Defect data processing and querying method and system for braid visual inspection

Abstract

The invention relates to the technical field of braid quality detection in industrial production, in particular to a defect data processing query method and system for braid visual inspection. The method comprises the steps of obtaining braid image data, performing enhancement processing on an image based on a dynamic illumination compensation algorithm, and generating a high-quality image in combination with material characteristics; inputting the enhanced image data into a deep learning defect detection model, detecting and classifying defect areas in the image, and updating the model weight based on defect feature parameters; performing multi-dimensional indexing processing on the detected defect feature data to generate a defect distribution database supporting real-time query; and the defect distribution database and the production process parameter sequence are combined, the causal relationship of defect occurrence is analyzed, process optimization parameters are generated and fed back to the detection module, and self-adaptive updating of model parameters is completed. According to the invention, high precision, high efficiency and closed-loop optimization of braid defect detection are realized, and intelligent support is provided for industrial production quality control.

Description

Technical Field

[0001] The present invention relates to the technical field of webbing quality inspection in industrial production, and in particular to a defect data processing query method and system for webbing visual inspection. Background Art

[0002] With the development of automation and intelligence in industrial manufacturing, defects such as broken yarn, cracks, and burrs often occur in the production process of webbing due to factors such as equipment operation and environmental conditions. If the above defects cannot be detected and located in time, it will directly affect the quality and performance of webbing products. At present, traditional webbing defect detection mostly relies on manual detection and rule-based machine vision methods, which have the following shortcomings:

[0003] (1) Insufficient detection accuracy: Traditional machine vision methods are mainly based on image enhancement and simple edge detection algorithms. They are difficult to accurately identify small non-significant defects in complex texture backgrounds, especially when the lighting is uneven and the ribbon materials are diverse.

[0004] (2) Poor real-time performance and adaptability: Since defect detection algorithms fail to combine dynamic optimization of production process parameters, traditional systems often have difficulty adapting to rapidly changing production conditions, resulting in high rates of missed detection and false detection, and are unable to meet the needs of modern and efficient production.

[0005] (3) Low data utilization: The existing system fails to fully utilize historical inspection data and process parameters for multimodal correlation analysis, making it difficult to optimize the production process and improve the performance of the inspection model through a data-driven approach.

[0006] Based on the above problems, there is an urgent need for an intelligent detection system that can integrate ribbon image data, production process parameters and defect feature information. Through multimodal data fusion and deep learning models, the defect area can be accurately detected and the detection model can be dynamically optimized, thereby realizing high-precision and high-efficiency ribbon quality detection and defect query functions, providing technical support for intelligent manufacturing. Summary of the invention

[0007] The present invention provides a defect data processing query method and system for webbing visual inspection, so as to solve the problem of how to achieve accurate detection and classification of defect areas based on image data, production process parameters and defect feature information in webbing visual inspection through multimodal data fusion and deep learning model, and dynamically optimize the detection model to improve the real-time and adaptability of defect detection.

[0008] In order to solve the above technical problems, the present invention provides a defect data processing and query method for webbing visual inspection, comprising:

[0009] Acquire ribbon image data, enhance the image based on a dynamic illumination compensation algorithm, and generate enhanced image data in combination with material characteristics;

[0010] The enhanced image data is input into the deep learning defect detection model to detect and classify the defect areas in the image, and the detection weights of the dynamic optimization model are updated based on the defect feature parameters;

[0011] Perform multi-dimensional index processing on the defect feature data obtained by detection, and generate a defect distribution database in combination with the defect time series;

[0012] Based on the defect distribution database and production process parameter sequence, the cause-effect relationship of defect occurrence is analyzed, process optimization parameters are generated and fed back to the dynamic optimization model to complete the adaptive update of model parameters.

[0013] Furthermore, the step of obtaining the ribbon image data includes:

[0014] The motion images of the webbing production process are captured in real time by industrial cameras, and the collected raw image data are preliminarily corrected and denoised to generate a denoised webbing image sequence.

[0015] Furthermore, the step of enhancing the image based on the dynamic illumination compensation algorithm includes:

[0016] Perform brightness distribution statistics on the denoised image and extract the lighting characteristics;

[0017] Adjust the local brightness value of the image based on the illumination characteristic mapping to generate an image with uniform illumination;

[0018] The texture of the enhanced image is optimized in combination with material properties to generate high-quality image data suitable for defect detection.

[0019] Furthermore, the steps of generating high-quality image data in combination with material characteristics include:

[0020] The material reflectivity and surface roughness information in the image are extracted and combined with the image texture feature enhancement to generate optimized image data.

[0021] Furthermore, the step of inputting the enhanced image data into the deep learning defect detection model includes:

[0022] Use deep learning models to detect image regions, locate defect areas, and extract defect features;

[0023] The defect features are input into the classification model to classify the defect area and generate defect type and location parameters.

[0024] Furthermore, the step of classifying the defective area includes:

[0025] Defects are classified based on their geometric shape, grayscale distribution and texture features, and structured data including type, location, time and production batch are generated.

[0026] Furthermore, the step of performing multi-dimensional index processing on the defect feature data obtained by detection includes:

[0027] Build a multidimensional index based on defect type, location, time, and batch information;

[0028] Generate a defect distribution database that supports real-time query and dynamically updates the defect index.

[0029] Furthermore, the steps of performing causal relationship analysis based on the defect distribution database and the production process parameter sequence include:

[0030] Match the defect distribution database with production process parameters to build a comprehensive data set;

[0031] Generate defect causal relationship sequences using causal inference models based on comprehensive datasets.

[0032] Furthermore, the steps of extracting key process parameters from the causal relationship sequence include:

[0033] Screen production process parameters that have a significant impact on defect occurrence, generate optimization suggestions and build a process optimization parameter set.

[0034] Furthermore, a defect data processing and query system for visual inspection of webbing includes:

[0035] A data acquisition module is used to obtain the ribbon image data and perform preliminary correction and denoising on the image;

[0036] The data preprocessing module is used to enhance the illumination of the denoised image based on the dynamic illumination compensation algorithm and generate high-quality image data in combination with material characteristics;

[0037] Defect detection module, which is used to input high-quality image data into the deep learning model to detect and classify defect areas in the image;

[0038] The data indexing module is used to perform multi-dimensional indexing on the detected defect feature data and generate a defect distribution database that supports real-time query;

[0039] Data analysis module, used to perform causal relationship analysis based on defect distribution database and production process parameters to generate defect causal relationship sequence;

[0040] The optimization feedback module is used to extract key process parameters from the defect causal relationship sequence, generate process optimization parameters and feed them back to the defect detection module to complete the dynamic optimization of the detection model.

[0041] The key innovative features of the present invention include:

[0042] (1) Image enhancement technology combining dynamic illumination compensation and material optimization. By combining the dynamic illumination compensation algorithm with the material characteristics of the ribbon, image enhancement processing under complex lighting conditions is achieved, providing high-quality input data and laying the foundation for improving the accuracy of defect detection.

[0043] (2) Dynamic optimization detection model based on deep learning. The deep learning model is used for defect detection and classification, and the model weight is adjusted in real time based on the defect feature parameters, realizing the adaptive optimization of the detection model and improving the model's adaptability to complex production scenarios.

[0044] (3) Construction of multidimensional index and defect distribution database. For the first time, multidimensional index technology was introduced into ribbon visual inspection, combining defect feature data with time series to build a defect distribution database that supports real-time query, providing strong data support for the tracing and optimization of production process problems.

[0045] (4) Multimodal causal analysis and process optimization closed-loop mechanism. Through the correlation analysis between the defect distribution database and the production process parameters, the causal relationship between defects and process conditions is established, and the optimization parameters are generated and fed back to the detection model, forming a closed-loop optimization process for process adjustment and defect detection.

[0046] The following are its main beneficial effects:

[0047] (1) Improve the accuracy and adaptability of defect detection. The present invention uses a dynamic illumination compensation algorithm to enhance the ribbon image data, effectively eliminating the interference of uneven illumination, and further optimizes the image texture details in combination with material characteristics, significantly improving the defect detection model's ability to identify small complex background defects. At the same time, the dynamic optimization mechanism based on the defect feature parameter updates the model weights, enabling the deep learning model to adapt to changes in process conditions during the ribbon production process, reducing missed detection and false detection rates.

[0048] (2) Realize efficient query and analysis of defect data. By performing multi-dimensional index processing on the defect feature data obtained by detection and combining the defect time series to generate a defect distribution database that supports real-time query, the present invention realizes efficient storage and rapid retrieval of defect data. The multi-dimensional index structure enables users to query the type, location, time and batch information of defects on demand, providing a convenient means for tracing and analyzing production problems.

[0049] (3) Enhance the optimization capability of the production process. Based on the defect distribution database and the production process parameter sequence, the cause-effect relationship of the defect is analyzed, key process optimization parameters are generated and fed back to the detection model to form a closed-loop optimization process. The cause-effect relationship analysis module of the present invention accurately identifies the key process factors that affect the occurrence of defects through multimodal data fusion, provides scientific process adjustment suggestions for ribbon production, and improves overall production efficiency and product quality.

[0050] (4) Support real-time and intelligent detection. The dynamic optimization mechanism and real-time database update function of the present invention ensure that the system can adapt to the rapidly changing production environment. Whether it is data collection and processing, or model optimization and feedback, it has real-time response capabilities, effectively meeting the high requirements of modern intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 A flowchart of a defect data processing and query method for visual inspection of webbing provided in an embodiment of the present application;

[0052] Figure 2 A structural block diagram of a defect data processing and query system for visual inspection of webbing provided in an embodiment of the present application. DETAILED DESCRIPTION

[0053] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by technicians in the technical field of the present application; the terms used in the specification of the application herein are only for the purpose of describing specific embodiments and are not intended to limit the present application; the terms "including" and "having" and any variations thereof in the specification and claims of the present application and the above-mentioned drawings are intended to cover non-exclusive inclusions. The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, not to describe a specific order.

[0054] Reference to "embodiments" herein means that the specific features and structural characteristics described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0055] Example 1: Reference Figure 1 , is a flow chart of a defect data processing query method for webbing visual inspection provided by an embodiment of the present invention, and the flow may at least include steps S100-S400:

[0056] S100, acquiring ribbon image data, enhancing the image based on a dynamic illumination compensation algorithm, and generating enhanced image data in combination with material characteristics.

[0057] S200: Input the enhanced image data into the deep learning defect detection model, detect and classify the defect areas in the image, and update the detection weights of the dynamic optimization model based on the defect feature parameters.

[0058] S300, performing multi-dimensional index processing on the defect feature data obtained by detection, and generating a defect distribution database in combination with the defect time series.

[0059] S400, based on the defect distribution database and the production process parameter sequence, analyze the cause and effect relationship of the defect, generate process optimization parameters and feed them back to the dynamic optimization model, and complete the adaptive update of the model parameters.

[0060] Step S100 at least includes steps S110-S130:

[0061] S110: acquiring real-time image data captured during the webbing production process, and performing preliminary correction and denoising processing on the image through a camera to obtain a denoised webbing image sequence.

[0062] First, the camera continuously captures the motion of the ribbon at a predetermined frame rate by synchronizing the speed parameters of the production line and records it as an image frame sequence I raw The image contains problems such as uneven illumination, noise interference, and motion blur.

[0063] Furthermore, the original image I raw After distortion correction and pixel alignment. The specific method is as follows:

[0064] Based on the camera intrinsic parameters and distortion model, the geometric distortion in the image caused by lens characteristics is corrected to generate a distortion-free image.

[0065] The grayscale value of each frame of the image is adjusted to standardize the grayscale distribution of different brightness areas and reduce the interference of grayscale offset on texture feature extraction.

[0066] Furthermore, raw Apply the noise suppression algorithm to generate the denoised image I denoiseI The methods used include:

[0067] The pixel grayscale and spatial position are weighted simultaneously to retain edge features while suppressing noise.

[0068] For Gaussian distributed noise, smoothing is used to reduce the intensity of random noise in the image.

[0069] The denoised image data is arranged in time series to generate a frame sequence IdenoiseI (t), and append the timestamp t and production batch number B to ensure that subsequent analysis can be performed by time dimension and production batch association.

[0070] The denoised ribbon image sequence is used as the input of S120 to provide clear and standardized basic data for subsequent illumination compensation.

[0071] S120: Performing illumination enhancement on the denoised ribbon image based on a dynamic illumination compensation algorithm to generate an enhanced image under uniform illumination conditions.

[0072] The denoised image sequence I output by S110 denoiseI Perform frame-by-frame brightness distribution statistics and extract illumination distribution characteristics:

[0073] The image is divided into multiple sub-regions, the average brightness of each region is calculated, and the illumination feature map L(x,y) is generated.

[0074] The brightness mean and variance of the entire image are counted to identify the impact of illumination changes on image contrast.

[0075] Furthermore, based on the illumination feature map L(x, y), the local brightness value of the image is compensated and adjusted.

[0076] For low-brightness areas, increase the brightness value; for over-bright areas, reduce the brightness value to ensure overall uniform brightness.

[0077] The adjusted local compensation results are further optimized through the equalization algorithm to make the overall brightness distribution of the image normalized.

[0078] Furthermore, the compensated uniform illumination image is fused with the original image at pixel level to further enhance the details and texture information. enhanced Maintain consistent visual properties in both bright and dim light conditions.

[0079] The enhanced uniform illumination image is used as the input of S130, providing high-quality basic data for material and texture property optimization.

[0080] S130: Combining the material characteristics and texture information of the ribbon, optimizing the features of the enhanced image to generate high-quality image data suitable for subsequent defect detection.

[0081] Combined with the material parameters of the ribbon (reflectivity, texture density), the material features in the image are extracted:

[0082] The reflectivity feature R(x,y) of each area in the enhanced image is calculated to locate the area with strong light reflection characteristics on the ribbon surface.

[0083] By calculating the distribution characteristics of the grayscale gradient, the roughness information of the ribbon surface is extracted to analyze the potential defective areas.

[0084] Furthermore, based on the structural characteristics of the ribbon, the texture area in the image is further enhanced:

[0085] Convert the image to the frequency domain, enhance the texture frequency components, and make the defect features more prominent.

[0086] An edge detection algorithm is applied to extract key texture edges of the webbing and enhance the contrast of said edges.

[0087] Combine material characteristics and texture features to generate the final optimized image I through weighted fusion optimized This image has clear texture features and uniform brightness distribution, providing ideal input data for subsequent defect detection and classification.

[0088] Optimized image data I optimized As input data for the next module, it ensures the accuracy and efficiency of subsequent detection.

[0089] Inter-module connection and association instructions

[0090] S110→S120: The denoised image sequence provides a unified input for illumination compensation to prevent noise from affecting illumination analysis results.

[0091] S120→S130: The enhanced uniform illumination image provides a reliable basis for material property extraction and texture feature optimization.

[0092] S130 → Subsequent modules: The optimized high-quality images are directly used as input data for the defect detection module to improve the robustness and accuracy of detection.

[0093] Step S200 at least includes steps S210-S230:

[0094] S210: Input the high-quality image data into a pre-trained deep learning defect detection model to detect potential defect areas in the ribbon image and extract defect features.

[0095] The optimized image data output from S130 is directly input to the S210 module. The optimized image contains enhanced information of material properties and texture features, and is annotated in chronological order and batch. The input data is further formatted into the deep learning model M detect The standard input format, data dimension, grayscale range and resolution meet the requirements of the model.

[0096] The optimized image is fed into a pre-trained deep learning defect detection model, which uses a convolutional neural network to extract multi-level features of the image. Through a sliding window strategy fully convolutional network, the model locates potential defect areas in the image and generates bounding boxes for candidate defect areas.

[0097] For each candidate defect region, the deep learning model further extracts key feature information, including the region's geometric shape, grayscale value distribution, texture features, etc. The feature data is stored as feature vectors, each of which contains multiple fields for describing the properties of the defect region.

[0098] The detected defective areas and their features are structured into a data set with timestamp and batch information attached to generate an intermediate result, which provides input for the S220 module.

[0099] S220: Based on the detected defect features, the defect area is classified and processed to generate defect type and position parameters.

[0100] The defect feature data output from the S210 module is input to the S220 module, and the defects are classified using a pre-trained classification model combined with information such as the texture, shape, and illumination characteristics of the defect area. The classification model uses a multi-layer perceptron or other deep learning structure suitable for classification tasks.

[0101] The classification model assigns specific type labels ("broken yarn", "crack", "burr", etc.) to each defect area according to the input feature vector, and stores the classification results in a structured manner. The classification labels provide a key basis for dynamic optimization in subsequent modules.

[0102] Combined with the coordinate information of the defect area in the original image, the global position parameters of the defect are calculated, and the time and production batch are marked. The position parameters are stored together with the classification results to form complete defect detection data.

[0103] The classification results and location parameters are integrated into a complete data structure to generate a detection result set, which contains the type, location and related features of each defect area and provides input for the S230 module.

[0104] Through the above steps, the S220 module realizes the classification processing of the detected defect areas, and the generated defect type and location parameters will provide a basis for model optimization.

[0105] S230: According to the defect type and characteristic parameters, the detection weight of the deep learning model is adjusted, the detection model is dynamically optimized, and an optimized detection model is generated.

[0106] The detection result set obtained from the S220 module is input to the S230 module. The detection result set includes the type, location and related feature information of each defect area. The S230 module analyzes the data, extracts the feature weight of each type of defect, and provides input for dynamic optimization.

[0107] According to the detection frequency and feature distribution of the defect type, load the corresponding weight adjustment strategy in the optimization strategy library. The adjustment strategy will assign a new weight factor to each type of defect based on the actual performance of the detection (classification accuracy, missed detection rate, etc.).

[0108] Based on the optimization strategy, the weight parameters of the deep learning model are dynamically adjusted. The weight adjustment is completed by the optimizer, making the model more sensitive to defect types with higher detection frequency and importance, thereby improving the overall detection accuracy.

[0109] The updated weights are loaded into the detection model M detect , generating an optimized deep learning model. The optimized model has higher defect detection accuracy and stronger adaptability, and can be used for subsequent detection tasks.

[0110] Through the above steps, the S230 module completes the dynamic optimization of the detection model and forms a closed-loop adaptive detection process.

[0111] Inter-module connection and association instructions

[0112] S210→S220: The detected defect feature data provides input for the classification model, and the classification results and feature information are integrated into a complete data set.

[0113] S220→S230: The classification results and position parameters provide a basis for weight adjustment, and the optimization strategy dynamically updates the model weights in combination with the detection data.

[0114] S230 → Subsequent detection task: The optimized model is returned to the detection task to further improve the detection efficiency and accuracy.

[0115] Step S300 at least includes steps S310-S330:

[0116] S310: Structural processing is performed on the defect type and location parameters, and a defect feature sequence is generated by combining time and production batch information.

[0117] The S310 module receives the defect detection result set output by the S220 module. The data includes characteristic fields such as the type, location parameters, timestamp, and production batch of each defect area. The S310 module parses the input data, extracts necessary fields, and performs normalization processing.

[0118] For each detected defect, a feature vector is generated, containing the following main fields:

[0119] Defect Type T i : Classification label indicating defects (such as “broken yarn”, “crack”, etc.).

[0120] Position parameter L i : Contains the center coordinates and bounding box information of the defect.

[0121] Timestamp i : Indicates the time when the defect detection occurs.

[0122] Production batch B i : Identifies the production batch to which the defect belongs.

[0123] The fields are combined into a structured feature vector F i , and stored in sequence form.

[0124] Furthermore, the generated defect feature vector is processed according to the timestamp t i Sort them to form a time-series defect feature sequence S feature This sequence makes it easier to associate defects with the timeline when constructing a defect distribution database.

[0125] After generating the defect feature sequence, the data is verified and cleaned for consistency. Invalid and abnormal data are eliminated, such as missing batch information and incomplete timestamp entries, to ensure the accuracy of subsequent processing.

[0126] Through the above steps, module S310 outputs the time-series defect feature sequence S feature , providing input for the S320 module.

[0127] S320: Input the defect feature sequence into a multi-dimensional index construction algorithm to generate a defect distribution database supporting real-time query.

[0128] The S320 module receives the defect feature sequence S output from S310. feature First, the parameter dimensions of the multidimensional index are defined according to the defect characteristics, including the following dimensions:

[0129] Type dimension: Based on defect type T i Construct a classification hierarchy.

[0130] Location dimension: using the defect location parameter L i Build a spatial index.

[0131] Time dimension: using timestamp t i Build a time series index.

[0132] Batch dimension: combined with production batch Bi Build the production batch index.

[0133] The dimension definitions form the basic framework of the index.

[0134] Furthermore, the defect feature sequence S feature Enter the multidimensional index construction algorithm and perform hierarchical indexing based on time, location, type, and batch as dimensions. The algorithms used are KD-Tree, R-Tree, and other technologies suitable for multidimensional indexing. After the construction is completed, a multidimensional index structure Θ is formed. index , its hierarchy is clear and supports fast query.

[0135] The multidimensional index structure Θ index Stored in defect distribution database D defect , and an additional query interface is provided to support users to quickly query relevant defect information by type, location, time and batch conditions.

[0136] Through the above steps, the S320 module generates a defect distribution database D that supports real-time query. defect , providing input for dynamic updates of the S330 module.

[0137] S330: Dynamically update the defect distribution database and generate the latest defect distribution index in combination with the new detection data.

[0138] The S330 module receives the new detection data set D from the S220 module new The data includes newly added defect type, location, time, and production batch information. The S330 module parses the new data and compares it with the index dimension θ of the current defect distribution database. index Compare and extract the dimension information that needs to be updated.

[0139] Furthermore, the parsed new data is serialized into a new feature vector and inserted into the multidimensional index structure Θ index Index updates follow the following rules:

[0140] If the type, location, and time dimensions of the newly added data match the existing index, it is directly appended to the corresponding index node.

[0141] If the dimension value of the newly added data exceeds the current index range, a new index node is created and the index structure is adjusted.

[0142] Furthermore, the updated index structure is stored back into the defect distribution database D defect After storage is completed, the database interface is refreshed to ensure that the query results can reflect the latest defect information in real time.

[0143] Through the above steps, the S330 module completes the dynamic update of the defect distribution database, ensuring that the database can reflect the newly added defect information in the production process in real time.

[0144] Inter-module connection and association instructions

[0145] S310→S320: Time-series defect feature sequence S feature It is the input for building a multidimensional index. The definition of index parameters is directly based on the fields in the feature sequence.

[0146] S320→S330: Generated defect distribution database D defect Provides basic data structure for dynamic update module, and new detection data is reflected in the database through index update.

[0147] S330→Subsequent modules: The updated database interface provides real-time data support for subsequent defect analysis and process optimization, ensuring the timeliness and accuracy of the analysis results.

[0148] Step S400 at least includes steps S410-S430:

[0149] S410: Based on the defect distribution database and the webbing production process parameters, a multimodal association analysis is performed to generate a defect cause-effect sequence.

[0150] Get the defect distribution database D from the S330 module defect The database contains multi-dimensional data such as defect type, location parameters, timestamp and batch information. At the same time, the production parameter sequence P = {P1, P2, ..., P m}, each production parameter P j Includes process settings (such as temperature, tension, speed, etc.) and their corresponding time and batch information.

[0151] Match the data in the defect distribution database with the production process parameter sequence P according to time and batch information to form a comprehensive data set D combined .

[0152] Using the comprehensive dataset D combined ,The relationship between defect characteristics and production parameters is mined through multimodal data analysis methods. Specifically, by analyzing the changing trends of defect types, locations, and quantities with production parameters, the potential influencing factors of defects are preliminarily identified.

[0153] Based on the data association results, a causal reasoning model is constructed to deduce the causal relationship between defect characteristics and production process parameters. The output causal relationship is expressed in sequence form to form a defect causal relationship sequence S causal ={(T i ,Pj )|i=1,2,…,n;j=1,2,…,m}, where each element (T i ,P j ) indicates defect type T i With production parameter P j causal relationship.

[0154] Through the above steps, the S410 module outputs the defect causal relationship sequence and provides input for the S420 module.

[0155] S420: extracting key process parameters from the defect causal relationship sequence, generating optimization suggestions and forming a process optimization parameter set.

[0156] The defect causal relationship sequence S output from the S410 module causal In the process, the production process parameters that have the greatest impact on defects are extracted to form the key process parameter set P key Specifically, the weight information in the causal inference model is used to select the parameter P with the greatest influence. j .

[0157] According to the key process parameters P key , combined with predefined optimization rules, generate process optimization suggestions. Optimization rule base R opt Contains optimization strategies related to each type of process parameter. For example, for tension parameters, the rules may recommend lowering the setpoint; for speed parameters, it may recommend increasing its stability.

[0158] Furthermore, the optimization suggestions are converted into specific parameter adjustment plans to generate the process optimization parameter set P opt Each entry in the optimization parameter set includes information such as parameter name, optimization direction, and adjustment range, ensuring that it can be directly applied to the production process system.

[0159] Through the above steps, the S420 module outputs the process optimization parameter set P opt , providing input for the S430 module.

[0160] S430: Feedback the process optimization parameter set to the defect detection module to complete the adaptive update of the detection model and form a closed-loop optimization process.

[0161] The S430 module receives the process optimization parameter set P from the S420 module. opt and compare it with the defect detection model M detect Specifically, by analyzing the historical performance of process optimization parameters and detection models, the potential impact of optimization parameters on model performance is determined.

[0162] Furthermore, the weights of the detection model are dynamically updated using the adjustment information in the process optimization parameter set. The weight adjustment rules are based on the degree of influence of the optimization parameters on different defect types. For example, if a certain process parameter has a significant improvement effect on the "broken yarn" defect, the relevant weights of the model will be adjusted accordingly.

[0163] Updated detection model M detect The optimized model can identify and classify defects more accurately, forming a closed-loop optimization process.

[0164] Inter-module connection and association instructions

[0165] S410→S420: Defect causal relationship sequence S causal It is the basis for extracting key process parameters and generating optimization suggestions.

[0166] S420→S430: Generated process optimization parameter set P opt It is used to dynamically adjust the detection model weights to ensure that the model performance is consistent with the process optimization direction.

[0167] S430 → Subsequent process: The optimized detection model is used for the next round of defect detection tasks to form a complete closed-loop process.

[0168] The key innovative features of the present invention include:

[0169] (1) Image enhancement technology combining dynamic illumination compensation and material optimization. By combining the dynamic illumination compensation algorithm with the material characteristics of the ribbon, image enhancement processing under complex lighting conditions is achieved, providing high-quality input data and laying the foundation for improving the accuracy of defect detection.

[0170] (2) Dynamic optimization detection model based on deep learning. The deep learning model is used for defect detection and classification, and the model weight is adjusted in real time based on the defect feature parameters, realizing the adaptive optimization of the detection model and improving the model's adaptability to complex production scenarios.

[0171] (3) Construction of multidimensional index and defect distribution database. For the first time, multidimensional index technology was introduced into ribbon visual inspection, combining defect feature data with time series to build a defect distribution database that supports real-time query, providing strong data support for the tracing and optimization of production process problems.

[0172] (4) Multimodal causal analysis and process optimization closed-loop mechanism. Through the correlation analysis between the defect distribution database and the production process parameters, the causal relationship between defects and process conditions is established, and the optimization parameters are generated and fed back to the detection model, forming a closed-loop optimization process for process adjustment and defect detection.

[0173] The following are its main beneficial effects:

[0174] (1) Improve the accuracy and adaptability of defect detection. The present invention uses a dynamic illumination compensation algorithm to enhance the ribbon image data, effectively eliminating the interference of uneven illumination, and further optimizes the image texture details in combination with material characteristics, significantly improving the defect detection model's ability to identify small complex background defects. At the same time, the dynamic optimization mechanism based on the defect feature parameter updates the model weights, enabling the deep learning model to adapt to changes in process conditions during the ribbon production process, reducing missed detection and false detection rates.

[0175] (2) Realize efficient query and analysis of defect data. By performing multi-dimensional index processing on the defect feature data obtained by detection and combining the defect time series to generate a defect distribution database that supports real-time query, the present invention realizes efficient storage and rapid retrieval of defect data. The multi-dimensional index structure enables users to query the type, location, time and batch information of defects on demand, providing a convenient means for tracing and analyzing production problems.

[0176] (3) Enhance the optimization capability of the production process. Based on the defect distribution database and the production process parameter sequence, the cause-effect relationship of the defect is analyzed, key process optimization parameters are generated and fed back to the detection model to form a closed-loop optimization process. The cause-effect relationship analysis module of the present invention accurately identifies the key process factors that affect the occurrence of defects through multimodal data fusion, provides scientific process adjustment suggestions for ribbon production, and improves overall production efficiency and product quality.

[0177] (4) Support real-time and intelligent detection. The dynamic optimization mechanism and real-time database update function of the present invention ensure that the system can adapt to the rapidly changing production environment. Whether it is data collection and processing, or model optimization and feedback, it has real-time response capabilities, effectively meeting the high requirements of modern intelligent manufacturing.

[0178] Embodiment 2: Figure 2 FIG. 2 shows a structural block diagram of a defect data processing and query system for visual inspection of webbing according to an embodiment of the present invention. Figure 2 As shown, the system may include:

[0179] Data acquisition module 10, the data acquisition module is used to obtain image data in the webbing production process in real time. The industrial camera is synchronized with the production line speed to capture the webbing motion picture at a predetermined frame rate. The captured image data includes problems such as uneven brightness, noise interference and motion blur. The collected image data is initially denoised and corrected to ensure image quality and provide reliable input for subsequent processing. The output data is formatted as a time series and batch information is attached.

[0180] Data preprocessing module 20, the data preprocessing module performs illumination compensation and material texture optimization on the collected image data to generate high-quality image data suitable for detection. Dynamic illumination compensation: analyzes the brightness distribution of the image, adjusts the local brightness value, and generates an image with uniform illumination. Material property extraction: extracts key material properties by combining the reflectivity and surface roughness information of the ribbon. Texture feature enhancement: enhances the texture details of the image through frequency domain analysis and edge detection algorithms to make potential defects more prominent.

[0181] Defect detection module 30, the defect detection module detects and classifies defect areas in high-quality image data based on a deep learning model. The optimized image data is input into the pre-trained deep learning model to locate potential defect areas and generate bounding boxes. The geometric shape, texture and grayscale features of the defect area are extracted to form a structured feature vector. The defects are classified using a classification model to generate type labels and location parameters.

[0182] The data indexing module 40 performs multi-dimensional indexing on the defect feature data obtained by the detection, and generates a defect distribution database that supports real-time query. The defect data is indexed and constructed according to dimensions such as type, location, time and batch. A multi-dimensional indexing algorithm (such as KD-TreeR-Tree) is used to generate an efficient query structure. The defect distribution database is dynamically updated to reflect the latest detection data.

[0183] Data analysis module 50: The data analysis module generates a defect causal relationship sequence by analyzing the correlation between the defect distribution database and the production process parameters. The correlation between the defect distribution data and the process parameters is comprehensively analyzed to identify the key process factors for the occurrence of defects. A causal reasoning model is constructed to output the causal relationship sequence between the defect type and the process parameters.

[0184] The optimization feedback module 60 generates process optimization parameters based on the defect causal relationship sequence and feeds back to the defect detection module to complete the closed-loop optimization. The key process parameters in the causal relationship are extracted and optimization suggestions are generated. The detection weight of the deep learning model is dynamically adjusted to optimize the model performance. The optimized model is deployed to the production line to realize a closed-loop process of real-time optimization.

[0185] User interaction module 70, the user interaction module provides an intuitive visual interface to display defect detection results and analysis data. It supports users to query defect distribution and display them by time, location and type. It provides production process parameter adjustment suggestions to assist operators in optimizing process flow.

[0186] System monitoring module 80: The system monitoring module continuously monitors the system operation status and issues early warnings and handles abnormal situations. It monitors the operation status of cameras, databases, and deep learning models. It provides abnormal logs and repair suggestions to ensure efficient and stable operation of the system.

[0187] Beneficial effects: The defect data processing query method and system for webbing visual inspection provided by the present invention has the following beneficial effects by constructing a closed-loop process integrating data collection, defect detection, multi-dimensional indexing and process optimization:

[0188] (1) Real-time and accuracy: Based on dynamic illumination compensation and deep learning models, it can efficiently and accurately detect defects in the webbing. Multi-dimensional indexing and dynamic update mechanisms support real-time query and data update, improving detection efficiency.

[0189] (2) Comprehensiveness and intelligence: Comprehensively consider the relationship between defect distribution and process parameters, build a causal analysis model, and achieve a comprehensive analysis of defect occurrence. Automatically generate optimization parameters and feed them back to the detection module to form a closed-loop optimization process.

[0190] (3) Adaptability and scalability: The modular design facilitates expansion of applications in different production scenarios. The dynamic optimization mechanism enables the system to adapt to changes in different webbing types and production conditions.

[0191] Obviously, the embodiments described above are only some embodiments of the present application, rather than all embodiments. The preferred embodiments of the present application are given in the accompanying drawings, but they do not limit the patent scope of the present application. The present application can be implemented in many different forms. On the contrary, the purpose of providing the embodiments is to make the understanding of the disclosure of the present application more thorough and comprehensive. Although the present application is described in detail with reference to the aforementioned embodiments, for those skilled in the art, it is still possible to modify the technical solutions recorded in the aforementioned specific implementation methods and to replace some of the technical features therein with equivalents. Any equivalent structure made using the contents of the specification and drawings of this application, directly or indirectly used in other related technical fields, is similarly within the scope of patent protection of this application.

Claims

1. A defect data processing query method for ribbon visual inspection, characterized in that: include: Acquire ribbon image data, enhance the image based on a dynamic illumination compensation algorithm, and generate enhanced image data in combination with material characteristics; The enhanced image data is input into the deep learning defect detection model to detect and classify the defect areas in the image, and the detection weights of the dynamic optimization model are updated based on the defect feature parameters; Perform multi-dimensional index processing on the defect feature data obtained by detection, and generate a defect distribution database in combination with the defect time series; Based on the defect distribution database and production process parameter sequence, the cause-effect relationship of defect occurrence is analyzed, process optimization parameters are generated and fed back to the dynamic optimization model to complete the adaptive update of model parameters.

2. The method according to claim 1, characterized in that: The steps of obtaining ribbon image data include: The motion images of the webbing production process are captured in real time by industrial cameras, and the collected raw image data are preliminarily corrected and denoised to generate a denoised webbing image sequence.

3. The method according to claim 1, characterized in that The steps of image enhancement based on the dynamic illumination compensation algorithm include: Perform brightness distribution statistics on the denoised image and extract the lighting characteristics; Adjust the local brightness value of the image based on the illumination characteristic mapping to generate an image with uniform illumination; The texture of the enhanced image is optimized in combination with material properties to generate high-quality image data suitable for defect detection.

4. The method according to claim 3, characterized in that The steps to generate high-quality image data in combination with material properties include: The material reflectivity and surface roughness information in the image are extracted and combined with the image texture feature enhancement to generate optimized image data.

5. The method according to claim 1, characterized in that The steps to feed the enhanced image data into the deep learning defect detection model include: Use deep learning models to detect image regions, locate defect areas, and extract defect features; The defect features are input into the classification model to classify the defect area and generate defect type and location parameters.

6. The method according to claim 5, characterized in that The steps to classify defect areas include: Defects are classified based on their geometric shape, grayscale distribution and texture features, and structured data including type, location, time and production batch are generated.

7. The method according to claim 1, characterized in that The steps of performing multi-dimensional index processing on the defect feature data obtained by detection include: Build a multidimensional index based on defect type, location, time, and batch information; Generate a defect distribution database that supports real-time query and dynamically updates the defect index.

8. The method according to claim 1, characterized in that: The steps of causal relationship analysis based on defect distribution database and production process parameter sequence include: Match the defect distribution database with production process parameters to build a comprehensive data set; Generate defect causal relationship sequences using causal inference models based on comprehensive datasets.

9. The method according to claim 8, characterized in that The steps to extract key process parameters from the causal sequence include: Screen production process parameters that have a significant impact on defect occurrence, generate optimization suggestions and build a process optimization parameter set.

10. A defect data processing and query system for visual inspection of webbing, characterized in that: include: A data acquisition module is used to obtain the ribbon image data and perform preliminary correction and denoising on the image; The data preprocessing module is used to enhance the illumination of the denoised image based on the dynamic illumination compensation algorithm and generate high-quality image data in combination with material characteristics; Defect detection module, which is used to input high-quality image data into the deep learning model to detect and classify defect areas in the image; The data indexing module is used to perform multi-dimensional indexing on the detected defect feature data and generate a defect distribution database that supports real-time query; Data analysis module, used to perform causal relationship analysis based on defect distribution database and production process parameters to generate defect causal relationship sequence; The optimization feedback module is used to extract key process parameters from the defect causal relationship sequence, generate process optimization parameters and feed them back to the defect detection module to complete the dynamic optimization of the detection model.

Images