Automatic detection system for front windshield bracket based on AI visual identification

Abstract

The invention discloses an AI visual identification-based front windshield support automatic detection system, and relates to the technical field of visual detection, and the system comprises a conflict rechecking module which locates a structure conflict region in a structure semantic conflict set through conflict judgment, carries out the targeted visual analysis processing of the structure conflict region in a structure alignment image, and carries out the automatic detection of a front windshield support. Generating a rechecking evidence item; and the causal judgment module is used for reversely deducing and rechecking the evidence item through causal, obtaining a minimum failure structure set, automatically detecting and judging the front windshield bracket image according to the minimum failure structure set, and outputting a grading detection conclusion and a structure tracing identifier. According to the method, by constructing a contradiction contrast rule and synchronously analyzing the support or negative relationship between the evidence and the initial assertion, high-sensitivity and high-reliability detection of correlation defects and hidden conflicts is realized, and the analysis capability and decision interpretability of visual detection for coping with a complex failure mode are enhanced.

Description

Technical Field

[0001] This invention relates to the field of visual inspection technology, and in particular to an automated inspection system for windshield brackets based on AI visual recognition. Background Technology

[0002] In the field of industrial automation quality inspection, machine vision-based visual inspection has gradually replaced traditional manual visual inspection, becoming a key means to improve production efficiency and ensure product quality stability. Early methods mainly relied on template matching, edge detection, and feature engineering to judge the size, position, or simple appearance defects of workpieces through preset geometric parameter thresholds. With the popularization of deep learning, especially convolutional neural networks, algorithms based on object detection and instance segmentation have improved the recognition and positioning accuracy of complex contours and multi-category structural elements. They can learn feature representations from large amounts of data and automatically identify elements such as hole positions and weld points, realizing a leap from rule-driven to data-driven approaches.

[0003] However, existing technologies still have certain shortcomings. They are weak in detecting defects in complex structural relationships and assembly logic. Existing methods usually detect the size, position, or surface quality of each structural element in isolation, or can only perform simple pairwise distance comparisons. For complex defects involving topological relationships, spatial dependencies, and assembly sequence compliance among multiple elements, it is difficult to formally define and verify high-level constraints, let alone identify implicit functional defects caused by relationship violations. They also lack causal reasoning mechanisms in defect root cause localization and impact assessment. When anomalies are detected, existing technologies can usually only point out where the defects do not meet expectations, but cannot reason why they do not meet expectations, or which are the primary failure points and which are cascading consequences. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an automated inspection system for windshield brackets based on AI visual recognition to solve the problems of difficulty in semantic defect detection of complex assembly relationships and lack of root cause reasoning for defects.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: This invention provides an automated inspection system for windshield brackets based on AI visual recognition, comprising: The structural semantic module acquires images of the windshield bracket and defines constraints on the relationships between various structural elements in the windshield bracket images, generating a set of structural semantic constraints. The structural benchmark module performs structural benchmark reconstruction processing on the front windshield bracket image based on the structural semantic constraint set, generates a structural alignment image, identifies the structural relationships in the structural alignment image and performs relationship feasibility determination, and forms structural evidence items. The disproving assertion module performs structural state interpretation processing on structural evidence items based on the structural semantic constraint set, generates a set of structural semantic assertions, identifies structural semantic assertion combinations that cannot be simultaneously true in the set of structural semantic assertions, performs disproving consistency verification, and generates a set of structural semantic conflicts. The conflict review module locates structural conflict regions in the set of structural semantic conflicts through conflict adjudication, performs targeted visual analysis processing on the structural conflict regions in the structural alignment image, and generates review evidence items. The causal adjudication module verifies the evidence items by causal inference, obtains the minimum set of failed structures, automatically detects and adjudicates the front windshield bracket image based on the minimum set of failed structures, and outputs graded detection conclusions and structural traceability identifiers.

[0007] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket detection system of the present invention, the specific steps for generating the structural semantic constraint set are as follows: The windshield bracket image is input into a convolutional neural network to obtain candidate image regions. By analyzing the structural geometric representation of the candidate image regions, the corresponding structural elements are identified and labeled. By extracting the stable co-occurrence relationships and assemblable relationships among the structural elements in the windshield bracket image, relationship modeling and semantic constraint partitioning are performed on each structural element to generate a set of structural semantic constraints.

[0008] As a preferred embodiment of the AI-based visual recognition-based automated inspection system for windshield supports described in this invention, the specific steps for generating the structure-aligned image are as follows: Using the set of structural semantic constraints as relational constraints, the constraint participation degree of structural elements in the windshield bracket image is statistically analyzed to generate a candidate list of structural references. The observable consistency of the candidate list of structural references in the windshield bracket image is evaluated, and benchmark adjudication processing is performed to output valid benchmark identifiers; Based on the valid reference identifier, relational coordinate mapping and structural representation reconstruction are performed on each structural element in the front windshield bracket image to generate a structurally aligned image.

[0009] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket detection system of the present invention, the specific steps for forming structural evidence items are as follows: Based on the structural semantic constraint set, the structural regions of the structure-aligned image are located, and the morphological and positional descriptive quantities of the structural regions are extracted through local appearance analysis and spatial representation analysis to generate a structural descriptive quantity table. Calculate the relative relational descriptive quantities of the structural descriptive scale, perform relational feasibility analysis on the relative relational descriptive quantities, and construct the assumptions that the relation holds true and the relation does not hold true. Structural elements are encapsulated as evidence with their corresponding structural regions, and assumptions of valid and invalid relationships, forming structural evidence items.

[0010] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket detection system of the present invention, the specific steps for generating the structural semantic assertion set are as follows: Based on the set of structural semantic constraints, structural evidence items are screened for constraint consistency to generate a subset of interpretable evidence. For multiple structural evidence items in the deducible evidence subset that point to the same structural element or have a constraint relationship, perform relational hypothesis combination analysis and consistency determination to generate a consistent relational combination item; Extract the structural relation state of the consistent relation combination items and perform semantic encapsulation to generate a set of structural semantic assertions.

[0011] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket detection system of the present invention, the specific steps for generating the structural semantic conflict set are as follows: Based on the set of structural semantic constraints, construct a non-coexistent comparison table by performing contradiction comparison rules on the same structural element or structural element pairs with structural semantic constraint relationships in the set of structural semantic assertions. Retrieve combinations of structural semantic assertions that match incompatible counterparts from the set of structural semantic assertions, and compile them into a table of conflict candidate combinations. For all structural semantic assertion combinations in the conflict candidate combination table, backtrack to read the structural evidence items, and verify the consistency of the refutation of the relational assumptions or relational non-assumptions encapsulated in the structural evidence items according to the structural semantic constraint set, and generate refutation-valid conflict items. Extract the structural elements and corresponding structural regions involved in the conflict terms of the proof by contradiction, encapsulate the conflict pointers, and generate a set of structural semantic conflicts.

[0012] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket detection system of the present invention, the specific steps for locating the structural conflict region in the set of semantic conflicts are as follows: Merge the structural regions pointed to by the same structural semantic conflict item in the set of structural semantic conflicts, and count the frequency of occurrence of each structural region in different structural semantic conflict items and the number of associated structural semantic assertion combinations to generate a regional conflict participation table. Based on the set of structural semantic constraints, conflict adjudication is performed on the conflict participation degree of each structural region in the regional conflict participation table to generate a high-confidence conflict region. The high-confidence conflict region is mapped to the structure alignment image and a consistency check is performed to obtain the structure conflict region.

[0013] As a preferred embodiment of the AI-based visual recognition-based automated windshield bracket inspection system of the present invention, the specific steps for generating verification evidence items are as follows: Extract local image sub-regions corresponding to structural conflict areas from the structure-aligned image, and impose structural semantic constraints on the local image sub-regions to generate a set of analytic subgraphs; Enhanced visual analysis processing is performed on the set of analytical subgraphs using edge detection and contour analysis methods to extract local morphological discrimination features and relative relationship discrimination features, and generate conflict region discrimination terms. The consistency of the conflict region discrimination item and the structural relationship state in the corresponding structural semantic assertion combination are compared and encapsulated as evidence to generate the verification evidence item.

[0014] As a preferred embodiment of the AI-based visual recognition-based automated detection system for windshield supports described in this invention, the specific steps for obtaining the minimum set of failed structures are as follows: By analyzing the combination of structural semantic assertions supported or denied by each piece of review evidence, the support and denial correspondences between the review evidence and the corresponding combination of structural semantic assertions are determined, and an evidence-assertion correspondence table is generated. Backtracking to extract evidence—the structural elements and corresponding structural regions involved in the structural semantic assertion combinations that are judged to be negative correspondences in the assertion correspondence table are used to generate a candidate failure structure table; Based on the set of structural semantic constraints, causal association resolution and causal convergence processing are performed on the candidate failure structure table to generate the minimum set of failure structures.

[0015] As a preferred embodiment of the AI-based visual recognition-based automated inspection system for windshield supports described in this invention, the specific steps for outputting graded inspection conclusions and structural traceability identifiers are as follows: The structural elements and corresponding structural regions contained in the minimum set of failed structures are mapped to the structural alignment image to generate a failed structure mapping table. Based on the set of structural semantic constraints, the failure impact range of each structural element in the failure structure mapping table and the degree of impact on the front windshield bracket assembly relationship are statistically analyzed, and adjudication is performed to generate adjudication judgment items. The number of structural elements in the minimum set of failed structures, the regional distribution range of structural elements in the windshield bracket image, and the degree of clustering of failure effects are statistically analyzed. The adjudication criteria are used as the grading benchmark to perform grading adjudication on the windshield bracket image, generating grading detection conclusions and structural traceability identifiers.

[0016] The beneficial effects of this invention are as follows: by constructing a set of structural semantic conflicts through contradictory comparison rules, it achieves high sensitivity and high reliability in detecting related defects and implicit conflicts, effectively eliminating misjudgments caused by accidental errors in single measurements; by simultaneously analyzing the support or denial relationship between evidence and initial assertions, it obtains the minimum set of failure structures, realizing a qualitative change in the detection output from a problem list to a diagnostic report, greatly improving the actionability of the results, and enhancing the analytical capabilities and decision interpretability of visual inspection in dealing with complex failure modes. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a schematic diagram of an automated front windshield bracket inspection system based on AI visual recognition.

[0019] Figure 2 A flowchart for generating a set of structural semantic constraints.

[0020] Figure 3 This is a flowchart for structural semantic conflict detection.

[0021] Figure 4 This is a flowchart for causal adjudication and hierarchical testing. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4As one embodiment of the present invention, this embodiment provides an automated detection system for windshield brackets based on AI visual recognition, comprising the following steps: The structural semantic module acquires images of the windshield bracket and defines constraints on the relationships between various structural elements in the windshield bracket images, generating a set of structural semantic constraints.

[0026] The windshield bracket image is input into a convolutional neural network to obtain candidate image regions. By analyzing the structural geometric representation of the candidate image regions, the corresponding structural elements are identified and labeled.

[0027] It should be noted that the windshield bracket image includes two-dimensional image information, spatial pose and imaging parameter information of the windshield bracket in the assembly or inspection station; the windshield bracket image acquisition process is completed by an industrial camera in a fixed station or under relatively stable shooting conditions. During the acquisition process, the shooting angle and shooting distance are kept stable to ensure comparability between different windshield bracket images.

[0028] During the training phase, the acquired windshield bracket images are divided into training and validation sets according to a uniform ratio (e.g., 70% training set, 30% validation set). The training set is scaled and normalized to a uniform resolution to form augmented samples. The Adam optimizer is used to backpropagate gradients onto the augmented samples, and a learning rate scheduler is applied to dynamically adjust the learning rate to obtain the training loss. On the validation set, the training loss is cross-validated to obtain the validation error. When the validation error converges and no longer decreases within several consecutive training epochs (e.g., 10 epochs), the trained convolutional neural network is output.

[0029] Pixel normalization and scaling are performed on the windshield bracket image to form the network input tensor. Multi-layer two-dimensional convolution operations are performed on the input tensor through convolution kernels. The results of the multi-layer two-dimensional convolution operations are processed by nonlinear activation functions and downsampling to form a multi-scale feature map containing spatial positioning information. The class score vector and bounding box parameter offset of each spatial location in the multi-scale feature map are obtained. The class score vector is obtained by the corresponding feature map through fully connected operation and softmax transformation, and the bounding box parameter offset is obtained by the corresponding feature map through linear regression. The class score vectors and bounding box parameter offsets at all spatial locations are combined to calculate the coordinates of multiple candidate bounding boxes and their corresponding class probabilities. Non-maximum suppression is performed based on the spatial overlap between candidate bounding boxes. Bounding boxes with high overlap (e.g., above 80%) and low class probability (e.g., below 40%) are removed from the candidate bounding boxes, and the retained bounding boxes are used as candidate image regions.

[0030] By visually detecting and statistically analyzing the grayscale changes of adjacent pixels in candidate image regions, pixels with consistent grayscale change directions and spatial contact are grouped into continuous boundary segments. The ratio of the distance between the beginning and end of a continuous boundary segment to its overall length is used as a measure of closure, and the gradient direction change of adjacent boundary pixels is used as a measure of curvature change. These two measures are normalized and combined to form a comprehensive evaluation value. The continuous boundary segment with the largest comprehensive evaluation value is selected as the outer contour, and the comprehensive evaluation value is combined with the category score vector corresponding to the candidate image region. The candidate image regions that are ranked in the top 20% (e.g., the top 20%) in both geometric consistency score and category probability are assigned structural element labels. Specifically, boundary segments with an approximately circular or regular opening shape and a fixed relative position to the outline are labeled as holes; continuous boundary segments with a compact shape and consistent distribution in multiple locations are labeled as weld points; continuous boundary segments with periodic boundary changes and consistent with the assembly direction are labeled as threads; and continuous boundary segments with a stable connection to the outline and a relatively gentle boundary shape are labeled as assembly surfaces. This completes the labeling of structural elements.

[0031] It should also be noted that multi-layer two-dimensional convolution operation refers to performing multiple two-dimensional convolution operations in series on the input tensor. The output of each convolution operation is used as the input of the next convolution operation. Each two-dimensional convolution operation performs weighted summation and activation transformation in its spatial neighborhood.

[0032] By extracting the stable co-occurrence relationships and assemblable relationships among the structural elements in the windshield bracket image, relationship modeling and semantic constraint partitioning are performed on each structural element to generate a set of structural semantic constraints.

[0033] It should be noted that the center position, boundary orientation, and connection boundary position of each structural element in the front windshield bracket image are read one by one. By statistically analyzing the relative distance, relative direction, and whether there is boundary contact or nesting relationship between structural elements, a spatial relationship description between structural element pairs is obtained. The spatial relationship description is compared in multiple front windshield bracket images. When the same type of structural element pair shows the same distance ordering relationship, directional consistency, and connection existence in different front windshield bracket images, it is recorded as a stable co-occurrence relationship. Based on directly observable spatial matching features such as the axial alignment of holes and threads in space, the attachment or covering relationship between weld points and assembly surfaces on the boundary, and the continuous connection relationship between assembly surfaces and the outer contour on the boundary, it is regarded as an assemblable relationship. The relationships between structural elements are divided according to the constraint strength. Relationships that satisfy stable co-occurrence and assemblability are classified as strong constraint relationships, relationships that only satisfy stable co-occurrence are classified as weak constraint relationships, and the remaining relationships are classified as exclusion relationships. All the classified relationships are sorted to form a structural semantic constraint set.

[0034] The structural baseline module performs structural baseline reconstruction processing on the front windshield bracket image based on the structural semantic constraint set, generates a structural alignment image, identifies the structural relationships in the structural alignment image and performs relationship feasibility determination, forming structural evidence items.

[0035] Using the set of structural semantic constraints as relational constraints, the constraint participation degree of structural elements in the windshield bracket image is statistically analyzed to generate a candidate list of structural references.

[0036] It should be noted that the statistical results of all structural elements in the same front windshield bracket image are used as the comparison benchmark. The relative position of the number of stable co-occurrence relationships and the number of assemblable relationships of a certain structural element in the overall statistical distribution is used as the basis for evaluating the constraint participation. Structural elements with high constraint participation (such as the top 20%) and involving multiple types (such as more than 2 types) of stable co-occurrence relationships are compiled into a structural reference candidate list.

[0037] The observable consistency of the candidate list of structural references in the windshield bracket image is evaluated, and benchmark adjudication processing is performed to output a valid benchmark identifier.

[0038] It should be noted that in the structural reference candidate list, the center position, boundary orientation, and relative distance and direction relationship between each structural element and other structural elements in the front windshield bracket image are read one by one. The relative distance order of the same structural element in different stable co-occurrence relationships is compared to determine whether the relative direction is reversed and whether the boundary contact or nesting relationship continues to exist. Structural elements with high observable consistency in the structural reference candidate list (such as the top 30%) and participating in both stable co-occurrence and assemblable relationships are assigned a valid reference label.

[0039] Based on the valid reference identifier, relational coordinate mapping and structural representation reconstruction are performed on each structural element in the front windshield bracket image to generate a structurally aligned image.

[0040] It should be noted that the structural element corresponding to the valid datum identifier is selected as the spatial reference body. Based on the center position and main direction of the spatial reference body in the windshield bracket image, a unified relational coordinate system is constructed, and the original pixel coordinates of the remaining structural elements in the windshield bracket image are converted into relative position coordinates relative to the relational coordinate system. Based on the stable co-occurrence relationship and assemblable relationship in the structural semantic constraint set, the main connection direction and relational sequence constraints between each structural element are derived and analyzed to form the assembly direction description and connection sequence description for reconstruction. With the structural element corresponding to the valid datum identifier as the origin of the relational coordinate, relational coordinate mapping is performed on each structural element in the windshield bracket image. During the relational coordinate mapping process, the structural morphology descriptions such as the contour principal axis direction, boundary curvature distribution, and key connection point spatial distribution of each structural element are temporarily extracted. The structural morphology descriptions are reordered in a consistent manner according to the assembly direction description and connection sequence description, and uniformly reconstructed into a structural alignment image with the valid datum identifier as the origin and the derived assembly principal direction as the positive axis.

[0041] It should also be noted that the derived assembly main direction refers to the main direction description quantity obtained by statistically adjudicating the assembly directions that repeatedly occur and have the same direction among structural elements, based on the stable co-occurrence relationship and assemblable relationship recorded in the structural semantic constraint set.

[0042] Based on the structural semantic constraint set, the structural regions of the structure-aligned image are located, and the morphological and positional descriptors of the structural regions are extracted through local appearance analysis and spatial representation analysis to generate a structural descriptor table.

[0043] It should be noted that, in the structural alignment image, the combination of structural elements satisfying strong and weak constraints is identified, and the corresponding front windshield bracket image range is limited to the structural region; within each structural region, continuous boundary segments are read along the boundaries of the labeled structural elements, and the boundary curvature change amplitude, boundary length ratio, and boundary closure degree are statistically analyzed as morphological description quantities; using the effective datum marker as the datum origin, the Euclidean distance from the center position of the structural region to the datum origin, the angle between the main extension direction and the main assembly direction, and the relative distance ranking relationship between adjacent structural regions are calculated to form positional description quantities; the morphological description quantities and positional description quantities corresponding to each structural region are collected according to the structural element markers to generate a structural description quantity table.

[0044] The expression for calculating Euclidean distance is: ; in, Indicates the first The Euclidean distance from the center of each structural region to the corresponding reference origin. Indicates the first The x-coordinate of the center position of each structural region. Indicates the first The ordinate of the center position of each structural region. Indicates valid reference identifier The horizontal coordinate in the structure alignment image Indicates valid reference identifier The vertical coordinate in the structure-aligned image.

[0045] Calculate the relative relational descriptors of the structural descriptor scale, perform relational feasibility analysis on the relative relational descriptors, and construct the assumptions that the relation holds true and the relation does not hold true.

[0046] It should be noted that the ratio of the boundary contact length of any two structural elements in the structural description table to the overall boundary length of one of the structural elements is used as the consistency quantity of the connection relationship. The weighted sum of the Euclidean distance, relative direction deviation, and consistency quantity of the connection relationship of any two structural elements in the structural description table is used to obtain the relative relationship description quantity of the current structural element. All relative relationship description quantities in the structural description table are compared with the relationship constraint states of the corresponding structural element pairs in the structural semantic constraint set. When the relative relationship description quantity is consistent with the stable interval corresponding to the strong or weak constraint relationship, the corresponding structural element pair is marked as having a valid relationship, and a hypothesis of a valid relationship is constructed. When the relative relationship description quantity deviates from the allowable range of variation of the relationship constraint state, the corresponding structural element pair is marked as having a invalid relationship, and a hypothesis of a invalid relationship is constructed.

[0047] The expression for calculating the relative direction offset is: ; in, Representing structural elements With structural elements The relative directional deviation. Indicates the structural elements Pointing to structural elements directional vector, Representing structural elements With structural elements The expected direction vector is derived from the relative direction relationship obtained statistically from the stable co-occurrence relationship; It represents the magnitude of the vector.

[0048] Structural elements are encapsulated as evidence with their corresponding structural regions, and assumptions of valid and invalid relationships, forming structural evidence items.

[0049] It should be noted that the structural elements are read from the corresponding structural region positions, boundary ranges, and index identifiers in the structural description table in the structural alignment image. The corresponding relative relation description quantities, relation validity assumptions, or relation invalidity assumptions are then associated and encapsulated as relation determination results. The relation determination results are then bound to the corresponding relation constraint types (strong or weak constraints) to form a structural evidence item containing structural element identifiers, structural region identifiers, relative relation description quantities, relation determination results, and the source of relation constraints.

[0050] The disproving assertion module performs structural state interpretation processing on structural evidence items based on the structural semantic constraint set, generates a set of structural semantic assertions, identifies structural semantic assertion combinations that cannot be simultaneously true in the set of structural semantic assertions, performs disproving consistency verification, and generates a set of structural semantic conflicts.

[0051] Based on the set of structural semantic constraints, the structural evidence items are screened for constraint consistency to generate an interpretable subset of evidence.

[0052] It should be noted that the structural evidence items are screened for constraint consistency based on the structural semantic constraint set. Specifically, when the relation determination result in the structural evidence item is consistent with the relation constraint type in the structural semantic constraint set, the current structural evidence item is marked as meeting the constraint consistency requirement; when the relation determination result is inconsistent with the relation constraint type in the structural semantic constraint set, or when there is no relation constraint definition for the current structural evidence item in the structural semantic constraint set, the current structural evidence item is determined to lack an interpretable basis and is removed; all structural evidence items that meet the constraint consistency requirement are gathered to form an interpretable evidence subset.

[0053] For multiple structural evidence items in the deducible evidence subset that point to the same structural element or have a constraint relationship, perform relational hypothesis combination analysis and consistency determination to generate a consistent relational combination item.

[0054] It should be noted that for multiple structural evidence items in the deducible evidence subset that point to the same structural element or have constraint relationships in the structural semantic constraint set, the structural evidence items are merged according to the structural element identifier and structural region identifier. Within each merged group, the relative relation description quantity, relation judgment result, and corresponding relation constraint type contained in each structural evidence item are read. Within the same merged group, the relation holding assumptions and relation not holding assumptions involving the same structural element pair in the structural evidence item are combined and listed to form a relation assumption combination set. According to the relation constraint type specified for the current structural element pair in the structural semantic constraint set, a consistency judgment is performed on each relation assumption combination. When all relation judgment results in the relation assumption combination remain compatible within the range allowed by the relation constraint type, they are encapsulated as a consistent relation combination item. When there is a relation judgment result in the relation assumption combination that conflicts with the relation constraint type, the judgment is invalid and it is removed, thereby outputting a consistent relation combination item set.

[0055] It should also be noted that when two structural elements are referenced simultaneously by the same relation constraint or the same set of relation constraints in the structural semantic constraint set, it is determined that there is a constraint relationship between the current structural element pair.

[0056] Extract the structural relation state of the consistent relation combination items and perform semantic encapsulation to generate a set of structural semantic assertions.

[0057] It should be noted that the relationship determination results and relationship constraint types corresponding to each structural evidence item in the consistency relationship combination are analyzed, and the relationship determination results of the same structural element pair are summarized and compared. When all structural evidence items in the consistency relationship combination give the same relationship determination result for the same structural element pair, and the relationship determination result does not violate the corresponding relationship constraint type, the current relationship determination result is determined as the structural relationship state. The structural relationship state is structurally integrated with the corresponding structural element identifier, structural region identifier, and relationship constraint type to form a set of structural semantic assertions. Each structural semantic assertion item contains a structural element identifier, a structural relationship state, and the corresponding source of structural semantic constraints.

[0058] Based on the set of structural semantic constraints, construct a non-coexistent comparison table by performing contradictory comparison rules on the same structural element or structural element pairs with structural semantic constraint relationships in the set of structural semantic assertions.

[0059] It should be noted that, for structural element pairs in the structural semantic assertion set that involve the same structural element or have direct constraint relationships (such as belonging to a stable co-occurrence relationship or an assemblable relationship), a comparative analysis is performed on the corresponding structural semantic assertion terms. If it is found that the structural relationship states of different structural semantic assertion terms for the same structural element pair are mutually exclusive under the same relationship constraint type, or if the combination of structural relationship states of the structural semantic assertion set for multiple structural element pairs with assembly dependency relationships violates the spatial logical consistency derived from the assemblable relationship, then the identified mutually exclusive structural relationship state combination and the triggered logical conflict condition are defined as a contradiction comparison rule. All structural semantic assertion sets to be compared are traversed, and each contradiction comparison rule and the corresponding structural semantic assertion term identifier are recorded to generate an incompatible comparison table.

[0060] Retrieve combinations of structural semantic assertions that match incompatible counterparts from the set of structural semantic assertions, and compile them into a table of conflict candidate combinations.

[0061] It should be noted that, based on the structural semantic assertion item identifiers recorded in the incompatible comparison table, structural semantic assertion items with the same identifiers are retrieved from the structural semantic assertion set, and then classified and grouped according to their respective directly mutually exclusive items. All groups and their corresponding contradiction comparison rules are organized and listed to generate a conflict candidate combination table.

[0062] It should also be noted that in the set of structural semantic assertions, when searching for all structural semantic assertion items pointing to the same pair of structural elements, if it is found that there are at least two structural semantic assertion items for the same pair of structural elements, one of which asserts the relation determination result as "relationship is established" and the other assertion the relation determination result as "relationship is not established", then the unique identifiers of these two or more assertions are recorded as a directly mutually exclusive entry in the incompatible lookup table.

[0063] For all structural semantic assertion combinations in the conflict candidate combination table, backtrack to read the structural evidence items, and verify the consistency of the refutation of the relational assumptions or relational non-assumptions encapsulated in the structural evidence items according to the structural semantic constraint set, and generate refutation-validated conflict items.

[0064] It should be noted that, based on the source identifier of the structural evidence item recorded in each structural semantic assertion combination in the conflict candidate combination table, the corresponding structural evidence item is traced back to; the relational validity assumption or relational invalidity assumption encapsulated in each structural evidence item and the corresponding relative relational descriptor are read; if the evidence supporting the relational invalidity assumption (i.e., the relative relational descriptor) is clearly outside the range of geometric parameters allowed by the relation constraint type, while the relative relational descriptor of the evidence supporting the relational validity assumption falls within the range of geometric parameters, then the relational invalidity assumption is determined to be supported by counter-evidence, and the structural semantic assertion combination representing the relational invalidity constitutes a valid conflict; all the structural semantic assertion combinations that have been verified and determined to be valid, together with the verification conclusion, are encapsulated into a counter-evidence conflict item.

[0065] It should also be noted that the range of geometric parameters is derived from the standard geometric parameters (such as average relative distance and angle) obtained from the statistical analysis of stable co-occurrence relationships in the structural semantic constraint set.

[0066] Extract the structural elements and corresponding structural regions involved in the conflict terms of the proof by contradiction, encapsulate the conflict pointers, and generate a set of structural semantic conflicts.

[0067] It should be noted that the mutually exclusive structural element pairs recorded in the counter-evidence conflict items are extracted, and the corresponding structural evidence items are traced back based on the structural element pairs. The structural region coordinate range encapsulated in the structural evidence items is read, and the verification conclusion recorded in the counter-evidence conflict items is extracted. The structural element identifiers, the corresponding structural region coordinate ranges, and the verification conclusions are bound and encapsulated to form a conflict record containing conflict element identifiers, conflict region coordinates, and counter-evidence support conclusions. All conflict records are organized to form a structural semantic conflict set.

[0068] The conflict review module locates structural conflict regions in the set of structural semantic conflicts through conflict adjudication, performs targeted visual analysis processing on the structural conflict regions in the structural alignment image, and generates review evidence items.

[0069] Merge the structural regions pointed to by the same structural semantic conflict item in the set of structural semantic conflicts, and count the frequency of occurrence of each structural region in different structural semantic conflict items and the number of associated structural semantic assertion combinations to generate a regional conflict participation table.

[0070] It should be noted that the coordinates of conflict regions in the structural semantic conflict set are extracted, and structural regions with completely identical or overlapping coordinate ranges (e.g., overlapping area exceeding 80%) are considered independent regions and merged. A unique independent region identifier is assigned to each independent region. All merged independent regions are traversed, and the number of records in the structural semantic conflict set whose conflict region coordinates are merged into the current independent region identifier is counted as the frequency of occurrence of conflict items. The conflict candidate combination table from which the conflict records originate is traced, and the number of different structural semantic assertion combinations involving the current independent region is counted as the number of associated assertion combinations. The independent region identifier, frequency of occurrence of conflict items, and number of associated assertion combinations of each independent region are compiled into a table to generate a regional conflict participation table.

[0071] Based on the set of structural semantic constraints, conflict adjudication is performed on the conflict participation degree of each structural region in the regional conflict participation table to generate a highly reliable conflict region.

[0072] It should be noted that, based on the structural semantic constraint set, the relation constraint types involved in each structural region in the regional conflict participation table are parsed, and the number of strong constraint relationships and the number of weak constraint relationships in the conflicts associated with each structural region are counted respectively. The number of strong constraint relationship conflicts is assigned a higher weight, and the number of weak constraint relationship conflicts is assigned a lower weight. The conflict participation evaluation value of the current structural region is obtained by weighted summation with the frequency of occurrence of conflict items and the number of associated assertion combinations of the structural region. The conflict participation evaluation values ​​of all structural regions are normalized, and structural regions with evaluation values ​​higher than the overall distribution (such as the top 20% quantile) are identified as high-confidence conflict regions.

[0073] The high-confidence conflict region is mapped to the structure alignment image and a consistency check is performed to obtain the structure conflict region.

[0074] It should be noted that, based on the relational coordinate mapping established when generating the structure-aligned image, the structural region coordinates of the high-confidence conflict area are transformed to the pixel coordinate system of the structure-aligned image. The corresponding image region is located in the structure-aligned image, and a consistency check is performed. Specifically, the structural relational state of all structural semantic assertion combinations that cause the image region to be marked as a high-confidence conflict is obtained. Within the corresponding local region of the structure-aligned image, the relative relational description of the involved structural elements is actually measured. The actually measured relative relational description is compared and verified with the structural relational state. If the actually measured relative relational description clearly deviates from the structural relational state, then it is confirmed that there is a real conflict in the current image region, and it is identified as a structural conflict region.

[0075] Extract local image sub-regions corresponding to structural conflict areas from the structure-aligned image, and impose structural semantic constraints on the local image sub-regions to generate a set of analytic subgraphs.

[0076] It should be noted that in the structural alignment image, local image sub-regions within the pixel range corresponding to the structural conflict area are extracted; based on the corresponding record of each local image sub-region in the structural semantic conflict set, it is associated with the specific structural element pair that caused the conflict and the corresponding relational constraint type; and according to the relational constraint type, the specific geometric parameter requirements of the stable co-occurrence relationship or assemblable relationship that the structural element pair must satisfy are read from the structural semantic constraint set, and the geometric parameter requirements are used as spatial and morphological constraints and attached to the corresponding local image sub-region to generate analytic subgraphs. All analytic subgraphs are then organized to form an analytic subgraph set.

[0077] Enhanced visual analysis processing is performed on the set of analytical subgraphs using edge detection and contour analysis methods. Local morphological discrimination features and relative relationship discrimination features are extracted respectively to generate conflict region discrimination terms.

[0078] It should be noted that, based on the conflict records associated with the analytic subgraph, structural element pairs are located, and the standard geometric parameters of the stable co-occurrence relationship or assemblable relationship corresponding to the structural element pairs are read from the structural semantic constraint set. Using the standard geometric parameters as a benchmark, the analytic subgraph is analyzed. For a single structural element, the boundary closure degree description and the boundary curvature change magnitude description are re-acquired as local morphological discrimination features. For paired structural elements with constraint relationships, the relative distance and relative direction deviation are re-measured as relative relationship discrimination features. The extracted measured discrimination feature values ​​(local morphological discrimination features and relative relationship discrimination features) are compared with the standard geometric parameters item by item, and each feature is recorded as to whether it conforms to the parameter tolerance range of the standard geometric parameters. Each analytic subgraph generates a conflict area discrimination item containing image region identifier, measured discrimination feature value, standard geometric parameter value, and the conformity status of each feature.

[0079] The consistency of the conflict region discrimination item and the structural relationship state in the corresponding structural semantic assertion combination are compared and encapsulated as evidence to generate the verification evidence item.

[0080] It should be noted that the feature conformity states recorded in the conflict region discrimination item are logically compared with the structural relationship states in the corresponding structural semantic assertion combinations. Specifically, if the structural relationship state is "relationship established" and the corresponding feature conformity state is "conformity", the comparison result is recorded as "supported"; if the structural relationship state is "relationship established" but the corresponding feature conformity state is "disconformity", it is recorded as "not supported"; if the structural relationship state is "relationship not established" and the corresponding feature conformity state is "disconformity", it is recorded as "supported"; if the structural relationship state is "relationship not established" but the corresponding feature conformity state is "conformity", it is recorded as "not supported". After completing the logical consistency comparison, the conflict region discrimination item, the compared structural semantic assertion combination identifier and structural relationship state, and the generated item-by-item comparison results are integrated, associated, and encapsulated to generate a verification evidence item.

[0081] The causal adjudication module verifies the evidence items by causal inference, obtains the minimum set of failed structures, automatically detects and adjudicates the front windshield bracket image based on the minimum set of failed structures, and outputs graded detection conclusions and structural traceability identifiers.

[0082] By analyzing the combination of structural semantic assertions supported or denied by each verification evidence item, the supporting and denying correspondences between the verification evidence item and the corresponding combination of structural semantic assertions are determined, and an evidence-assertion correspondence table is generated.

[0083] It should be noted that the comparison results (i.e., whether the conclusion is supported or not supported) recorded in each review evidence item are analyzed. If the comparison results of all conflict area discrimination items in the review evidence item are supported, it is determined that there is a supporting correspondence between the review evidence item and the corresponding structural semantic assertion combination. If the comparison result of at least one conflict area discrimination item in the review evidence item is not supported, it is determined that there is a negative correspondence. All review evidence items are traversed, and the identifier of each conflict area discrimination item, the identifier of the corresponding structural semantic assertion combination, and the type of the determined correspondence (support or negation) are recorded as a mapping relationship. All mapping relationships are sorted out to generate an evidence-assertion correspondence table.

[0084] Backtracking to extract evidence—the structural elements and corresponding structural regions involved in the structural semantic assertion combinations that are judged to be negative correspondences in the assertion correspondence table are used to generate a candidate failure structure table.

[0085] It should be noted that the evidence-assertion correspondence table is traversed to filter out all record entries with a negative correspondence type. For each filtered record entry, the structural element identifier associated with the corresponding structural semantic assertion combination is extracted. Based on the structural element identifier, the structural evidence item on which these structural semantic assertion combinations are based is traced back. The structural region coordinates corresponding to each structural element are read from the structural evidence item. Each involved structural element identifier and its corresponding structural region coordinates are treated as a structural record. All structural records are organized to generate a candidate failure structure table.

[0086] Based on the set of structural semantic constraints, causal association resolution and causal convergence processing are performed on the candidate failure structure table to generate the minimum set of failure structures.

[0087] It should be noted that the process involves traversing each structural record in the candidate failure structure table to obtain the structural element identifiers contained in the structural record; in the structural semantic constraint set, it is searched to determine whether there is an assemblable relationship between the structural element identifiers, and the directional attribute of the assemblable relationship is confirmed. For paired structural elements with an assemblable relationship and a clear relationship direction, the structural element identifier located in the starting direction of the assemblable relationship is marked as the superior dependent element of the other structural element identifier. If a superior dependent element of a structural element identifier of a structural record in the candidate failure structure table also exists in another structural record in the candidate failure structure table, then the current structural record is determined to be a convergent term. All structural records in the candidate failure structure table that are determined to be convergent terms are removed, and the structural element identifiers and corresponding structural region coordinates contained in the remaining structural records are organized to form the minimum set of failure structures.

[0088] The structural elements and corresponding structural regions contained in the minimum set of failed structures are mapped to the structure alignment image to generate a failed structure mapping table.

[0089] It should be noted that the coordinates of the structural regions contained in each structural record in the minimum set of failed structures are transformed into the pixel coordinate system of the structural alignment image according to the relational coordinate mapping relationship established when generating the structural alignment image, so as to obtain the corresponding image coordinate range; the structural element identifier of each structural record is associated with the transformed image coordinate range to form a mapping record; all mapping records are structurally integrated to generate a failed structure mapping table.

[0090] Based on the set of structural semantic constraints, the failure impact range of each structural element in the failure structure mapping table and its impact on the front windshield bracket assembly relationship are statistically analyzed, and adjudication is performed to generate adjudication judgment items.

[0091] It should be noted that the structural element identifiers contained in each structural record in the minimum failure structure set are used as the analysis object. All constraint edges (i.e., stable co-occurrence relationships or assemblable relationships) with the analysis object as one of their endpoints are searched in the structural semantic constraint set. The total number of constraint edges is counted as the failure impact range value. All associated edges are weighted and summed to obtain the degree of influence on the front windshield bracket assembly relationship. Structural elements whose failure impact range value and influence degree value are both higher than their respective medians are judged as critical failures, and the rest are judged as minor failures. Each structural element identifier, the corresponding structural region coordinates, the failure impact range value and influence degree value, and the adjudication classification results (the results of being judged as critical failures and minor failures) are encapsulated to generate adjudication judgment items.

[0092] The number of structural elements in the minimum set of failed structures, the regional distribution range of structural elements in the windshield bracket image, and the degree of clustering of failure effects are statistically analyzed. The adjudication criteria are used as the grading benchmark to perform grading adjudication on the windshield bracket image, generating grading detection conclusions and structural traceability identifiers.

[0093] It should be noted that the total number of different structural element identifiers in the minimum failure structure set is taken as the total number of failure elements; based on the structural region coordinates of each structural record in the minimum failure structure set, the total area of ​​the pixel range covered in the structural alignment image is obtained, and the ratio of the total area of ​​the pixel range to the total area of ​​the region defined by the outline in the structural alignment image is taken as the region distribution range index; the number of structural element identifiers classified as critical failures in the statistical decision item, and the reciprocal of the average Euclidean distance between the structural region coordinates corresponding to the critical failure elements are defined as the clustering degree index of the failure impact; if there is a critical failure and both the region distribution range index and the clustering degree index exceed the statistical reference level of historical qualified samples, a serious defect conclusion is generated; if only a minor failure exists, or although there is a critical failure but the clustering degree index does not exceed the statistical reference level at the same time, a general defect conclusion is generated; if the minimum failure structure set is empty, a no-defect conclusion is generated; all data records in the minimum failure structure set and the decision item set that lead to the corresponding conclusion are structured and integrated, and a unique structural traceability identifier is generated through hash operation.

[0094] It should also be noted that the statistical reference level of historical qualified samples is established by statistically analyzing the distribution of various indicators of qualified windshield bracket images and selecting high percentiles (such as the 95th percentile) as dynamic benchmarks.

[0095] In summary, this invention achieves high sensitivity and high reliability in detecting related defects and implicit conflicts by constructing contradiction comparison rules to generate a set of structural semantic conflicts, effectively eliminating misjudgments caused by accidental errors in single measurements; and simultaneously analyzes the support or denial relationship between evidence and initial assertions to obtain a minimum set of failure structures, thus realizing a qualitative change in the detection output from a problem list to a diagnostic report, greatly improving the actionability of the results, and enhancing the analytical capabilities and decision interpretability of visual inspection in dealing with complex failure modes.

[0096] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An automated inspection system for windshield brackets based on AI visual recognition, characterized in that: include, The structural semantic module acquires images of the windshield bracket and defines constraints on the relationships between various structural elements in the windshield bracket images, generating a set of structural semantic constraints. The structural benchmark module performs structural benchmark reconstruction processing on the front windshield bracket image based on the structural semantic constraint set, generates a structural alignment image, identifies the structural relationships in the structural alignment image and performs relationship feasibility determination, and forms structural evidence items. The disproving assertion module performs structural state interpretation processing on structural evidence items based on the structural semantic constraint set, generates a set of structural semantic assertions, identifies structural semantic assertion combinations that cannot be simultaneously true in the set of structural semantic assertions, performs disproving consistency verification, and generates a set of structural semantic conflicts. The conflict review module locates structural conflict regions in the set of structural semantic conflicts through conflict adjudication, performs targeted visual analysis processing on the structural conflict regions in the structural alignment image, and generates review evidence items. The causal adjudication module verifies the evidence items by causal inference, obtains the minimum set of failed structures, automatically detects and adjudicates the front windshield bracket image based on the minimum set of failed structures, and outputs graded detection conclusions and structural traceability identifiers.

2. The automated front windshield bracket detection system based on AI visual recognition as described in claim 1, characterized in that: The specific steps for generating the structural semantic constraint set are as follows: The windshield bracket image is input into a convolutional neural network to obtain candidate image regions. By analyzing the structural geometric representation of the candidate image regions, the corresponding structural elements are identified and labeled. By extracting the stable co-occurrence relationships and assemblable relationships among the structural elements in the windshield bracket image, relationship modeling and semantic constraint partitioning are performed on each structural element to generate a set of structural semantic constraints.

3. The automated front windshield bracket detection system based on AI visual recognition as described in claim 2, characterized in that: The specific steps for generating the structure-aligned image are as follows. Using the set of structural semantic constraints as relational constraints, the constraint participation degree of structural elements in the windshield bracket image is statistically analyzed to generate a candidate list of structural references. The observable consistency of the candidate list of structural references in the windshield bracket image is evaluated, and benchmark adjudication processing is performed to output valid benchmark identifiers; Based on the valid reference identifier, relational coordinate mapping and structural representation reconstruction are performed on each structural element in the front windshield bracket image to generate a structurally aligned image.

4. The automated front windshield bracket detection system based on AI visual recognition as described in claim 3, characterized in that: The specific steps for forming the structural evidence item are as follows. Based on the structural semantic constraint set, the structural regions of the structure-aligned image are located, and the morphological and positional descriptive quantities of the structural regions are extracted through local appearance analysis and spatial representation analysis to generate a structural descriptive quantity table. Calculate the relative relational descriptive quantities of the structural descriptive scale, perform relational feasibility analysis on the relative relational descriptive quantities, and construct the assumptions that the relation holds true and the relation does not hold true. Structural elements are encapsulated as evidence with their corresponding structural regions, and assumptions of valid and invalid relationships, forming structural evidence items.

5. The automated front windshield bracket inspection system based on AI visual recognition as described in claim 4, characterized in that: The specific steps for generating the set of structural semantic assertions are as follows: Based on the set of structural semantic constraints, structural evidence items are screened for constraint consistency to generate a subset of interpretable evidence. For multiple structural evidence items in the deducible evidence subset that point to the same structural element or have a constraint relationship, perform relational hypothesis combination analysis and consistency determination to generate a consistent relational combination item; Extract the structural relation state of the consistent relation combination items and perform semantic encapsulation to generate a set of structural semantic assertions.

6. The automated front windshield bracket detection system based on AI visual recognition as described in claim 5, characterized in that: The specific steps for generating the set of structural semantic conflicts are as follows: Based on the set of structural semantic constraints, construct a non-coexistent comparison table by performing contradiction comparison rules on the same structural element or structural element pairs with structural semantic constraint relationships in the set of structural semantic assertions. Retrieve combinations of structural semantic assertions that match incompatible counterparts from the set of structural semantic assertions, and compile them into a table of conflict candidate combinations. For all structural semantic assertion combinations in the conflict candidate combination table, backtrack to read the structural evidence items, and verify the consistency of the refutation of the relational assumptions or relational non-assumptions encapsulated in the structural evidence items according to the structural semantic constraint set, and generate refutation-valid conflict items. Extract the structural elements and corresponding structural regions involved in the conflict terms of the proof by contradiction, encapsulate the conflict pointers, and generate a set of structural semantic conflicts.

7. The automated front windshield bracket detection system based on AI visual recognition as described in claim 6, characterized in that: The specific steps for locating structural conflict regions in the set of semantic conflicts are as follows. Merge the structural regions pointed to by the same structural semantic conflict item in the set of structural semantic conflicts, and count the frequency of occurrence of each structural region in different structural semantic conflict items and the number of associated structural semantic assertion combinations to generate a regional conflict participation table. Based on the set of structural semantic constraints, conflict adjudication is performed on the conflict participation degree of each structural region in the regional conflict participation table to generate a high-confidence conflict region. The high-confidence conflict region is mapped to the structure alignment image and a consistency check is performed to obtain the structure conflict region.

8. The automated front windshield bracket detection system based on AI visual recognition as described in claim 7, characterized in that: The specific steps for generating the review evidence items are as follows. Extract local image sub-regions corresponding to structural conflict areas from the structure-aligned image, and impose structural semantic constraints on the local image sub-regions to generate a set of analytic subgraphs; Enhanced visual analysis processing is performed on the set of analytical subgraphs using edge detection and contour analysis methods to extract local morphological discrimination features and relative relationship discrimination features, and generate conflict region discrimination terms. The consistency of the conflict region discrimination item and the structural relationship state in the corresponding structural semantic assertion combination are compared and encapsulated as evidence to generate the verification evidence item.

9. The automated front windshield bracket detection system based on AI visual recognition as described in claim 8, characterized in that: The specific steps for obtaining the minimum set of failed structures are as follows: By analyzing the combination of structural semantic assertions supported or denied by each piece of review evidence, the support and denial correspondences between the review evidence and the corresponding combination of structural semantic assertions are determined, and an evidence-assertion correspondence table is generated. Backtracking to extract evidence—the structural elements and corresponding structural regions involved in the structural semantic assertion combinations that are judged to be negative correspondences in the assertion correspondence table are used to generate a candidate failure structure table; Based on the set of structural semantic constraints, causal association resolution and causal convergence processing are performed on the candidate failure structure table to generate the minimum set of failure structures.

10. The automated detection system for windshield brackets based on AI visual recognition as described in claim 9, characterized in that: The specific steps for outputting the hierarchical detection conclusions and structural traceability identifiers are as follows. The structural elements and corresponding structural regions contained in the minimum set of failed structures are mapped to the structural alignment image to generate a failed structure mapping table. Based on the set of structural semantic constraints, the failure impact range of each structural element in the failure structure mapping table and the degree of impact on the front windshield bracket assembly relationship are statistically analyzed, and adjudication is performed to generate adjudication judgment items. The number of structural elements in the minimum set of failed structures, the regional distribution range of structural elements in the windshield bracket image, and the degree of clustering of failure effects are statistically analyzed. The adjudication criteria are used as the grading benchmark to perform grading adjudication on the windshield bracket image, generating grading detection conclusions and structural traceability identifiers.

Images