.

[0172] The anomaly marking submodule locates the regional segments where the contour trend differences are concentrated according to the contour difference offset coefficient, screens the contour blocks whose continuous trend differences exceed the offset judgment interval, marks and overlays the corresponding layer areas, and obtains the local anomaly map of the chip structure.

[0173] First, path segments with continuous large offsets are extracted from the contour path. The direction difference concentration area is determined by statistically analyzing the consistency of the offset direction of each segment with the direction of adjacent segments. In a typical example, if the offset direction of five consecutive path segments is outward, and the offset amplitude of each segment is 3.5, 3.7, 3.9, 4.2, and 4.0 pixels, all greater than the set threshold of 3 pixels, and the direction is continuous and consistent, they are determined to be segments with concentrated difference areas. Then, the offset judgment interval is set to "3 pixels starting and 3 or more segments appearing continuously" as a screening condition. Segments that meet the conditions are classified as abnormal contour blocks. The starting and ending coordinates of the contour block are recorded and mapped to the original layer structure. The minimum enveloping rectangular area corresponding to the block is extracted. For example, this area contains the layer range from (48, 32) to (55, 40). This range is defined as the marker block area in the image data and assigned a uniform highlight grayscale marker value, such as 230. The original pixel value is overwritten to complete the layer modification. The same operation is performed on all contour blocks that meet the screening conditions, completing the marking process of all contour abnormal areas, and finally forming a complete local anomaly map of the chip structure.

[0174] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in the present invention should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A silicon photonic chip detection system, characterized in that: include: The structural perspective control module is used to obtain the edge diffraction image area, extract the horizontal and vertical grayscale differences and three brightness change points, calculate the trend difference, determine the jump boundary, construct the cutting plane and the incident direction, record the grayscale fluctuations of consecutive frames, and rotate the acquisition direction until the fluctuation is lower than the reference value, thereby generating the edge alignment image layer; A reflection anomaly capture module, based on the edge alignment image layer, records illumination uniformity, collects reflection brightness traces, identifies and marks breaks, filters areas with more intense diffusion than neighboring areas, and generates a distribution map of abnormal reflection points; The direction misalignment verification module extracts the brightness trend based on the abnormal reflection point set distribution map, compares the angle difference with the neighboring direction, selects the segments with consistent direction to form a trajectory, and generates a continuous direction offset morphology map; A response distribution balance module, based on the continuous directional offset morphology map, extracts brightness fluctuations in each area of the offset morphology map, calculates the difference in light intensity inside and outside the structure, determines whether the boundary difference continues to expand outward, and generates a regional response offset layer; The structural state calibration module compares the boundary drift area in the response layer with the initial contour path based on the regional response offset layer, identifies the tension and compression differences, and generates a local anomaly map of the chip structure.

2. The silicon photonic chip detection system according to claim 1, wherein: The edge alignment image layer includes the grayscale jump point position, incident direction parameters, image stability fluctuation area, and brightness difference structure information; the abnormal reflection point set distribution map includes the brightness trajectory break position, abnormal reflection intensity point, and neighborhood reflection abnormal diffusion area; the continuous direction offset morphology map includes the brightness trend path, angle limit fragment, and continuous offset direction line segment; the regional response offset layer includes the boundary response extension area, the illumination intensity difference area, and the brightness fluctuation amplitude map; the chip structure local abnormality map includes the stretching abnormality area, the compression abnormality area, and the boundary contour drift area.

3. The silicon photonic chip detection system according to claim 1, characterized in that: The structural perspective control module includes: The edge extraction submodule is used to obtain the image area where the edge diffraction structure on the silicon photonic chip is located, extract the position difference of the horizontal and vertical image grayscale values, calculate the position points in the grayscale difference grid that are greater than the brightness jump boundary threshold, and construct the edge path line to obtain the jump boundary difference coefficient; Perform differential calculation on the grayscale values of adjacent pixels, extract the position difference of the horizontal and vertical image grayscale values, form a grayscale difference set in two directions, and expand the set into an image grid; The trend recognition submodule extracts three consecutive grayscale change points based on the jump boundary difference coefficient, calculates the grayscale change direction index, selects the point group with the same direction to construct the brightness change trend path, and obtains the grayscale change trend value; The direction adjustment submodule constructs a structural cutting plane and extracts the incident direction according to the grayscale change trend value, records the regional grayscale fluctuation range in continuous frames, gradually rotates the acquisition direction and compares the grayscale fluctuation with the set reference fluctuation range, selects the acquisition angle with the smallest grayscale fluctuation and establishes an image layer to obtain an edge-aligned image layer.

4. The silicon photonic chip detection system according to claim 3, characterized in that: The grayscale change direction index calculation formula is specifically: Among them, H represents the grayscale change direction index, κ represents the sudden edge strength coefficient, I i Represents the original grayscale value of the i-th point, i = 1 or 2 or 3, w i represents the position weight distribution factor, Defined as when I i >I i-1 It takes +1 when , otherwise it takes -1, ν represents the direction consistency weight, ξ represents the grayscale distribution normalization factor, and ρ represents the denominator offset constant.

5. The silicon photonic chip detection system according to claim 3, characterized in that: The reflection anomaly capture module includes: An illumination recording submodule records uniform illumination of the illuminated area based on the edge alignment image layer, collects surface reflection intensity values within the illuminated area, records the brightness value of each point in the plane layer and arranges them in order according to the layer coordinates, constructs a brightness extension trajectory sequence based on the grayscale values of adjacent points, and generates a brightness extension trajectory value; The track break submodule calls the brightness extension track value to determine whether there is a grayscale mutation position in the track layer, detects whether the grayscale difference between the points before and after the mutation is greater than the brightness continuity threshold, records the coordinates of the break point, marks the break position to form a track gap layer, and obtains the brightness track break point set; The anomaly screening submodule screens the reflection area around the breakpoint according to the brightness trajectory breakpoint set, compares the brightness diffusion amplitude of the breakpoint with the average diffusion amplitude in the neighborhood, screens the breakpoints whose brightness diffusion amplitude is greater than the diffusion intensity benchmark value, and marks the layer position coordinates to generate a distribution map of the abnormal reflection point set; The difference between the two values of the brightness mutation before and after the breakpoint in the trajectory is extracted as the brightness diffusion amplitude of the point. The brightness diffusion amplitude of the breakpoint is compared with the average diffusion amplitude of the neighborhood. If it is greater than the set diffusion intensity benchmark value, the point is marked as an abnormal reflection point.

6. The silicon photonic chip detection system according to claim 5, characterized in that: The direction misalignment verification module includes: The direction extraction submodule extracts the brightness direction according to the layer arrangement order of the abnormal reflection point set distribution map, records the direction of change of the brightness value of adjacent pixels, constructs a brightness trend sequence based on the point sequence, and eliminates segments with brightness value fluctuation amplitudes less than the brightness trend fluctuation threshold to obtain the brightness trend sequence value; An angle judgment submodule, which calls the brightness trend sequence value to obtain the brightness extension direction of the adjacent area and performs angle measurement, calculates the angle value between the adjacent trend and the extension direction, compares the angle difference with the set offset reference angle threshold, marks the trend segment position exceeding the threshold, and generates the direction angle deviation; The trajectory construction submodule selects the direction segments whose angle differences are continuously less than the offset reference angle threshold according to the direction angle deviation, extracts the layer coordinates as connecting nodes, connects the continuous segments to form a trajectory structure line, and superimposes it on the original layer to obtain a continuous direction offset morphology map.

7. The silicon photonic chip detection system according to claim 6, characterized in that: The specific calculation formula for the angle between the adjacent strike direction and the extension direction is: Among them, θ y,j Represents the angle between the yth segment strike and the jth adjacent strike and extension direction, φ y,g Represents the strike angle of the g-th position point in the y-th segment, Represents the sum of the brightness values of all points in the y-th segment, Represents the direction angle of the midpoint of the y-th segment, represents the direction angle of the jth adjacent area, n represents the total number of position points involved in the calculation in the yth segment, Represents the average brightness value in the jth adjacent area.

8. The silicon photonic chip detection system according to claim 6, characterized in that: The response distribution balancing module includes: The fluctuation extraction submodule extracts the time series layer of the brightness value in the region based on the continuous directional offset morphology map, records the brightness variation range of the pixels in the structure region and the background region, eliminates the data segments whose variation range is lower than the brightness fluctuation threshold, constructs the brightness fluctuation region distribution map, and generates the regional brightness fluctuation interval value; The intensity difference submodule uses the regional brightness fluctuation interval value to extract the illumination intensity range of the two regions inside and outside the structure boundary, calculates the difference in the grayscale values of the two regions, and samples the brightness difference sequence outward from the boundary to obtain the grayscale expansion relationship of consecutive pixel positions to obtain the illumination difference coefficient of the structure boundary; The boundary diffusion submodule determines the illumination difference change trend in the boundary direction based on the structural boundary illumination difference coefficient, makes a consistency judgment on the difference growth direction at multiple adjacent boundary points, screens the area that continues to expand outward, and overlays the original layer to obtain the regional response offset layer.

9. The silicon photonic chip detection system according to claim 8, characterized in that: The structural state calibration module includes: The boundary extraction submodule extracts the coordinate sequence in the regional boundary layer based on the boundary drift area in the regional response offset layer and constructs the current contour path sequence, simultaneously extracts the boundary path of the corresponding area in the initial contour image, records the point mapping relationship between the two paths in the layer coordinate system, and obtains the regional boundary path control group; The contour comparison submodule calls the regional boundary path control group, calculates the displacement direction and amplitude indicators between the current path and the initial path, determines whether there is an outward stretching and inward compression trend in the path direction, and selects the boundary segments whose direction difference is greater than the contour deformation judgment threshold to obtain the contour difference offset coefficient; The anomaly marking submodule locates the regional segments where the contour trend differences are concentrated according to the contour difference offset coefficient, screens the contour blocks whose continuous trend differences exceed the offset judgment interval, marks and overlays the corresponding layer areas, and obtains the local anomaly map of the chip structure.

10. The silicon photonic chip detection system according to claim 9, characterized in that: The calculation formula for the displacement direction and amplitude index between the current path and the initial path is as follows: Among them, Δv represents the displacement direction and amplitude index between the current path and the initial path, P cz Represents the zth point of the current path, P oz represents the zth point of the initial path, d z Represents the direction difference of the z-th point, and m represents the total number of points.

Images