Visual identification method and system for non-defective products, medium and program product

Abstract

The invention discloses a good product visual identification method and system, a medium and a program product, and relates to the field of image data processing, and the method comprises the steps: determining the detection type of a to-be-traced product; the method comprises the following steps: when a process detection request is requested, acquiring abnormal process information for triggering detection, calculating a process correlation degree between each completed process and the abnormal process, and generating a to-be-detected sequence; in a final inspection request, extracting defect feature data of a to-be-traced product, determining an associated process set from the process flow database, and sequencing to generate a to-be-detected sequence; sequentially acquiring to-be-detected process images of each process ranked in the front from the to-be-detected sequence, and inputting the to-be-detected process images into a parallel detection channel for detection; when the result of the parallel detection channel is passing, updating the to-be-detected sequence, and obtaining a first process image for detection; and when the result of the parallel detection channel is abnormal, determining an abnormal process and stopping the detection. By implementing the method, the problem positioning efficiency based on visual identification can be improved.

Description

Technical Field

[0001] The present application relates to the field of image data processing, and in particular to a method, system, medium and program product for visually identifying good products. Background Art

[0002] The production process of laptop computer shells involves multiple processes such as material selection, mold design, injection molding, processing, surface treatment, painting, quality inspection, etc. With the continuous growth of market demand for laptop computers, the scale of production continues to expand, and the production efficiency and quality control requirements of each process are also increasing. In the actual production process, due to the large number of process links and the interrelationships between processes, once a quality problem occurs, it is necessary to quickly locate the problem process and take corresponding measures to ensure the continued efficient operation of the production line.

[0003] In related technologies, multiple inspection points are set up on the notebook shell production line, and inspectors check the products according to the preset inspection procedures. When abnormalities are found, relevant information is recorded and possible causes are analyzed. At the same time, automatic inspection equipment is installed in key processes to monitor product appearance, size and other parameters in real time, and the inspection data is stored in the database for subsequent analysis.

[0004] However, as the scale of production continues to expand, especially when the production line is long, due to the process correlation between each process, problems in a certain process will be transmitted to subsequent processes, and the efficiency of problem location based on visual recognition is low. Summary of the invention

[0005] The present application provides a method, system, medium and program product for visually identifying good products, which are used to improve the efficiency of problem location based on visual recognition.

[0006] In a first aspect, the present application provides a method for visual recognition of good products, which is applied to a visual recognition system, and the method includes: in response to a product traceability application, determining the detection type of the product to be traced; the detection type includes a process detection request or a final inspection request; when the detection type is a process detection request, obtaining abnormal process information that triggers the detection; calculating the process correlation between each completed process and the abnormal process, and sorting each completed process based on the process correlation to generate a sequence to be detected; when the detection type is a final inspection request, extracting defect feature data of the product to be traced; based on the defect feature data, determining a set of associated processes from a process flow database, and sorting them to generate a sequence to be detected; starting multiple parallel detection channels, and sequentially obtaining the process images to be detected of each process ranked in the top row from the sequence to be detected, and inputting the process images to be detected into the parallel detection channel for detection; when the parallel detection channel completes the current process detection and the result is passed, updating the sequence to be detected, and obtaining the first process image in the updated sequence to be detected for detection; when the parallel detection channel detects a process abnormality, the corresponding process is determined as an abnormal process, and the detection of all parallel detection channels is stopped.

[0007] In the above embodiment, the visual recognition system is able to determine the inspection type, optimize the inspection sequence based on process correlation during process inspection, use defect feature data to spot-check related processes during final inspection, and simultaneously process the top-ranked process images by starting multiple parallel inspection channels, dynamically update the sequence when the inspection passes, and stop inspection in time when an abnormality is found, which significantly improves the efficiency of problem location and reduces unnecessary waste of inspection resources.

[0008] In combination with some embodiments of the first aspect, in some embodiments, before the step of determining the detection type of the product to be traced in response to a product traceability application, the method also includes: obtaining product monitoring data of multiple monitoring devices in the product production process; performing feature recognition on the product monitoring data, determining marked products whose product feature values ​​deviate from a preset threshold range, and generating abnormal prompt information; generating a product traceability application after a preset waiting time, or when receiving a product traceability instruction input by the user after viewing the abnormal prompt information.

[0009] In the above embodiment, the visual recognition system acquires product monitoring data from multiple monitoring devices in real time, compares the characteristic values ​​with preset thresholds, promptly detects anomalies and generates prompt information, and automatically triggers the traceability application after a preset waiting time or user confirmation, thereby achieving early warning and rapid response to anomalies.

[0010] In combination with some embodiments of the first aspect, in some embodiments, when the parallel detection channel completes the current process detection and the result is passed, the sequence to be detected is updated, and the image of the first process in the updated sequence to be detected is obtained for detection. The steps specifically include: when the parallel detection channel completes the current process detection and the result is passed, generating detection pass data based on the current process; based on the detection pass data, removing the associated processes in the sequence to be detected to obtain the optimized sequence to be detected; re-sorting the optimized sequence to be inspected, and obtaining the sorted first process image to the parallel detection channel for detection.

[0011] In the above embodiment, the visual recognition system generates detection pass data when the detection passes, and dynamically removes the associated processes in the sequence to be detected based on this data, and reorders the optimized sequence. This sequence optimization mechanism based on the detection results avoids repeated detection, improves detection efficiency, and ensures the integrity and accuracy of the detection.

[0012] In combination with some embodiments of the first aspect, in some embodiments, after the step of generating detection pass data based on the current process when the parallel detection channel completes the current process detection and the result is passed, the method also includes: when other parallel detection channels complete the corresponding current process detection and the result is passed, updating the detection pass data based on the corresponding current process detection.

[0013] In the above embodiment, the visual recognition system supports multiple parallel detection channels to update the detection data at the same time, realizes the real-time summary of the detection results, improves the concurrent processing capability of the system, and speeds up the problem location.

[0014] In combination with some embodiments of the first aspect, in some embodiments, before the step of removing associated processes in the sequence to be detected based on the detection pass data to obtain the optimized sequence to be detected, the method also includes: obtaining a confidence value of the detection pass data of the current process; when the confidence value is lower than a preset confidence threshold, obtaining the first process of the sequence to be detected to the parallel detection channel for detection, and adding the current process to the first position of the sequence to be detected.

[0015] In the above embodiment, the visual recognition system evaluates the reliability of the detection results through the confidence value, and automatically adjusts the detection order when the confidence is low, ensuring that key suspicious processes are inspected first, thereby improving the accuracy of problem location.

[0016] In combination with some embodiments of the first aspect, in some embodiments, the step of starting multiple parallel detection channels, and sequentially obtaining images of the process to be inspected of the top-ranked processes in the sequence to be inspected, and inputting the images of the process to be inspected into the parallel detection channels for inspection specifically includes: starting multiple parallel detection channels, and obtaining hardware parameter information corresponding to each parallel detection channel; performing computing power grading on each parallel detection channel according to the hardware parameter information to obtain a computing power grading result; sequentially obtaining images of the process to be inspected of the top-ranked processes in the sequence to be inspected, and calculating the estimated inspection time of the images of the process to be inspected; and performing channel allocation on the images of the process to be inspected based on the computing power grading result and the estimated inspection time.

[0017] In the above embodiment, the visual recognition system classifies the computing power of the detection channels based on the hardware parameters, and allocates channels according to the estimated detection time of the process images, thereby achieving load balancing of the detection tasks.

[0018] In combination with some embodiments of the first aspect, in some embodiments, when a parallel detection channel detects a process abnormality, the corresponding process is determined as an abnormal process, and after the step of stopping the detection of all parallel detection channels, the method also includes: acquiring the detection image and detection parameters of the abnormal process, and generating an abnormal analysis report; generating abnormal detection results and parameter adjustment suggestions based on the abnormal analysis report, and receiving process adjustment instructions input by the user; and adjusting the process parameters of the production equipment of the abnormal process according to the process adjustment instructions.

[0019] In the above embodiment, the visual recognition system automatically generates an abnormality analysis report and parameter adjustment suggestions, and supports parameter adjustment of production equipment through process adjustment instructions, thereby improving the automation level and response speed of the production line.

[0020] In the second aspect, an embodiment of the present application provides a visual recognition system, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the visual recognition system to execute the method described in the first aspect and any possible implementation method of the first aspect.

[0021] In a third aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when the computer program product is run on a visual recognition system, enables the visual recognition system to perform the method described in the first aspect and any possible implementation method of the first aspect.

[0022] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium comprising instructions. When the instructions are executed on a visual recognition system, the visual recognition system executes the method described in the first aspect and any possible implementation method of the first aspect.

[0023] It is understandable that the visual recognition system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the method provided in the embodiment of the present application. Therefore, the beneficial effects that can be achieved can refer to the beneficial effects in the corresponding method, which will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Due to the adoption of a dual-path detection strategy based on detection type and a combined mechanism of multi-channel parallel detection and dynamic sequence optimization, the system can select the optimal detection path according to different scenarios and improve detection efficiency through parallel processing, effectively solving the problems of single detection path and low serial processing efficiency in the existing technology, thereby achieving efficient utilization of detection resources and rapid location of problems; by determining the detection type to distinguish the processing path, optimizing the detection sequence based on process correlation during process detection, using defect feature data to spot-check related processes during final inspection, and starting multiple parallel detection channels to simultaneously process process images, the efficiency of problem location has been significantly improved.

[0025] 2. Due to the adoption of real-time feature recognition and automatic triggering mechanism based on monitoring data, the system can detect anomalies in the production process in a timely manner and automatically start the traceability process, effectively solving the problems of delayed abnormality detection and time-consuming manual intervention in the existing technology, thereby achieving early warning and rapid response to abnormalities; by continuously acquiring monitoring equipment data and performing feature recognition, products with abnormal feature values ​​are identified. When an abnormality is found, the system will automatically generate prompt information, improving the timeliness of abnormality handling.

[0026] 3. Due to the use of a sequence optimization mechanism and confidence assessment strategy based on test results, the system can dynamically adjust the test order and ensure the reliability of the test results, effectively solving the problem of fixed test sequence and repeated test that wastes resources in the prior art, thereby achieving a double improvement in test efficiency and accuracy. After passing the test, the test pass data will be generated and the sequence to be tested will be optimized accordingly, unnecessary related processes will be removed, resource waste will be avoided, and the reliability of the test will be ensured. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flow chart of a method for visually identifying good products in an embodiment of the present application; Figure 2is another flow chart of the method for visually identifying good products in an embodiment of the present application; Figure 3 It is a schematic diagram of the structure of a physical device of the visual recognition system in the embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to be used as limitations to the present application. As used in the specification of the present application, the singular expressions "one", "a kind of", "above", "the" and "this" are intended to also include plural expressions, unless there is a clear indication to the contrary in the context. It should also be understood that the term "and / or" used in the present application refers to any or all possible combinations comprising one or more of the listed items.

[0029] In the following, the terms "first" and "second" are used for descriptive purposes only and are not to be understood as suggesting or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, unless otherwise specified, "plurality" means two or more.

[0030] For ease of understanding, the application scenarios of the embodiments of the present application are introduced below.

[0031] The production process of laptop computer shells involves multiple processes such as injection molding, machining, surface treatment, and painting. With the growth of market demand and the expansion of production scale, production lines are getting longer and longer, and the correlation between processes is becoming more and more complex. For example, temperature control problems in the injection molding process may cause product size deviations, which in turn affect the subsequent surface treatment effects; abnormalities in the painting process may be caused by improper surface treatment in the previous process. When product quality problems are found, it is necessary to quickly locate the specific problem process and deal with it in a timely manner to avoid the continued production of defective products and cause greater losses. This puts higher requirements on quality monitoring and problem tracing during the production process.

[0032] In the related art, real-time monitoring of product quality can be achieved by adopting fixed inspection points and preset inspection procedures. Specifically, multiple manual inspection posts and automatic inspection equipment are arranged on the production line, and each inspection is carried out according to the predetermined inspection sequence and standards. When an abnormality is found, it is recorded and analyzed. The following introduces the scenario of using the visual recognition method of good products in the related art.

[0033] In the prior art, quality control is usually carried out by combining manual inspection with automatic inspection. Multiple inspection points are set up on the production line, and inspectors inspect the products according to preset inspection procedures. At the same time, automatic inspection equipment is installed in key processes such as injection molding and painting to monitor product appearance, size and other parameters in real time. When an abnormality is found, the inspector needs to judge the possible source of the problem based on experience and check the relevant processes one by one. For example, when scratches are found on the surface of the product, it is necessary to check multiple processes such as painting, polishing, and surface treatment in turn, and each process usually contains multiple sub-processes, involving a long time period before and after. This serial inspection method is inefficient and easy to miss, and it is difficult to meet the needs of rapid production.

[0034] The method for visually identifying good products in the embodiment of the present application is used to achieve rapid location of abnormalities and improve detection efficiency through dynamic process correlation analysis and multi-channel parallel detection mechanism. The following describes the scenario in which the method for visually identifying good products in the present application is used.

[0035] After adopting the visual recognition method of the present application, the system can automatically analyze the process correlation according to the specific characteristics of the anomaly and determine the process sequence where the problem is most likely to occur. For example, when scratches are detected on the surface of the product, the system will analyze the morphological characteristics of the scratches, and combine the empirical data in the process database to quickly determine that it may be related to the surface treatment and painting processes. Then start multiple parallel detection channels and detect these processes at the same time. If an abnormality is found in a detection channel, the system will immediately determine the problem process; if the test passes, the sequence to be tested will be dynamically updated. This intelligent parallel detection method significantly improves the efficiency of problem location.

[0036] It can be seen that the visual recognition method for good products in the embodiment of the present application can not only realize automatic monitoring of product quality, but also effectively solve the problem of low efficiency of traditional serial detection, thereby realizing intelligent quality control of the production process.

[0037] For ease of understanding, the following describes the process of the method provided by this implementation in combination with the above scenario. Figure 1 , which is a flow chart of the method for visually identifying good products in an embodiment of the present application.

[0038] S101. In response to a product traceability application, determine a test type for the product to be traced.

[0039] Among them, product traceability application refers to a traceability test request manually triggered by the monitoring system or the user during the product production process. The product to be traced represents a specific product instance that needs to be traced for quality, and can be uniquely identified by information such as product number and production batch. The test type is used to indicate the specific scenario of the current traceability test, including process test request and final inspection request. The process test request indicates a real-time test triggered when an abnormality is found during the product production process. The final inspection request indicates the final quality inspection performed after the product completes all production processes.

[0040] After receiving the product traceability application, the visual recognition system needs to first determine the detection type to select the appropriate detection strategy. Specifically, the visual recognition system will determine the current detection type based on the trigger source, current product status, and production stage information carried in the traceability application. If the traceability application comes from an abnormal monitoring trigger during the production process, and the product has not completed all production processes, it is determined to be a process detection request; if the traceability application comes from the quality inspection link after the product is completed, or the product has completed all production processes, it is determined to be a final inspection request. The visual recognition system will also verify the legitimacy of the traceability application, check the integrity of the necessary parameters, and record the timestamp and basis information for determining the detection type.

[0041] In some embodiments, the determination of the detection type can be achieved in a variety of ways: Optionally, the visual recognition system can establish a decision tree model, take the various characteristic parameters of the traceability application as input, and output the detection type after multiple layers of conditional judgment, specifically including: extracting the trigger time, product status, number of completed processes and other characteristic parameters in the traceability application; inputting these parameters into a pre-trained decision tree model; and determining the detection type based on the output results of the decision tree. Optionally, the visual recognition system can determine the detection type through a rule engine, specifically including: pre-setting a series of judgment rules and weights; performing feature extraction and rule matching on the traceability application; and calculating the final detection type based on the matching results and rule weights. It is understandable that other machine learning methods or expert systems can also be used to determine the detection type, which is not limited here.

[0042] In the process of determining the type of inspection, there may be situations where the traceability application information is incomplete or the parameter value is abnormal. In response to this, the visual recognition system uses multiple verification and fault-tolerant mechanisms: first, the preset parameter validity check rules are used to verify whether each parameter is within a reasonable range; for missing non-critical parameters, default values ​​or historical data are used to supplement; for abnormal parameter values, corrections are made based on contextual information and historical experience; when the inspection type cannot be accurately determined, a more stringent inspection strategy is given priority, that is, it is classified as a final inspection request. For example, when product status information is missing, the visual recognition system will query the production record of the product. If it is found that the process completion rate of the most recent record exceeds 95%, it will be judged as a final inspection request.

[0043] S102: When the detection type is a process detection request, obtain abnormal process information that triggers the detection.

[0044] Among them, process detection request indicates the detection type triggered when an abnormality is found in the product production process. Abnormal process information refers to the process data detected in the production process that does not meet the quality standards, including process ID, abnormal type, abnormal parameter value, occurrence time, etc. Triggered detection means the detection behavior initiated automatically by the abnormal monitoring system or manually by the operator. Process information contains complete data such as basic attributes of the process, quality parameters, equipment status, operation records, etc.

[0045] After determining that the current request is a process detection request, the visual recognition system needs to obtain the specific abnormal process information that caused this detection. Specifically, the visual recognition system will extract the complete record of the abnormal process from the database of the abnormal monitoring system, including the specific process link where the abnormality occurred, the specific manifestation of the abnormality, the degree of deviation of the abnormal parameters, the time point when the abnormality occurred, the duration of the abnormality, the relevant equipment status data, operator records, etc. At the same time, the visual recognition system will also collect the operation data of the related processes before and after the abnormal process to establish a complete abnormal context information set.

[0046] In some embodiments, the acquisition of abnormal process information can be achieved in a variety of ways: Optionally, the visual recognition system can monitor the quality parameters of each process on the production line in real time, and automatically record and upload abnormal information when the parameters deviate from the preset threshold, including setting multi-level warning thresholds, real-time acquisition of parameter data, calculation of parameter deviation, judgment of abnormal level, and recording of abnormal data; Optionally, the visual recognition system can establish an abnormal feature library and combine machine learning algorithms to identify abnormal patterns, including collecting historical abnormal data, extracting abnormal features, training recognition models, matching abnormal patterns in real time, and outputting abnormal information. It is understandable that other data collection and analysis methods can also be used to achieve the acquisition and processing of abnormal process information, which is not limited here.

[0047] In the process of acquiring abnormal process information, problems such as incomplete data collection or distorted abnormal data may occur. In response to this, the visual recognition system adopts a data completion and verification mechanism: by setting up multiple data collection channels to ensure the reliability of data collection; using data redundant storage to prevent data loss; establishing data consistency verification rules to identify and process abnormal data; when the data is incomplete, interpolation and supplementation are performed through similar working condition data. For example, when a sensor collecting a quality parameter fails, the system will automatically switch to a backup sensor and correct the collected abnormal data based on historical data.

[0048] S103: Calculate the process correlation between each completed process and abnormal process, and sort each completed process based on the process correlation to generate a sequence to be detected.

[0049] Among them, process correlation refers to the degree of mutual influence between different processes in the process flow, including quantitative indicators such as direct correlation between processes, correlation of process parameters, and transmission relationship of quality characteristics. Completed processes refer to all production processes that have been completed before the detection trigger point. Sorting refers to the priority arrangement of processes according to the calculation results of process correlation. The sequence to be detected refers to the process queue sorted by detection priority.

[0050] After obtaining the abnormal process information, the visual recognition system needs to calculate the degree of correlation between each process and the abnormal process to determine the source process that is most likely to cause the current abnormality. Specifically, the visual recognition system first extracts basic data such as the topological relationship between processes, the process parameter influence matrix, and the quality feature transfer chain from the process flow database. Then, combined with the specific characteristics of the abnormal process, the influence weight of each completed process on the abnormal process is calculated. The calculation process considers multiple dimensions such as direct influence, indirect influence, and cumulative influence between processes, and introduces a time series attenuation factor to adjust the degree of influence of processes completed in different time periods. Finally, the processes are sorted based on the calculated process correlation to generate a sequence to be detected.

[0051] In some embodiments, the calculation and ranking of process correlation can be achieved in a variety of ways: Optionally, the visual recognition system can establish a process correlation network model, and calculate the correlation strength between processes through a graph theory algorithm, specifically including: constructing a process correlation graph, setting edge weights, calculating the shortest path, evaluating the impact transfer, and summarizing the correlation strength; Optionally, the visual recognition system can use a multi-factor weighted scoring method, taking into account factors such as process parameter correlation, quality feature consistency, and historical abnormal correlation, specifically including: determining scoring factors, setting weight coefficients, calculating individual scores, weighted summation, and standardized processing. It is understandable that other mathematical models or algorithms can also be used to achieve quantitative calculations of process correlation, which are not limited here.

[0052] In the process of calculating process correlation, the problem of high calculation complexity caused by the large number of processes will be encountered. In this regard, the visual recognition system adopts a hierarchical calculation strategy: first, the processes are coarsely classified to identify the process categories with high correlation with abnormal features; then fine-grained correlation calculations are performed within the relevant process categories; finally, the calculation results are merged and normalized. At the same time, the system also uses parallel computing and caching mechanisms to improve computing efficiency. For example, for common abnormal types, the system will pre-calculate and store the process correlation template, and only local updates and adjustments are required during actual calculations.

[0053] S104: When the inspection type is a final inspection request, extract defect feature data of the product to be traced.

[0054] Among them, the final inspection request refers to the final quality inspection process after the product completes all production processes. Defect feature data refers to the quantitative description of various quality defects found in the final inspection process of the product, including multi-dimensional feature parameters such as defect type, defect location, defect size, and defect degree. The extraction process refers to the acquisition of a complete information set of defect features through image processing, data analysis, and other means. Products to be traced refer to specific individual products that require quality traceability analysis.

[0055] After the visual recognition system determines that it is a final inspection request, it needs to comprehensively collect and analyze the defect characteristics of the product. Specifically, the visual recognition system first calls the image acquisition equipment of the final inspection station to obtain multi-angle high-definition images of the product; then extracts surface defect characteristics through image processing algorithms, including visible defects such as color difference, scratches, bumps, bubbles, etc.; at the same time, it combines the data of other inspection methods such as size inspection and performance testing to form a complete defect feature data set. For each defect detected, the system will record its specific parameters and compare them with the product quality standards to evaluate the severity of the defect.

[0056] In some embodiments, defect feature data extraction can be achieved in a variety of ways: Optionally, the visual recognition system can use a deep learning model for defect detection and feature extraction, specifically including: preprocessing image data, using a pre-trained neural network model for defect recognition, extracting morphological features of defect areas, calculating defect parameters, and generating feature vectors; Optionally, the visual recognition system can construct a feature extraction process based on traditional image processing methods, specifically including: image enhancement, edge detection, region segmentation, feature calculation, and parameter statistics. It is understandable that other computer vision technologies or signal processing methods can also be used to achieve defect feature extraction and quantification, which is not limited here.

[0057] In the process of defect feature extraction, the problem of inconsistent feature expression of different types of defects will be encountered. In response to this, the visual recognition system adopts a standardized feature expression framework: establish a unified defect feature description system to map defect features from different sources and types to the standard feature space; design feature conversion rules to ensure that data from different inspection equipment can be effectively integrated; and construct a feature similarity measurement method to support comparative analysis of cross-type defects. For example, for products that contain both dimensional deviations and appearance defects, the system will uniformly express various defects as standardized feature vectors through feature conversion to facilitate subsequent correlation analysis.

[0058] S105 . Based on the defect feature data, determine a set of related processes from a process flow database, and sort them to generate a sequence to be inspected.

[0059] Among them, the process flow database refers to a data system that stores the complete production process information of the product, including process flow, process parameters, quality characteristics, equipment parameters and other data. The associated process set represents the combination of production processes that have a causal relationship or association relationship with the current defect characteristics. Process sorting refers to the priority ranking of processes based on the strength of the association. The sequence to be tested represents the process queue to be traced and analyzed after being sorted by priority.

[0060] After obtaining complete defect feature data, the visual recognition system needs to find out the key processes that may cause such defects from the historical production data. Specifically, the visual recognition system first matches the defect feature data with the defect pattern in the process knowledge base to identify the potential causative processes; then analyzes the parameter transfer relationship in the process flow to determine the impact mechanism of these processes on the current defects; then calculates the correlation strength between each related process and the defect, considering direct and indirect impacts; finally, sorts the processes according to the correlation strength to form a sequence to be detected with clear priority.

[0061] In some embodiments, process association analysis and sorting can be achieved in a variety of ways: Optionally, the visual recognition system can establish a defect-process association model, and analyze the association rules in the historical data through data mining methods, specifically including: constructing a feature-process association matrix, calculating support and confidence, extracting strong association rules, evaluating rule reliability, and generating a set of associated processes; Optionally, the visual recognition system can use causal reasoning methods to construct a causal network based on process expert knowledge, specifically including: establishing a process cause-effect diagram, setting conditional probabilities, performing probabilistic reasoning, calculating posterior probabilities, and determining key processes. It is understandable that other intelligent analysis methods can also be used to achieve the discovery and quantification of process associations, which are not limited here.

[0062] In the process of process association analysis, there will be problems with incorrect associations leading to inaccurate detection sequences. In response to this, the visual recognition system adopts a multiple verification mechanism: first, through historical case verification, check whether the analysis results are consistent with the known defect-process correspondence; then through process expert rule verification, ensure that the association results are consistent with the process logic; finally, through cross-validation, compare the results obtained by different analysis methods. For example, when the system finds that a process is highly associated with a specific defect, it will automatically retrieve the abnormal records of the process in the historical data to verify the universality and reliability of this association.

[0063] S106, start a plurality of parallel detection channels, and sequentially obtain the process images to be inspected of the top-ranked processes in the sequence to be inspected, and input the process images to be inspected into the parallel detection channels for inspection.

[0064] Among them, parallel detection channels refer to multiple processing units that can perform image detection tasks simultaneously, including a combination of hardware resources and software algorithms. The process image to be inspected refers to the image data of the process product that needs to be inspected for quality. The top ranking refers to the important processes selected according to the priority sorting results determined in the previous steps. Image input refers to the process of transferring process image data to the detection unit for processing.

[0065] After determining the sequence to be detected, the visual recognition system needs to efficiently complete the image detection tasks of multiple processes. Specifically, the visual recognition system first evaluates the available computing resources and initializes multiple parallel detection channels; then assigns the corresponding detection algorithm and parameter configuration to each detection channel; then extracts the process images from the sequence to be detected in order of priority, and assigns the images to the most suitable detection channel according to the image characteristics and detection requirements; at the same time, monitors the operating status of each detection channel, dynamically adjusts the task allocation strategy, and ensures maximum detection efficiency. The system also collects the detection progress and results of each channel in real time to ensure the synchronization and consistency of the data.

[0066] In some embodiments, the scheduling and execution of parallel detection can be achieved in a variety of ways: Optionally, the visual recognition system can adopt a dynamic load balancing strategy to allocate tasks according to the performance characteristics of the detection channel and the current load situation, including: evaluating image complexity, calculating detection difficulty, estimating processing time, selecting the optimal channel, and dynamically adjusting allocation; Optionally, the visual recognition system can establish a hierarchical detection framework to divide the detection task into subtasks of different granularities for parallel processing, including: task decomposition, resource allocation, parallel execution, result merging, and performance optimization. It is understandable that other parallel computing or task scheduling methods can also be used to achieve efficient image detection, which is not limited here.

[0067] During the parallel detection process, the problem of uneven performance of detection channels leading to reduced processing efficiency will be encountered. In response to this, the visual recognition system adopts an adaptive scheduling mechanism: real-time monitoring of the processing capacity and work efficiency of each detection channel; establishing a performance evaluation model to predict the processing time of different types of images; dynamically adjusting the task allocation strategy based on the evaluation results; and timely reallocation of tasks when performance bottlenecks occur. For example, when the processing speed of a certain detection channel is significantly lower than that of other channels, the system will automatically reduce the number of tasks assigned to the channel and transfer the tasks to channels with better performance.

[0068] S107, when the parallel detection channel completes the current process detection and the result is passed, the sequence to be detected is updated, and the first process image in the updated sequence to be detected is obtained for detection.

[0069] Among them, the test result of passing indicates that the product quality test of the current process meets the preset standards. Updating the sequence to be tested refers to dynamically adjusting the queue of processes to be tested according to the test results. The first process image represents the test image corresponding to the process with the highest priority in the updated sequence. The sequence update process includes operations such as deleting the processes that have passed the test, recalculating the priorities of the remaining processes, and adjusting the test order.

[0070] After obtaining the result of passing the inspection, the visual recognition system needs to adjust the subsequent inspection strategy in time. Specifically, the visual recognition system first verifies the reliability of the inspection results, including whether the inspection parameters are within the effective range, whether the inspection process is stable, and whether the confidence of the inspection results meets the standards; then analyzes the correlation between the current process and other processes in the sequence to be inspected, and evaluates the impact of the inspection on other processes; then updates the correlation weights between processes, and recalculates the inspection priority of each process; finally, generates an updated sequence to be inspected, and immediately starts the inspection of the new first process. The system will also save detailed information on the inspection for subsequent quality traceability analysis.

[0071] In some embodiments, dynamic updating of the sequence to be detected can be achieved in a variety of ways: Optionally, the visual recognition system can establish a process dependency network model and dynamically adjust the detection priority through a graph structure update algorithm, specifically including: constructing a process dependency graph, calculating node influence, updating edge weights, reordering sequences, and optimizing paths; Optionally, the visual recognition system can use a heuristic algorithm to adjust the detection strategy based on the detection history and empirical rules, specifically including: analyzing historical patterns, extracting decision rules, evaluating adjustment plans, executing sequence updates, and verifying update effects. It is understandable that other sequence optimization or scheduling algorithms can also be used to achieve efficient updates of the sequence to be detected, which are not limited here.

[0072] During the sequence update process, the problem of local optimum leading to decreased detection efficiency will be encountered. In response to this, the visual recognition system adopts a global optimization strategy: establish a multi-objective evaluation system, comprehensively consider factors such as detection efficiency, resource utilization, and result reliability; introduce a random perturbation mechanism to avoid falling into local optimum; set a dynamic adjustment threshold to improve processing speed while ensuring detection quality. For example, when the system finds that the detection efficiency has not been improved after multiple consecutive sequence updates, it will automatically trigger a global re-optimization mechanism to re-evaluate the priority of all processes to be inspected.

[0073] S108. When a parallel detection channel detects that a process is abnormal, the corresponding process is determined as an abnormal process, and detection of all parallel detection channels is stopped.

[0074] Among them, process abnormality means that the test results do not meet the preset quality standards or process requirements. Abnormal process refers to the specific production link where quality problems are found during the test. Stopping the test means interrupting the operation of all current test channels and saving the test status. The stopping process of parallel test channels includes tasks termination, state saving, resource release and other operations. The process of determining the test results includes steps such as abnormality assessment, confidence calculation, and multi-dimensional verification.

[0075] When the visual recognition system finds an abnormality in any detection channel, it needs to respond quickly and take corresponding measures. This step can be implemented at any time, that is, step S108 can be detected and executed after step S106, that is, after the first round of task allocation of the parallel detection channel, or after step S107, that is, after the second round of task allocation of the parallel detection channel; it can be simply deduced that step S107 can be plural, that is, this step can also be detected and executed after the Nth round of task allocation of the parallel detection channel. Specifically, the visual recognition system first verifies the reliability of the detected anomaly, including multiple repeated detections, cross-validation, threshold comparison, etc.; then evaluates the severity of the anomaly, including calculating the anomaly parameters, analyzing the scope of influence, and determining the anomaly level; then sends a stop command to all parallel detection channels to ensure that each channel stops the current detection task in an orderly manner; at the same time, saves the intermediate status of all detection channels and the completed detection results; finally generates an abnormal detection report, recording the complete information such as the time, location, characteristics, and severity of the abnormal discovery.

[0076] In some embodiments, anomaly detection and processing can be achieved in a variety of ways: Optionally, the visual recognition system can establish a multi-level anomaly detection model and identify complex abnormal patterns through deep learning algorithms, including: feature extraction, pattern matching, anomaly scoring, threshold judgment, result verification, and anomaly confirmation; Optionally, the visual recognition system can use statistical process control methods to establish anomaly discrimination criteria based on historical data, including: data collection, statistical analysis, control limit calculation, trend monitoring, anomaly identification, and result output. It is understandable that other intelligent detection or anomaly identification methods can also be used to achieve accurate judgment of process anomalies, which is not limited here.

[0077] During the exception handling process, there will be problems with false exceptions leading to unnecessary detection interruptions. To this end, the visual recognition system adopts multiple anti-error mechanisms: establish an exception feature library to distinguish common exceptions from interference factors; set up a multi-level exception confirmation process, requiring high-confidence exceptions to trigger stop operations; implement a fast recovery mechanism to allow for rapid recovery of detection in the event of a false alarm; save complete decision-making basis to support post-analysis of exception judgments. For example, when a detection channel reports an exception, the system will immediately start cross-validation, confirm the authenticity of the exception through other detection methods or historical data comparison, and only execute the stop operation when multiple verifications confirm the existence of an exception.

[0078] In the above embodiment, the detection priority is determined by process correlation analysis, and multi-channel parallel detection is used to improve processing efficiency. In actual applications, the system will also dynamically update the sequence to be inspected according to the real-time detection results to ensure the optimal configuration of detection resources. The following is a supplement to the scenario of this embodiment.

[0079] In further applications, the visual recognition system can also be combined with the real-time parameters of production equipment for early warning. For example, when the temperature, pressure and other parameters of the surface treatment equipment fluctuate, the system will automatically record the relevant product numbers and focus on the surface quality of these products in subsequent inspections. At the same time, the system will continuously accumulate empirical data based on the test results and optimize the calculation model of process correlation. For the processes that pass the inspection, the system will record their process parameter combinations to form the best process window; for the processes that are found to be abnormal, the system will analyze the causes of the abnormalities and automatically generate parameter adjustment suggestions. This intelligent quality control system has greatly improved production efficiency and product yield.

[0080] After combining the above scenarios, the following is a more detailed description of the process of the method provided by this implementation. Figure 2 , is another flow chart of the method for visually identifying good products in an embodiment of the present application.

[0081] S201. Obtain product monitoring data from multiple monitoring devices in a product production process.

[0082] Among them, the product production process refers to the complete manufacturing process of the laptop shell from raw materials to finished products, including injection molding, processing, surface treatment, painting and other processes. Monitoring equipment refers to data acquisition devices installed in each process of the production line, including image collectors, temperature sensors, pressure sensors, etc. Product monitoring data refers to information such as process parameters, image data and quality indicators collected in real time by each monitoring device, which is used to reflect the real-time status of the product production process. Multiple monitoring devices refer to a collection of multiple types of data acquisition devices arranged in different processes and locations.

[0083] The visual recognition system needs to continuously obtain monitoring data during the production process to achieve real-time monitoring of product quality. Specifically, the visual recognition system first determines the acquisition parameters and data format of each monitoring device; then establishes a data communication connection with each device, sets the data acquisition frequency and cache strategy; then obtains monitoring data from each device according to the preset acquisition rules, including image data, process parameters, etc.; at the same time, pre-processes the collected data, including data cleaning, format conversion, time synchronization and other operations; finally, stores the processed data in a database in a unified format, and creates an index for subsequent analysis.

[0084] In some embodiments, the collection and processing of monitoring data can be achieved in a variety of ways: Optionally, the visual recognition system can adopt a distributed data collection architecture to obtain monitoring data by deploying local collection nodes in each process and centrally summarizing them, specifically including: configuring the collection parameters of local collection nodes, establishing a communication network between nodes, implementing data collection, local preprocessing, data transmission, and central aggregation; Optionally, the visual recognition system can build a data collection system based on the industrial Internet of Things platform, specifically including: accessing Internet of Things devices, configuring collection protocols, establishing data models, collecting data, real-time processing, and cloud storage. It is understandable that other data collection and processing methods can also be used to achieve the acquisition of monitoring data, which is not limited here.

[0085] During the data collection process, there will be problems with data loss due to device communication interruption. In response to this, the visual recognition system adopts multiple fault-tolerant mechanisms: establish a backup channel for data collection, and automatically switch to the backup channel when the main channel fails; set up data cache at the local node to temporarily store data when the network is disconnected and automatically retransmit it after recovery; set data integrity verification rules to interpolate and repair lost data; record data anomaly logs to support traceability analysis of the data collection process. For example, when the image acquisition device of a certain process is interrupted in communication, the system will automatically switch to the backup camera to continue collecting, and temporarily store the process parameter data during the interruption locally, and retransmit and verify the data after communication is restored.

[0086] S202: Perform feature recognition on product monitoring data, determine marked products whose product feature values ​​deviate from a preset threshold range, and generate abnormal prompt information.

[0087] Among them, feature recognition means analyzing and processing monitoring data to extract key features that reflect the quality status of the product. Product feature values ​​refer to various indicator parameters that characterize product quality, including quantitative indicators such as size, appearance, and performance. The preset threshold range represents the normal fluctuation range of each feature value, which is used to determine whether the product is abnormal. Marked products refer to product instances that have been identified as having abnormal feature values. Abnormal prompt information is used to record and display specific information about abnormal products, including abnormality type, abnormality degree, and occurrence time.

[0088] The visual recognition system needs to analyze the collected monitoring data in real time to detect abnormal products in time. Specifically, the visual recognition system first reads the latest monitoring data from the database, including image data and process parameters; then preprocesses the data, including denoising, enhancement, standardization and other operations; then uses the feature extraction algorithm to analyze the data and calculate various feature values; compares the calculated feature values ​​with the preset threshold range; when it is found that the feature value exceeds the threshold range, records the relevant product information and determines the alarm level according to the degree of abnormality; finally generates a prompt message containing the details of the abnormality and pushes it to the relevant personnel.

[0089] In some embodiments, feature recognition and anomaly judgment can be achieved in a variety of ways: Optionally, the visual recognition system can use a deep learning model for feature extraction and anomaly detection, specifically including: image preprocessing, feature extraction, pattern recognition, anomaly scoring, threshold judgment, and result output; Optionally, the visual recognition system can build an anomaly detection framework based on statistical process control methods, specifically including: data collection, statistical analysis, control limit calculation, trend monitoring, rule judgment, and anomaly marking. It is understandable that other feature analysis and anomaly detection methods can also be used, which are not limited here.

[0090] In the process of feature recognition, there will be problems with false positives and false negatives. To this end, the visual recognition system adopts a multiple verification mechanism: establish a multi-level threshold system, adopt different processing strategies for different degrees of anomalies; introduce time series correlation analysis, and combine historical data to judge the reliability of anomalies; set cross-validation rules to improve accuracy through the combination of multiple features; record the basis for judgment and support the later review of abnormal judgments. For example, when suspected scratches are detected on the surface of a product, the system will simultaneously analyze multiple dimensions such as regional grayscale value, edge features, texture features, etc., and combine the judgment results of similar cases in historical data to finally determine whether to mark the product as abnormal.

[0091] S203: After a preset waiting time, or when a product tracing instruction is received from a user after viewing abnormal prompt information, a product tracing application is generated.

[0092] Among them, the preset waiting time refers to the time interval for the system to automatically trigger the traceability application, which is used to automatically start the traceability process without human intervention. Abnormal prompt information refers to the abnormal product alarm information generated by the system. The product traceability instruction refers to the traceability detection command issued after the user confirms to view the abnormal information. The product traceability application refers to the formal request for starting the traceability detection process, which includes abnormal information, triggering method, timestamp, etc.

[0093] The visual recognition system needs to choose the appropriate time to generate a traceability application according to different scenarios. Specifically, the visual recognition system first sets a preset waiting time to control the time interval for automatic triggering; when the system generates an abnormal prompt message, it starts the waiting timer; during the waiting period, if the user's traceability instruction is received, the traceability application is immediately generated; if the user's instruction is not received after the preset waiting time, the traceability application is automatically generated; at the same time, the triggering method, time point and related abnormal information of the traceability application are recorded; the traceability application is stored in the database and the relevant modules are notified to start the detection process.

[0094] In some embodiments, the generation management of traceability applications can be achieved in a variety of ways: Optionally, the visual recognition system can use a state machine mechanism to manage the generation process of traceability applications, including: anomaly detection, waiting time, trigger judgment, application generation, state transition, and process control; Optionally, the visual recognition system can build an application trigger framework based on an event-driven model, including: event monitoring, condition judgment, priority evaluation, resource inspection, application packaging, and task distribution. It is understandable that other process management or task scheduling methods can also be used, which are not limited here.

[0095] In the process of generating traceability applications, a large number of exceptions in a short period of time will cause the application to pile up. In this regard, the visual recognition system adopts an intelligent scheduling mechanism: establish a traceability application priority evaluation model, determine the processing order according to factors such as the degree of exception and the scope of influence; set application merging rules to batch process exceptions in the same process and similar time points; implement a resource pre-inspection mechanism to dynamically adjust the time interval for automatic triggering according to the system load; record application processing logs to support optimization analysis of the traceability process. For example, when similar surface defects appear on multiple products at the same time, the system will perform cluster analysis on these exceptions and generate a batch traceability application to improve processing efficiency.

[0096] S204: In response to the product traceability application, determine the test type of the product to be traced.

[0097] Referring to step S101 , the visual recognition system determines the detection type.

[0098] It should be noted that the determination of the detection type involves a complex decision-making analysis process. The visual recognition system adopts a hierarchical decision model to construct a detection demand evaluation function by comprehensively evaluating factors such as the characteristics of the abnormal type, the production stage of the product, and historical detection records. Based on the evaluation results, the system can also calculate the applicability scores of the two types of process detection and final inspection, and select the type with the higher score. For example, when a local scratch is detected on the surface of the shell, the system will analyze the formation characteristics, location distribution and historical cases of the defect. If it is found that similar defects are usually caused by specific processes, it tends to choose the process detection type.

[0099] S205. When the detection type is a process detection request, obtain abnormal process information that triggers the detection.

[0100] Referring to step S102 , the visual recognition system determines abnormal process information when a process inspection is requested.

[0101] S206: Calculate the process correlation between each completed process and the abnormal process, and sort each completed process based on the process correlation to generate a sequence to be detected.

[0102] Referring to step S103 , the visual recognition system generates a sequence to be detected.

[0103] It should be noted that the process correlation calculation can use an improved Markov random field model to represent the correlation between processes as a conditional probability network. The visual recognition system builds a complete correlation evaluation system by establishing the conditional influence probability and process weight matrix between processes, analyzing the process parameter transfer chain and the evolution law of quality characteristics. In practical applications, the system will consider the direct influence and indirect transmission effects between processes, and dynamically adjust the correlation calculation parameters based on the abnormal correlation patterns in historical data. For example, during the shell processing process, if the finishing process is abnormal, the system will analyze the correlation of the previous rough processing process and consider the influence weights of factors such as tooling and processing parameters.

[0104] S207: When the inspection type is a final inspection request, extract defect feature data of the product to be traced.

[0105] Referring to step S104 , the visual recognition system will extract defect feature data when a final inspection is requested.

[0106] S208. Based on the defect feature data, determine a set of related processes from a process flow database, and sort them to generate a sequence to be inspected.

[0107] Referring to step S105 , the visual recognition system generates a sequence to be detected.

[0108] It should be noted that the defect feature data analysis adopts a multimodal feature fusion method. The feature expression model constructed by the visual recognition system includes multiple dimensions such as visual features, texture features, and position features. These features are extracted through a deep learning network, and a mapping relationship with process parameters is established to trace possible problem processes. This mapping relationship takes into account the importance weights of different feature dimensions, and continuously optimizes feature extraction and mapping rules based on successful cases in historical data. For example, when the final inspection finds that there are bubble defects on the surface of the shell, the system will analyze the morphological characteristics, spatial distribution characteristics, and depth characteristics of the bubbles, and locate possible problem processes through feature-process mapping relationships.

[0109] S209, start multiple parallel detection channels, and sequentially obtain the process images to be inspected of each process ranked top in the sequence to be inspected, and input the process images to be inspected into the parallel detection channels for inspection.

[0110] Referring to step S106 , the visual recognition system will perform recognition detection.

[0111] It should be noted that the scheduling of parallel detection channels adopts a dynamic load balancing algorithm. The visual recognition system establishes a resource allocation model by evaluating the complexity of the detection task and the resource occupancy. The model achieves the optimal allocation of detection resources by minimizing the load difference between channels. At the same time, the system implements an adaptive task allocation strategy, which comprehensively considers factors such as the current load status, detection quality requirements, and processing priority. For example, when multiple processes need to be detected at the same time, the system will intelligently allocate detection tasks based on the image feature complexity and processor load of each process.

[0112] In some embodiments, the visual recognition system will allocate detection tasks according to the hardware performance of each parallel detection channel and make rational use of detection resources. That is, the visual recognition system will start multiple parallel detection channels and obtain the hardware parameter information corresponding to each parallel detection channel; perform computing power classification on each parallel detection channel according to the hardware parameter information to obtain a computing power classification result; obtain the images of the processes to be inspected of the top-ranked processes in the sequence to be inspected in turn, and calculate the estimated detection time of the images of the processes to be inspected; and perform channel allocation on the images of the processes to be inspected based on the computing power classification result and the estimated detection time.

[0113] Among them, the hardware parameter information represents the computing performance indicators of the parallel detection channel, including processor speed, memory capacity, cache size, etc. The computing power classification represents the classification result of the processing capacity of the detection channel. The estimated detection time refers to the estimated time required to complete the process image detection. Channel allocation refers to the reasonable allocation of detection resources according to task requirements and channel capabilities.

[0114] The visual recognition system needs to perform resource assessment and task allocation before starting the inspection. Specifically, the visual recognition system first obtains the hardware configuration information of all inspection channels; then evaluates the channels according to performance indicators; then analyzes the feature complexity of each process image in the sequence to be inspected; estimates the time required for inspection based on image features; matches and analyzes the inspection task with the channel capacity; formulates the optimal task allocation plan; and reserves some high-performance channels for emergency tasks; finally, executes task allocation and starts the inspection process; continuously monitors the inspection progress and makes dynamic adjustments when necessary.

[0115] In some embodiments, resource evaluation and task allocation can be achieved in a variety of ways: Optionally, the visual recognition system can use performance modeling methods for resource management, including hardware performance evaluation, load capacity calculation, resource pool division, task demand analysis, matching degree calculation, and allocation strategy generation; Optionally, the visual recognition system can optimize the detection process based on the task scheduling algorithm, including task decomposition, dependency analysis, priority evaluation, resource reservation, load balancing, and real-time scheduling. It is understandable that other resource management or scheduling algorithms can also be used to improve detection efficiency, which is not limited here.

[0116] In the process of resource allocation, the problem of uneven distribution of task loads leading to low resource utilization will be encountered. In response to this, the visual recognition system adopts a dynamic load balancing mechanism: establish a real-time load monitoring system to track the resource usage of each channel; implement a dynamic task migration strategy to flexibly allocate tasks between channels; set load threshold control to prevent overload of a single channel; record resource utilization data to support the optimization and adjustment of allocation strategies. For example, when a channel is found to be overloaded during the detection process, the visual recognition system will dynamically migrate some tasks to a channel with a lower load to ensure the optimization of the overall detection efficiency.

[0117] S210: When the parallel inspection channel completes the inspection of the current process and the result is passed, generate inspection pass data based on the current process.

[0118] Among them, the parallel detection channel refers to multiple processing units that perform image detection at the same time. The current process refers to the production link that is being tested. The test result of passing indicates that the product quality of the current process meets the standard requirements. The test pass data refers to the complete information set that records the test pass situation, including process information, test parameters, confidence, etc.

[0119] The visual recognition system needs to process the test results in a timely manner and generate corresponding data. Specifically, the visual recognition system first verifies the reliability of the test results, including checking whether the test parameters are within the valid range and whether the test process is stable and complete; then records the detailed information of the test, including process ID, test time, test parameters, image features, etc.; then calculates the confidence of the test results, considering test conditions, parameter stability and other factors; packages the test data in a standard format; at the same time, updates the process status and marks that the process has completed the test; finally, stores the test data in the database for subsequent analysis and optimization.

[0120] In some embodiments, the generation and management of detection data can be achieved in a variety of ways: Optionally, the visual recognition system can use a multi-dimensional feature analysis method to evaluate the detection results, including: parameter verification, feature extraction, stability analysis, confidence calculation, data encapsulation, and result storage; Optionally, the visual recognition system can build a detection result evaluation system based on the knowledge graph, including: feature mapping, rule matching, association analysis, reliability evaluation, data generation, and status update. It is understandable that other data analysis or result evaluation methods can also be used, which are not limited here.

[0121] In the process of generating test data, the problem of unstable test results leading to difficulty in confidence assessment will be encountered. In response to this, the visual recognition system adopts a dynamic evaluation mechanism: establish a multi-dimensional stability evaluation model, comprehensively consider test parameters, environmental factors, equipment status, etc.; set dynamic threshold rules, and adaptively adjust the judgment criteria based on historical data; introduce time series analysis to evaluate the trend changes of test results; record the evaluation process to support traceability analysis of the reliability of the results. For example, when the test results of a certain process fluctuate greatly, the system will increase the number of sampling times and analyze the reasons for the fluctuations in combination with historical data to ensure that the generated test data is sufficiently reliable.

[0122] In some embodiments, the visual recognition system will summarize the inspection results and record the information of passed inspection in the inspection pass data to make the optimization more accurate. That is, the visual recognition system will update the inspection pass data based on the corresponding current process inspection when other parallel inspection channels complete the corresponding current process inspection and the result is passed.

[0123] Among them, the parallel detection channel refers to multiple independent processing units that perform detection tasks at the same time, which is used to improve detection efficiency. The current process refers to the production link that is being tested. The test result of passing indicates that the process quality inspection meets the preset standard requirements. The test pass data is used to record the test information of the qualified process, including process parameters, test time, test results, etc. The update operation means adding the new test results to the existing test pass data set.

[0124] The visual recognition system needs to update the inspection record in time when any parallel inspection channel completes the inspection and the result is qualified. Specifically, the visual recognition system first confirms the inspection status of the inspection channel and verifies the integrity of the inspection process; then conducts a reliability assessment on the inspection results to ensure the accuracy of the inspection data; then extracts the inspection parameters and result data of the current process; packages the new inspection data in a standard format; updates the relevant records in the inspection database; and updates the process status mark to indicate that the inspection of the process is completed; finally, triggers the subsequent data analysis and optimization process.

[0125] In some embodiments, the update processing of the detection data can be implemented in a variety of ways: Optionally, the visual recognition system can use a distributed data synchronization mechanism to update the detection record, including data verification, incremental update, state synchronization, consistency check, version management, and log record; Optionally, the visual recognition system can manage the data update process based on an event-driven model, including event triggering, data encapsulation, update strategy formulation, data merging, integrity verification, and state push. It is understandable that other data management or update methods can also be used to implement the update of the detection data, which is not limited here.

[0126] S211 . Based on the detection passing data, remove the associated processes in the sequence to be detected to obtain the optimized sequence to be detected.

[0127] Among them, the test pass data refers to the complete test information of the process that has been confirmed to be qualified. The associated process refers to other processes that have a process dependency or influence relationship with the process that has passed the test. The sequence to be tested refers to the process queue that has not yet completed the test. The optimized sequence to be tested refers to the process queue to be tested after optimization and adjustment.

[0128] The visual recognition system needs to dynamically optimize the sequence to be inspected based on the inspection results. Specifically, the visual recognition system first analyzes the process characteristics of the inspection process and identifies its key impact in the production process; then extracts the correlation between the processes from the process database and builds an influence transmission network; calculates the degree of correlation between other processes to be inspected and the current process; analyzes the processes with correlation exceeding the threshold and evaluates whether they can be removed from the sequence to be inspected; generates a new optimized sequence to ensure the integrity of the inspection coverage; and records the optimization results in the database for subsequent inspection optimization.

[0129] In some embodiments, sequence optimization can be achieved in a variety of ways: Optionally, the visual recognition system can use a process dependency graph analysis method, specifically including: building a dependency network, calculating influence strength, determining optimization strategy, sequence adjustment, integrity verification, and result update; Optionally, the visual recognition system can build a sequence optimization model based on a heuristic algorithm, specifically including: feature extraction, correlation analysis, optimization target setting, sequence reconstruction, verification evaluation, and dynamic update. It is understandable that other optimization algorithms or decision-making methods can also be used, which are not limited here.

[0130] During the sequence optimization process, there will be problems with missed detections due to over-optimization. In response to this, the visual recognition system adopts a conservative optimization strategy: establish a process impact assessment model to strictly control the conditions for process removal; set up multi-level verification rules to ensure that optimization does not affect the reliability of detection; retain key node processes and keep them in the detection sequence even if there is a correlation; record the basis for optimization and support retrospective analysis of sequence adjustments. For example, when the system determines that certain processes can be removed, it will first evaluate the historical abnormality probability of these processes, and only perform the removal operation when the abnormality probability is extremely low and highly correlated with the processes that have passed the inspection.

[0131] In some embodiments, the visual recognition system will perform a reliability assessment on the detection results and initiate a re-inspection mechanism for results with low confidence levels. That is, the visual recognition system will first obtain the confidence value of the detection pass data of the current process; then, when the confidence value is lower than a preset confidence threshold, the first process in the sequence to be detected is obtained and sent to the parallel detection channel for detection, and the current process is added to the first position in the sequence to be detected.

[0132] The confidence value represents a quantitative indicator of the reliability of the test result, which is used to evaluate the credibility of the test quality. The preset confidence threshold refers to the standard value for determining whether the test result is reliable. The first position in the sequence to be tested represents the process to be tested with the highest priority. Adding to the first position means adjusting the process to the highest priority position in the sequence to be tested.

[0133] The visual recognition system needs to evaluate the reliability of the detection results and trigger re-detection when the reliability is insufficient. Specifically, the visual recognition system first extracts the confidence index of the current process from the detection data; then compares the confidence value with the preset threshold; when the confidence is lower than the threshold, the current process is marked as needing re-detection; then the process is added to the front of the sequence to be detected; at the same time, the original first process is obtained from the sequence to be detected; the detection tasks of the two processes are respectively assigned to the appropriate detection channels; a new detection process is started; and finally, the process status and detection plan are updated.

[0134] In some embodiments, confidence assessment and re-inspection processing can be achieved in a variety of ways: Optionally, the visual recognition system can adopt a multi-dimensional confidence assessment method, including feature matching analysis, detection stability assessment, environmental impact assessment, historical data comparison, comprehensive score calculation, and decision threshold judgment; Optionally, the visual recognition system can manage the detection process based on an adaptive re-inspection strategy, including abnormality assessment, resource status check, priority recalculation, channel reservation, task reallocation, and status synchronization. It is understandable that other evaluation methods or re-inspection strategies can also be used to achieve detection quality assurance, which is not limited here.

[0135] During the confidence assessment and re-inspection process, repeated inspections may lead to reduced inspection efficiency. To address this, the visual recognition system uses an intelligent re-inspection optimization mechanism: establish an adaptive adjustment strategy for inspection parameters, and optimize re-inspection parameters based on the first inspection results; implement an inspection resource reservation mechanism to ensure timely execution of re-inspection tasks; set a maximum re-inspection limit to avoid invalid repetitions; record re-inspection reason analysis to support continuous optimization of inspection strategies. For example, when the confidence of the first inspection of a process is insufficient, the visual recognition system will analyze the specific factors that affect the confidence, and adjust the inspection parameters or switch the inspection method in a targeted manner to improve the success rate of re-inspection.

[0136] S212, re-arrange the optimized sequence to be inspected, and obtain the sorted first process image to the parallel inspection channel for inspection.

[0137] Among them, reordering means adjusting the inspection order of the processes in the inspection optimization sequence according to a specific strategy. The first process refers to the process to be inspected with the highest priority after sorting. The process image refers to the product image data corresponding to the process to be inspected. The parallel inspection channel refers to a set of multiple processing units that can perform image inspection tasks simultaneously.

[0138] The visual recognition system needs to scientifically and reasonably arrange the inspection sequence and allocate inspection resources. Specifically, the visual recognition system first analyzes the characteristics of each process in the optimized sequence to be inspected, including process type, abnormal risk, inspection difficulty, etc.; then calculates the inspection priority of each process based on multi-dimensional evaluation indicators, considering factors such as process importance, resource consumption, and time constraints; sorts the sequence based on priority to generate the final inspection sequence; allocates inspection resources to the first process after sorting, including determining the inspection channel to be used, loading inspection parameters, etc.; retrieves the image data of the first process from the image database; distributes the image data to the designated parallel inspection channel, and starts the inspection process.

[0139] In some embodiments, detection sorting and resource allocation can be achieved in a variety of ways: Optionally, the visual recognition system can use a multi-objective optimization algorithm to make sorting decisions, including: feature quantification, weight calculation, priority evaluation, sequence rearrangement, resource allocation, and task dispatching; Optionally, the visual recognition system can build a detection scheduling model based on a dynamic programming method, including: state evaluation, objective function construction, optimal path solution, sequence generation, resource scheduling, and task initiation. It is understandable that other scheduling algorithms or resource allocation methods can also be used, which are not limited here.

[0140] During the sorting and detection allocation process, the problem of unbalanced detection resource load will be encountered. In response to this, the visual recognition system adopts an adaptive scheduling mechanism: establish a detection channel load evaluation model to monitor the resource occupancy of each channel in real time; set load balancing rules to dynamically adjust the task allocation strategy according to the channel status; introduce a task estimation mechanism to plan resource allocation plans in advance; record the scheduling process to support the optimization analysis of resource utilization efficiency. For example, when the processing queue of a certain detection channel is long, the system will select a channel with a lighter load and suitable for this type of detection to perform the task based on the detection characteristics of the first process, thereby achieving rational use of resources. In addition, the system will dynamically adjust the parameters of the priority calculation model based on the feedback of the detection results and continuously optimize the sorting strategy.

[0141] S213. When a parallel detection channel detects that a process is abnormal, the corresponding process is determined as an abnormal process, and detection of all parallel detection channels is stopped.

[0142] Referring to step S108, the visual recognition system will stop the detection when it determines that there is an abnormality in the process. Similarly, this step can be implemented at any time, that is, step S213 can be detected and executed after step S209, that is, after the first round of task allocation of the parallel detection channel, or after step S212, that is, after the second round of task allocation of the parallel detection channel; it can be simply deduced that steps S210 to S212 can be plural, that is, this step can also be detected and executed after the Nth round of task allocation of the parallel detection channel.

[0143] It should be noted that the termination judgment of anomaly detection can adopt a quick response mechanism. The anomaly propagation model constructed by the visual recognition system takes into account multiple dimensions such as defect feature evolution, detection confidence and impact range, and realizes real-time evaluation of abnormal status. The model can quickly analyze the reliability and potential impact of anomalies, and trigger termination instructions when it is confirmed that the impact of anomalies exceeds the allowable range. For example, when a critical dimension of a process is detected to be out of tolerance, the system will evaluate the impact of the anomaly on subsequent processes, analyze the risk of anomaly propagation, and promptly terminate the operation of all detection channels after confirming the reliability of the anomaly.

[0144] In some embodiments, the visual recognition system will perform intelligent analysis of abnormal situations and provide processing suggestions to optimize and adjust production equipment, that is, the visual recognition system will obtain detection images and detection parameters of abnormal processes and generate an abnormal analysis report; generate abnormal detection results and parameter adjustment suggestions based on the abnormal analysis report, and receive process adjustment instructions input by the user; and adjust the process parameters of production equipment for abnormal processes according to the process adjustment instructions.

[0145] Among them, the abnormal analysis report refers to the detailed record and analysis results of the abnormal conditions found in the detection. The parameter adjustment suggestion refers to the process parameter optimization plan given based on the abnormal analysis. The process adjustment instruction refers to the specific parameter adjustment requirements after the user's confirmation. The production equipment refers to the specific equipment that performs the process processing.

[0146] The visual recognition system needs to analyze and diagnose after discovering anomalies and guide process adjustments. Specifically, the visual recognition system first collects the detection images and parameter data of the abnormal process; then extracts and analyzes the abnormal features; generates an analysis report containing abnormal descriptions, cause analysis, and impact assessment; formulates parameter adjustment suggestions based on the causes of the abnormalities; provides the analysis results and adjustment suggestions to the user; receives and verifies the user's adjustment instructions; and finally executes the process parameter adjustment operation; and monitors the adjusted production status at the same time.

[0147] In some embodiments, abnormal analysis and parameter adjustment can be achieved in a variety of ways: Optionally, the visual recognition system can use a knowledge graph method to perform abnormal diagnosis, including feature extraction, pattern matching, causal analysis, impact assessment, solution generation, and verification evaluation; Optionally, the visual recognition system can optimize parameter adjustment based on a feedback control model, including state assessment, target setting, deviation analysis, parameter calculation, adjustment execution, and effect verification. It is understandable that other analysis methods or optimization strategies can also be used to achieve process improvement, which is not limited here.

[0148] During the parameter adjustment process, the problem of unstable adjustment effect may occur. In response to this, the visual recognition system adopts a progressive adjustment strategy: establish a parameter sensitivity analysis model to evaluate the adjustment impact of each parameter; implement a step-by-step adjustment mechanism to achieve the optimization goal through multiple small adjustments; set parameter adjustment range limits to ensure the safety of the adjustment; record the adjustment process data to support the continuous optimization of the adjustment strategy. For example, when it is necessary to adjust the temperature parameters of the injection molding process, the visual recognition system will first conduct a small-scale trial adjustment, and gradually optimize according to the effect feedback until the expected improvement effect is achieved.

[0149] In the embodiment of the present application, due to the use of a dual-path detection strategy based on process correlation and a multi-channel parallel detection mechanism, it is possible to intelligently analyze possible problem processes based on abnormal characteristics, and simultaneously carry out detection through parallel channels, effectively solving the problems of single detection path and low efficiency of serial processing in traditional technologies, thereby achieving efficient use of detection resources and rapid location of problems. At the same time, the system also has functions such as real-time monitoring and early warning, dynamic sequence optimization, and abnormal analysis and diagnosis, which can promptly detect potential problems and give improvement suggestions, significantly improving the automation level of the production line and product quality control capabilities. Through this intelligent visual recognition method, not only the efficiency of problem location is improved and the waste of detection resources is reduced, but also data support is provided for process optimization, achieving continuous improvement of the production process.

[0150] The visual recognition system in the embodiment of the present invention is described below from the perspective of hardware processing. Figure 3 , which is a schematic diagram of a physical device structure of a visual recognition system in an embodiment of the present application.

[0151] It should be noted that Figure 3 The structure of the visual recognition system shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0152] like Figure 3 As shown, the visual recognition system includes a CPU 301, which can perform various appropriate actions and processes according to the program stored in the ROM 302 or the program loaded from the storage part 308 into the RAM 303, such as executing the method described in the above embodiment. In the RAM 303, various programs and data required for system operation are also stored. The CPU 301, the ROM 302 and the RAM 303 are connected to each other via a bus 304. An I / O interface 305 is also connected to the bus 304.

[0153] The following components are connected to the I / O interface 305: an input section 306 including an audio input device, a button switch, etc.; an output section 307 including a liquid crystal display (LCD) and an audio output device, an indicator light, etc.; a storage section 308 including a hard disk, etc.; and a communication section 309 including a network interface card such as a LAN (Local Area Network) card, a modem, etc. The communication section 309 performs communication processing via a network such as the Internet. A drive 310 is also connected to the I / O interface 305 as needed. A removable medium 311, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 310 as needed so that a computer program read therefrom is installed into the storage section 308 as needed.

[0154] In particular, according to an embodiment of the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer readable medium, and the computer program includes a computer program for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 309, and / or installed from the removable medium 311. When the computer program is executed by the CPU 301, various functions defined in the present invention are executed.

[0155] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present invention. Each box in the flowchart or block diagram may represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for implementing the specified logical functions. It should also be noted that in some alternative implementations, the functions marked in the box may also occur in an order different from that marked in the accompanying drawings.

[0156] Specifically, the visual recognition system of this embodiment includes a processor and a memory. The memory stores a computer program. When the computer program is executed by the processor, the visual recognition method for good products provided in the above embodiment is implemented.

[0157] As another aspect, the present invention further provides a computer-readable storage medium, which may be included in the visual recognition system described in the above embodiment; or may exist independently without being assembled into the visual recognition system. The above storage medium carries one or more computer programs, and when the above one or more computer programs are executed by a processor of the visual recognition system, the visual recognition system implements the method for visually identifying a good product provided in the above embodiment.

[0158] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.

[0159] As used in the above embodiments, the term "when..." may be interpreted to mean "if..." or "after..." or "in response to determining..." or "in response to detecting...", depending on the context. Similarly, the phrases "upon determining..." or "if (the stated condition or event) is detected" may be interpreted to mean "if determining..." or "in response to determining..." or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)", depending on the context.

Claims

1. A method for visually identifying a good product, characterized in that: Applied to a visual recognition system, the method comprises: In response to the product traceability application, determine the test type of the product to be traced; the test type includes a process test request or a final inspection request; When the detection type is the process detection request, obtaining abnormal process information that triggers the detection; Calculating the process correlation between each completed process and the abnormal process, and sorting the completed processes based on the process correlation to generate a sequence to be detected; When the inspection type is the final inspection request, extracting defect feature data of the product to be traced; Based on the defect feature data, determine a set of related processes from a process flow database, and sort them to generate a sequence to be inspected; Start a plurality of parallel detection channels, and sequentially obtain the process images to be inspected of the top-ranked processes in the sequence to be inspected, and input the process images to be inspected into the parallel detection channels for inspection; When the parallel inspection channel completes the current process inspection and the result is passed, the sequence to be inspected is updated, and the first process image in the updated sequence to be inspected is obtained for inspection; When the parallel detection channel detects that a process is abnormal, the corresponding process is determined as an abnormal process, and the detection of all the parallel detection channels is stopped.

2. The method according to claim 1, characterized in that Before the step of determining the detection type of the product to be traced in response to the product traceability application, the method further includes: Obtain product monitoring data from multiple monitoring devices in the product production process; Performing feature recognition on the product monitoring data, determining marked products whose product feature values ​​deviate from a preset threshold range, and generating abnormal prompt information; After a preset waiting time, or when a product tracing instruction input by the user after viewing the abnormal prompt information is received, a product tracing application is generated.

3. The method according to claim 1, characterized in that When the parallel detection channel completes the current process detection and the result is passed, the step of updating the sequence to be detected and obtaining the first process image in the updated sequence to be detected for detection specifically includes: When the parallel detection channel completes the current process detection and the result is passed, generating detection pass data based on the current process; Based on the detection passing data, removing the associated processes in the sequence to be detected to obtain an optimized sequence to be detected; The optimized sequence to be inspected is reordered, and the first process image after the order is acquired to the parallel inspection channel for inspection.

4. The method according to claim 3, characterized in that: When the parallel detection channel completes the current process detection and the result is passed, after the step of generating detection pass data based on the current process, the method further includes: When other parallel detection channels complete the corresponding current process detection and the result is passed, the detection pass data is updated based on the corresponding current process detection.

5. The method according to claim 3, characterized in that: Before the step of removing the associated processes in the sequence to be detected based on the detection pass data to obtain the optimized sequence to be detected, the method further includes: Obtaining a confidence value of the detection pass data of the current process; When the confidence value is lower than a preset confidence threshold, the first process of the sequence to be detected is acquired to the parallel detection channel for detection, and the current process is added to the first position of the sequence to be detected.

6. The method according to claim 1, characterized in that The step of starting a plurality of parallel detection channels, sequentially acquiring images of the processes to be detected of the top-ranked processes from the sequence to be detected, and inputting the images of the processes to be detected into the parallel detection channels for detection specifically includes: Starting multiple parallel detection channels and obtaining hardware parameter information corresponding to each of the parallel detection channels; Performing computing power classification on each of the parallel detection channels according to the hardware parameter information to obtain a computing power classification result; Sequentially acquiring the process images to be inspected of the top-ranked processes in the sequence to be inspected, and calculating the estimated inspection time of the process images to be inspected; Based on the computing power classification result and the estimated detection time, channels are allocated to the process images to be inspected.

7. The method according to claim 1, characterized in that After the step of determining the corresponding process as an abnormal process and stopping the detection of all the parallel detection channels when the parallel detection channels detect that the process is abnormal, the method further includes: Acquire the detection image and detection parameters of the abnormal process and generate an abnormal analysis report; Generate abnormality detection results and parameter adjustment suggestions based on the abnormality analysis report, and receive process adjustment instructions input by the user; The process parameters of the production equipment of the abnormal process are adjusted according to the process adjustment instruction.

8. A visual recognition system, characterized in that: The visual recognition system includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code includes computer instructions, and the one or more processors call the computer instructions to enable the visual recognition system to execute the method described in any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that: When the instructions are executed on a visual recognition system, the visual recognition system is caused to execute the method according to any one of claims 1 to 7.

10. A computer program product, characterized in that When the computer program product is run on a visual recognition system, the visual recognition system is caused to execute the method according to any one of claims 1 to 7.

Images