An AI-based detection and recognition method and system

By constructing an AI-based defect detection model and utilizing fault feature trees and multi-level hash analysis, the problem of low accuracy and efficiency in the detection of automotive parts in existing technologies has been solved, achieving efficient and accurate identification of multiple types of faults and processing guidance.

CN119559161BActive Publication Date: 2026-11-17WUXI YOSHIOKA PRECISION TECH CORP LTD
View PDF 1 Cites 0 Cited by

Patent Information

Application Number
CN202510029453.5
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2025-01-08
Publication Date
2026-11-17
Estimated Expiration
2045-01-08

AI Technical Summary

Technical Problem

Existing methods for detecting defects in automotive parts suffer from low accuracy and efficiency, and are ill-suited to handling complex and varied defect types.

Method used

An AI-based detection and recognition method is adopted. A defect detection model is constructed, which includes a pixel sampling layer, a data array detection layer and an evaluation output layer. Sample-driven training is carried out using fault feature trees and multi-level hash analysis. Iterative sampling and recognition are performed by combining reverse loops and sampling adjustment rules, and the output detection chain is used for processing guidance.

Benefits of technology

It significantly improves the accuracy and efficiency of defect detection in automotive parts, enabling more comprehensive identification and classification of various types of faults and providing precise processing guidance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119559161B_ABST
    Figure CN119559161B_ABST
Patent Text Reader

Abstract

This invention discloses an artificial intelligence-based detection and recognition method and system, relating to the field of image data processing technology. The method includes: for a target automotive component, retrieving fault records and mining a fault feature tree; based on the fault feature tree, using multi-level hash analysis based on a pixel array as the detection method, performing sample-driven training to construct a defect detection model; acquiring images of the target automotive component by controlling a front-end sensing device to determine the component image; transmitting the component image back to the defect detection model for iterative sampling and pixel array-based component defect identification, outputting a detection chain; and providing processing guidance for the target automotive component based on the detection chain. This invention solves the technical problems of existing automotive component defect detection methods, which suffer from low detection accuracy and efficiency, and difficulty in handling complex and varied defect types.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of image data processing technology, specifically to an artificial intelligence-based detection and recognition method and system. Background Technology

[0002] With the continuous development of the automotive industry, the quality and reliability of automotive parts play a crucial role in the performance and safety of the entire vehicle. Traditional defect detection methods mainly rely on manual inspection and rule-based image processing technology. These methods suffer from low efficiency, insufficient accuracy, and high dependence on operators when dealing with complex and diverse defect types.

[0003] Therefore, existing automotive component defect detection methods suffer from low detection accuracy and efficiency, and are unable to cope with complex and varied defect types. Summary of the Invention

[0004] This application provides an artificial intelligence-based detection and identification method and system, which solves the technical problems of low detection accuracy and efficiency, and difficulty in dealing with complex and varied defect types in existing automotive component defect detection methods. It significantly improves the accuracy and efficiency of automotive component defect detection, enabling more comprehensive identification and classification of various types of faults.

[0005] This application provides an artificial intelligence-based detection and identification method, the method comprising: for a target automotive component, retrieving fault records and mining a fault feature tree, wherein the fault features include fault area-fault type; based on the fault feature tree, performing sample-driven training using multi-level hash analysis based on pixel arrays as the detection method to construct a defect detection model, wherein the defect detection model includes a pixel sampling layer, a pixel array detection layer, and an evaluation output layer, and the pixel array detection layer and the pixel sampling layer establish an inverse loop; by controlling a front-end sensing device, acquiring images of the target automotive component to determine the component image; transmitting the component image back to the defect detection model, performing component defect identification based on iterative sampling of pixel arrays, and outputting a detection chain; and providing processing guidance for the target automotive component based on the detection chain.

[0006] In the implementation, the construction of the defect detection model includes: sequentially connecting the pixel sampling layer, the array detection layer, and the evaluation output layer, and establishing a reverse loop from the array detection layer to the pixel sampling layer; setting an initial sampling interval and sampling adjustment rules, and co-training the pixel sampling layer and the array detection layer, wherein the sampling adjustment rule is an interval adjustment ratio.

[0007] In the implementation, the collaborative training of the pixel sampling layer and the array detection layer includes: embedding the fault feature tree, using fault recognition accuracy as the convergence condition, and supervising the training of the array detection layer based on the fault records; extracting the first fault record to determine the original sample image; performing pixel discrete sampling on the original sample image based on the initial sampling interval to determine the first pixel array; transferring the first pixel array to the array detection layer for fault recognition to determine the first pixel fault array; analyzing the first pixel fault array to locate the blurred detection region; determining a second sampling method for the blurred detection region based on sampling adjustment rules; transferring the second sampling method to the pixel sampling layer based on the reverse loop for resampling to determine the second pixel array of the blurred detection region; transferring the second pixel array to the array detection layer for iterative sampling training to complete one training based on the first fault record.

[0008] In the implementation, based on the initial sampling interval, pixel discrete sampling is performed on the original sample image to determine the first pixel array, including: acquiring pixel sampling points, wherein the pixel sampling points are represented as [x,y,z,f(t)], (x,y,z) are the relative distribution positions of the sampling points to the target automobile parts, and f(t) is the pixel feature of the sampling point; for the initial sampling interval, N pixel sampling points are determined; based on the relative distribution positions, the N pixel sampling points are arranged in an array to determine the first pixel array.

[0009] In the implementation, the component defect identification using iterative sampling of a pixel array includes: transmitting the component image back, performing background segmentation to determine the valid image of the target automotive component; performing pixel sampling based on the initial sampling interval on the valid image using the pixel sampling layer to determine the target pixel array; transferring the target pixel array to the array detection layer for defect determination, localization, and iterative sampling detection, superimposing multi-layer iterative detection results to determine the defect detection result; and transferring the defect detection result to the evaluation output layer for damage assessment and rework assessment to determine the detection result.

[0010] In the implementation method, calling the fault records and mining the fault feature tree includes: performing clustering processing on the fault records based on fault type to determine M groups of fault events; extracting a group of fault events, performing fault feature identification as the root feature layer, refining the root feature layer at multiple levels to determine the first fault feature tree; traversing the M groups of fault events until the determination of the Mth fault feature tree is completed, integrating the first fault feature tree until the Mth fault feature tree is obtained, and generating the fault feature tree.

[0011] In the implementation, processing guidance for the target automotive parts is provided based on the detection chain, including: identifying the detection chain, wherein the detection chain includes a part image, an array element digital image, and a detection result, and the array element digital image includes at least one item; determining processing guidance information based on the detection chain, wherein the processing guidance includes part rework guidance and production line processing guidance; and providing processing guidance for the target automotive parts based on the processing guidance information.

[0012] This application also provides an artificial intelligence-based detection and recognition system, comprising: a fault tree construction module, used to call fault records and mine a fault feature tree for a target automotive component, wherein the fault features include fault area-fault type; a defect detection model construction module, used to construct a defect detection model based on the fault feature tree, using multi-level hash analysis based on pixel arrays as the detection method, through sample-driven training, wherein the defect detection model includes a pixel sampling layer, an array detection layer, and an evaluation output layer, and the array detection layer and the pixel sampling layer establish an inverse loop; a component image acquisition module, used to acquire images of the target automotive component by controlling a front-end sensing device to determine the component image; a detection chain acquisition module, used to transmit the component image back to the defect detection model, perform component defect identification based on iterative sampling of pixel arrays, and output a detection chain; and a processing guidance module, used to provide processing guidance for the target automotive component based on the detection chain.

[0013] This application proposes an artificial intelligence-based detection and identification method and system. The method includes: for a target automotive component, retrieving fault records and mining a fault feature tree, wherein the fault features include fault region-fault type; based on the fault feature tree, using multi-level hash analysis based on pixel arrays as the detection method, performing sample-driven training to construct a defect detection model, wherein the defect detection model includes a pixel sampling layer, a pixel array detection layer, and an evaluation output layer, and the pixel array detection layer and the pixel sampling layer establish an inverse loop; by controlling a front-end sensing device, acquiring images of the target automotive component to determine the component image; transmitting the component image back to the defect detection model, performing iterative sampling of the pixel array for component defect identification, and outputting a detection chain; based on the detection chain, providing processing guidance for the target automotive component. This method solves the technical problems of low detection accuracy and efficiency in existing automotive component defect detection methods, which are unable to cope with complex and varied defect types. It significantly improves the accuracy and efficiency of automotive component defect detection, enabling more comprehensive identification and classification of multiple types of faults. Attached Figure Description

[0014] To more clearly illustrate the technical solutions of the embodiments of this disclosure, the accompanying drawings of the embodiments of this disclosure will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of this application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0015] Figure 1 This is a schematic diagram of an artificial intelligence-based detection and recognition method provided in an embodiment of this application.

[0016] Figure 2 This is a schematic diagram of an artificial intelligence-based detection and recognition system provided in an embodiment of this application.

[0017] Figure labeling: Fault tree construction module 11, Defect detection model construction module 12, Component image acquisition module 13, Detection chain acquisition module 14, Processing guidance module 15. Detailed Implementation

[0018] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application.

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description of this application will be provided in conjunction with the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0020] In the following description, references to "some embodiments" describe a subset of all possible embodiments. However, it is understood that "some embodiments" may be the same or different subsets of all possible embodiments and may be combined with each other without conflict. The terms "first" and "second" are used merely to distinguish similar objects and do not represent a specific ordering of objects. The terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or modules not explicitly listed or inherent to these processes, methods, products, or devices. Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application only.

[0021] This application provides an artificial intelligence-based detection and recognition method and system, such as... Figure 1 As shown, the method includes: for the target automotive component, retrieving fault records and mining a fault feature tree, wherein the fault features include fault region-fault type. Based on the fault feature tree, using multi-level hash analysis based on pixel arrays as the detection method, sample-driven training is performed to construct a defect detection model, wherein the defect detection model includes a pixel sampling layer, an array detection layer, and an evaluation output layer, and the array detection layer and the pixel sampling layer establish an inverse loop.

[0022] In the production process of automotive parts, a certain amount of fault data is usually accumulated. These fault records typically include detailed information about the fault, such as the fault location, type, and image data. For target automotive parts that require intelligent detection and identification, the fault records are retrieved and a fault feature tree is constructed. Subsequently, based on the fault feature tree, a sample-driven training method is used, employing multi-level hash analysis based on pixel arrays, to construct a defect detection model. This defect detection model includes a pixel sampling layer, an array detection layer, and an evaluation output layer, with the array detection layer and the pixel sampling layer establishing an inverse loop.

[0023] The method provided in this application embodiment further includes: sequentially connecting the pixel sampling layer, the array detection layer, and the evaluation output layer, and establishing a reverse loop from the array detection layer to the pixel sampling layer. An initial sampling interval and sampling adjustment rule are set, and the pixel sampling layer and the array detection layer are trained collaboratively, wherein the sampling adjustment rule is an interval adjustment ratio.

[0024] The basic architecture of the defect detection model is constructed, consisting of three main layers: a pixel sampling layer, a data array detection layer, and an evaluation output layer. These layers are connected sequentially to form a feedforward network. Furthermore, a reverse loop is established between the data array detection layer and the pixel sampling layer to provide feedback on the detection results and optimize the sampling strategy. Further, an initial sampling interval is obtained by setting the initial interval distance of selected pixels in the component image. Simultaneously, a sampling adjustment rule, primarily an interval adjustment ratio, is set to dynamically adjust the sampling interval during iteration. For example, the sampling adjustment rule is set to sample at 1 / 4 of the initial sampling interval in blurred detection regions. The pixel sampling layer and the data array detection layer are then trained collaboratively.

[0025] The method provided in this application embodiment further includes: embedding the fault feature tree, using fault identification accuracy as the convergence condition, and supervising the training of the array detection layer based on the fault records. A first fault record is extracted to determine the original sample image. Based on the initial sampling interval, pixel discrete sampling is performed on the original sample image to determine a first pixel array. The first pixel array is transferred to the array detection layer for fault identification to determine a first pixel fault array. The first pixel fault array is analyzed to locate the blurred detection region. For the blurred detection region, a second sampling method is determined based on sampling adjustment rules. The second sampling method is transferred to the pixel sampling layer based on the reverse loop for resampling to determine a second pixel array of the blurred detection region. The second pixel array is transferred to the array detection layer for iterative sampling training to complete one training iteration based on the first fault record.

[0026] The collaborative training of the pixel sampling layer and the array detection layer includes: integrating the constructed fault feature tree structure into the defect detection model. The fault feature tree organizes fault features in a hierarchical manner, helping the model better understand and classify different types of faults. Collected fault records are input into the fault feature tree for supervised training. Supervised training uses labeled data (i.e., records of known fault images and corresponding fault types) to train the model, enabling it to learn and recognize different types of faults. A predetermined fault recognition accuracy is set as the convergence condition for model training. When the model's recognition accuracy on the validation set reaches or exceeds this standard, training is complete, and the array detection layer is obtained. A first set of fault records is extracted from the fault record database, and the corresponding original fault sample image is determined based on this record. These images will serve as the basic data for training the model, used for subsequent pixel sampling and defect recognition. According to a preset initial sampling interval, the original sample images are discretely sampled to form the first pixel array. This process ensures that the model can cover the key areas of the image in the initial stage, improving the comprehensiveness of detection. The initial sampling interval is the distance between pixels when sampling in the image. Larger intervals help cover a wider area, while smaller intervals facilitate detailed detection. The first pixel array is a data array structure formed by the set of pixel points selected through the initial sampling interval, used for subsequent defect identification. The first pixel array is transferred to the data array detection layer, where fault identification is performed based on the fault feature tree structure in the data array detection layer to determine the first pixel fault array, which is the set of pixel points with faults identified by the data array detection layer.

[0027] Further analysis of the identified first-pixel fault array determines which regions have ambiguous or uncertain detection results, resulting in ambiguous detection regions. These ambiguous detection regions are those where the detection results are not clear enough, requiring further sampling and analysis to confirm the existence of defects. These ambiguous detection regions require more detailed sampling to improve detection accuracy. For example, if adjacent pixels exist, one faulty and the other normal, it is difficult to determine the fault boundary. Therefore, sampling with a smaller spacing is needed, or the initial sampling spacing can be shifted overall and fed back to the sampling layer for resampling analysis until the detection accuracy is met.

[0028] Based on preset sampling adjustment rules, a more refined second sampling method is determined for the located fuzzy detection area. This second sampling method is then transferred to the pixel sampling layer via the reverse loop for resampling, determining the second pixel array for the fuzzy detection area. The second pixel array is a set of pixel points generated in the fuzzy detection area according to the second sampling method, used for further defect identification. The second pixel array is input to the array detection layer for more refined fault identification. Through iterative sampling and analysis, the model's detection capability is further optimized, ultimately completing a full training process based on the first fault record.

[0029] The method provided in this application embodiment further includes: acquiring pixel sampling points, wherein the pixel sampling points are represented as [x,y,z,f(t)], (x,y,z) are the relative distribution positions of the sampling points to the target automotive parts, and f(t) is the pixel feature of the sampling point. For the initial sampling interval, N pixel sampling points are determined. Based on the relative distribution positions, the N pixel sampling points are arranged in a matrix to determine the first pixel matrix.

[0030] Based on the initial sampling interval, the original sample image is subjected to pixel discrete sampling to determine a first pixel array. This includes: acquiring pixel sampling points, where each pixel sampling point is represented as [x,y,z,f(t)], (x,y,z) is the relative distribution position of the sampling point to the target automotive component, and f(t) is the pixel feature of the sampling point. For the initial sampling interval, N pixel sampling points are determined. Further, based on the relative distribution positions (x,y,z) of the sampling points, the N pixel sampling points are arranged in an array according to certain rules to form a first pixel array. The first pixel array contains a matrix structure formed by the array arrangement, containing information on all sampling points under the initial sampling interval, for subsequent detection and analysis.

[0031] The method provided in this application embodiment further includes: performing clustering processing on the fault records based on fault type to determine M groups of fault events; extracting a group of fault events and performing fault feature identification as a root feature layer; refining the root feature layer at multiple levels to determine a first fault feature tree; traversing the M groups of fault events until the determination of the Mth fault feature tree is completed; integrating the first fault feature tree up to the Mth fault feature tree to generate the fault feature tree.

[0032] The process of retrieving fault records and mining a fault feature tree includes: analyzing all collected fault records, clustering them according to fault type, grouping records with similar fault features (including visual features and fault type) into the same group to simplify subsequent feature extraction and analysis, and determining M groups of fault events, where M is a positive integer greater than 1, and each group of fault events represents a specific type of fault. From the M clustered fault events, fault images of each group are extracted one by one, and detailed fault feature recognition is performed to obtain multiple recognition features, which are used as the root feature layer. The root feature layer is refined at multiple levels of granularity to determine the first fault feature tree. That is, the features of the root feature layer are enriched, such as by truncating, rotating, scaling, and translating the features. This ensures that the root feature layer after multi-level granularity refinement can cover the display information at different feature sizes, so that no matter what the sampled features are, as long as a fault exists, it can be matched and detected. Finally, the M groups of fault events are traversed until the determination of the Mth fault feature tree is completed, and the first fault feature tree is integrated up to the Mth fault feature tree to generate the fault feature tree.

[0033] By controlling the front-end sensing device, images of the target automotive parts are acquired, and the parts images are determined. The parts images are then fed back to the defect detection model for iterative sampling and pixel-based defect identification, outputting a detection chain. Based on this detection chain, processing guidance is provided for the target automotive parts.

[0034] By controlling a front-end sensing device (an industrial camera) to acquire images of the target automotive parts, the system identifies the parts' images. The acquired parts images are then transmitted back to a defect detection model. Defect identification is performed using a pixel array obtained through iterative sampling, and a detection chain is output. This detection chain records the data flow of the entire detection process, including the parts image, the pixel digital image, the detection results, and feedback information from each stage. Finally, based on the detection chain, processing guidance is provided for the target automotive parts. This approach solves the technical problems of low detection accuracy and efficiency in existing automotive parts defect detection methods, which struggle to handle complex and varied defect types. It significantly improves the accuracy and efficiency of automotive parts defect detection, enabling more comprehensive identification and classification of various fault types.

[0035] The method provided in this application embodiment further includes: transmitting the component image back, and determining the valid image of the target automotive component by performing background segmentation. Based on the pixel sampling layer, performing pixel sampling on the valid image based on the initial sampling interval to determine the target pixel array. Transferring the target pixel array to the array detection layer for defect determination, location, and iterative sampling detection, superimposing multi-layer iterative detection results to determine the defect detection result. Transferring the defect detection result to the evaluation output layer for damage degree evaluation and rework evaluation to determine the detection result.

[0036] The component defect identification using iterative sampling and pixel array analysis includes: transmitting the acquired component images back to the detection system; extracting the effective region of the target automotive component from the image using background segmentation technology, removing background noise and irrelevant information to ensure the accuracy of subsequent sampling and detection; after determining the effective image, using a pixel sampling layer to sample pixels in the image according to a preset initial sampling interval to determine the target pixel array; inputting the determined target pixel array into the array detection layer, and analyzing these pixels using a trained defect detection model; determining the corresponding defect category and the location of the defect category through defect judgment and localization, with the location determined based on pixel position, thereby identifying potential defect areas; further, fuzzy detection areas within the potential defect areas are located, and a second sampling method is determined through sampling adjustment rules, performing iterative sampling detection and defect identification analysis to obtain identification results; superimposing the multi-layer detection results to finally determine the specific location and corresponding category of the defect, obtaining the defect detection result; transferring the defect detection result to the evaluation output layer, where the evaluation output layer performs corresponding damage degree evaluation matching based on the defect detection result, obtains the corresponding rework evaluation, and determines the detection result. The average damage level corresponding to different defect detection results and the corresponding resumption evaluation results for whether to resume production are recorded in the evaluation output layer.

[0037] The method provided in this application further includes: identifying the detection chain, wherein the detection chain comprises a component image, an array element digital image, and a detection result, and the array element digital image comprises at least one of these. Based on the detection chain, processing guidance information is determined, wherein the processing guidance includes component rework guidance and production line processing guidance. Based on the processing guidance information, processing guidance is provided for the target automotive component.

[0038] Based on the detection chain, processing guidance for the target automotive component is provided, including: identifying the detection chain, wherein the detection chain comprises a component image, an array element digital image, and a detection result, and the array element digital image comprises at least one item; determining processing guidance information based on the detection chain, wherein the processing guidance comprises component rework guidance and production line processing guidance, and the rework guidance is extracted based on the detection result; the production line processing guidance data is set by professional technicians; and providing processing guidance for the target automotive component based on the processing guidance information.

[0039] In the above text, refer to Figure 1 A detection and recognition method based on artificial intelligence according to an embodiment of the present invention has been described in detail. Next, reference will be made to... Figure 2 This invention describes an artificial intelligence-based detection and recognition system according to an embodiment of the present invention.

[0040] An artificial intelligence-based detection and identification system according to an embodiment of the present invention solves the technical problems of low detection accuracy and efficiency, and difficulty in dealing with complex and varied defect types in existing automotive component defect detection methods. It significantly improves the accuracy and efficiency of automotive component defect detection, and can more comprehensively identify and classify multiple types of faults. The artificial intelligence-based detection and identification system includes: a fault tree construction module 11, a defect detection model construction module 12, a component image acquisition module 13, a detection chain acquisition module 14, and a processing guidance module 15.

[0041] The fault tree construction module 11 is used to call fault records and mine fault feature trees for target automotive parts, where fault features include fault region-fault type.

[0042] The defect detection model construction module 12 is used to construct a defect detection model based on the fault feature tree and using multi-level hash analysis based on pixel array as the detection method, through sample-driven training. The defect detection model includes a pixel sampling layer, an array detection layer and an evaluation output layer, and the array detection layer and the pixel sampling layer establish an inverse loop.

[0043] The component image acquisition module 13 is used to acquire images of the target automotive component by controlling the front-end sensing device and determine the component image.

[0044] The detection chain acquisition module 14 is used to transmit the component image back to the defect detection model, perform component defect identification by iterative sampling of the pixel array, and output the detection chain.

[0045] The processing guidance module 15 is used to provide processing guidance for the target automotive parts based on the detection chain.

[0046] The specific configuration of the defect detection model construction module 12 will be described in detail below. The defect detection model construction module 12 may further include: constructing the defect detection model, which includes: sequentially connecting the pixel sampling layer, the array detection layer, and the evaluation output layer, and establishing a reverse loop from the array detection layer to the pixel sampling layer. An initial sampling interval and sampling adjustment rule are set, and the pixel sampling layer and the array detection layer are trained collaboratively, wherein the sampling adjustment rule is an interval adjustment ratio.

[0047] The specific configuration of the defect detection model construction module 12 will be described in detail below. The defect detection model construction module 12 further includes: the collaborative training of the pixel sampling layer and the array detection layer, comprising: embedding the fault feature tree, using fault recognition accuracy as the convergence condition, and supervising the training of the array detection layer based on the fault records. The first fault record is extracted to determine the original sample image. Based on the initial sampling interval, pixel discrete sampling is performed on the original sample image to determine the first pixel array. The first pixel array is transferred to the array detection layer for fault recognition to determine the first pixel fault array. The first pixel fault array is analyzed to locate the blurred detection region. For the blurred detection region, a second sampling method is determined based on sampling adjustment rules. The second sampling method is transferred to the pixel sampling layer based on the reverse loop for resampling to determine the second pixel array of the blurred detection region. The second pixel array is transferred to the array detection layer for iterative sampling training to complete one training based on the first fault record.

[0048] The specific configuration of the defect detection model construction module 12 will be described in detail below. The defect detection model construction module 12 may further include: performing pixel discrete sampling on the original sample image based on the initial sampling interval to determine a first pixel array, including: acquiring pixel sampling points, wherein the pixel sampling points are represented as [x,y,z,f(t)], (x,y,z) are the relative distribution positions of the sampling points to the target automotive parts, and f(t) is the pixel feature of the sampling point. For the initial sampling interval, N pixel sampling points are determined. Based on the relative distribution positions, the N pixel sampling points are arranged in an array to determine the first pixel array.

[0049] The specific configuration of the detection chain acquisition module 14 will be described in detail below. The detection chain acquisition module 14 further includes: component defect identification based on iterative sampling of a pixel array, comprising: transmitting the component image back; determining the valid image of the target automotive component by performing background segmentation; performing pixel sampling based on the initial sampling interval on the valid image based on the pixel sampling layer to determine the target pixel array; transferring the target pixel array to the array detection layer for defect determination, localization, and iterative sampling detection; superimposing multi-layer iterative detection results to determine the defect detection result; and transferring the defect detection result to the evaluation output layer for damage assessment and rework assessment to determine the detection result.

[0050] The specific configuration of the fault tree construction module 11 will be described in detail below. The fault tree construction module 11 further includes: calling fault records and mining a fault feature tree, including: performing clustering processing on the fault records based on fault type to determine M groups of fault events; extracting a group of fault events, performing fault feature identification as the root feature layer, refining the root feature layer at multiple levels to determine the first fault feature tree; traversing the M groups of fault events until the determination of the Mth fault feature tree is completed, integrating the first fault feature tree up to the Mth fault feature tree to generate the fault feature tree.

[0051] The specific configuration of the processing guidance module 15 will be described in detail below. The processing guidance module 15 further includes: providing processing guidance for the target automotive part based on the detection chain, including: identifying the detection chain, wherein the detection chain includes a part image, an array element digital image, and a detection result, and the array element digital image includes at least one item; determining processing guidance information based on the detection chain, wherein the processing guidance includes part rework guidance and production line processing guidance; and providing processing guidance for the target automotive part based on the processing guidance information.

[0052] The detection and recognition system based on artificial intelligence provided in this embodiment of the invention can execute the detection and recognition method based on artificial intelligence provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0053] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0054] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A detection and recognition method based on artificial intelligence, characterized in that, The method includes: For the target automotive parts, fault records are retrieved and a fault feature tree is mined, where fault features include fault region and fault type; Based on the fault feature tree, a defect detection model is constructed by using multi-level hash analysis based on pixel arrays as the detection method and performing sample-driven training. The defect detection model includes a pixel sampling layer, an array detection layer and an evaluation output layer, and the array detection layer and the pixel sampling layer establish an inverse loop. By controlling the front-end sensing device, images of the target automotive parts are acquired, and the images of the parts are determined; The component images are fed back to the defect detection model for iterative sampling of component defects in the pixel array, and a detection chain is output. Based on the detection chain, processing guidance is provided for the target automotive parts; Retrieve fault logs and mine the fault feature tree, including: The fault records are clustered based on fault type to determine M groups of fault events; Extract a set of fault events, identify fault features as the root feature layer, refine the root feature layer at multiple levels, and determine the first fault feature tree. Traverse the M groups of fault events until the determination of the Mth fault feature tree is completed, integrate the first fault feature tree until the Mth fault feature tree is obtained, and generate the fault feature tree. Based on the detection chain, processing guidance for the target automotive parts is provided, including: Identify the detection chain, wherein the detection chain comprises a component image - an array element digital image - a detection result, and the array element digital image comprises at least one item; Based on the detection chain, processing guidance information is determined, wherein the processing guidance includes guidance for resuming work on parts and guidance for processing on the production line; Based on the processing guidance information, processing guidance is provided for the target automotive parts; The construction of the defect detection model includes: The pixel sampling layer, the array detection layer, and the evaluation output layer are sequentially connected, and a reverse loop from the array detection layer to the pixel sampling layer is established. An initial sampling interval and sampling adjustment rule are set, and the pixel sampling layer and the array detection layer are trained collaboratively. The sampling adjustment rule is the interval adjustment ratio. The collaborative training of the pixel sampling layer and the array detection layer includes: The built-in fault feature tree is used as the convergence condition, and the array detection layer is trained in a supervised manner based on the fault records. Extract the first fault record and determine the original image of the sample; Based on the initial sampling interval, the original image of the sample is subjected to pixel discrete sampling to determine the first pixel array; The first pixel array is transferred to the array detection layer for fault identification to determine the faulty first pixel array. Analyze the fault array of the first pixel to locate the blurred detection area; For the aforementioned fuzzy detection region, a second sampling method is determined based on sampling adjustment rules; The second sampling method is transferred to the pixel sampling layer based on the reverse loop for resampling to determine the second pixel array of the blurred detection region; The second pixel array is transferred to the array detection layer for iterative sampling training, completing one training session based on the first fault record; Based on the initial sampling interval, pixel discrete sampling is performed on the original image of the sample to determine the first pixel array, including: Obtain pixel sampling points, wherein the pixel sampling points are represented as [x,y,z,f(t)], (x,y,z) are the relative distribution positions of the sampling points to the target automotive parts, and f(t) are the pixel features of the sampling points; For the initial sampling interval, N pixel sampling points are determined; Based on their relative distribution positions, the N pixel sampling points are arranged in a number array to determine the first pixel array; The component defect identification based on the pixel array under iterative sampling includes: The component images are returned, and valid images of the target automotive component are determined by performing background segmentation. Based on the pixel sampling layer, pixel sampling based on the initial sampling interval is performed on the effective image to determine the target pixel array; The target pixel array is transferred to the array detection layer for defect determination, location and iterative sampling detection. The results of multi-layer iterative detection are superimposed to determine the defect detection result. The defect detection results are transferred to the evaluation output layer for damage assessment and rework evaluation to determine the detection results.

2. An artificial intelligence-based detection and recognition system, characterized in that, The system is used to perform the method of claim 1, the system comprising: The fault tree construction module is used to retrieve fault records and mine fault feature trees for target automotive parts. The fault features include fault region and fault type. The defect detection model construction module is used to construct a defect detection model based on the fault feature tree and using multi-level hash analysis based on pixel array as the detection method, through sample-driven training. The defect detection model includes a pixel sampling layer, an array detection layer and an evaluation output layer, and the array detection layer and the pixel sampling layer establish an inverse loop. The component image acquisition module is used to acquire images of the target automotive component by controlling the front-end sensing device and to determine the component image; The detection chain acquisition module is used to transmit the component image back to the defect detection model, perform component defect identification under iterative sampling of the pixel array, and output the detection chain; The processing guidance module is used to provide processing guidance for the target automotive parts based on the detection chain.

Citation Information

Patent Citations

  • Industrial surface defect detection system and method based on data analysis

    CN118469996A