A real-time visual pattern recognition method and system for industrial inspection

Abstract

This invention relates to the field of data processing technology, specifically to a real-time visual pattern recognition method and system for industrial inspection. The method includes acquiring visual inspection data collected from products on a production line over a period of time, equipment process parameters of welding equipment, and inspection resource allocation information during product acquisition; determining a risk assessment coefficient characterizing the current operating status of the production line based on the visual inspection data and equipment process parameters; adjusting the inspection resource allocation information using the risk assessment coefficient to obtain adjusted inspection resource allocation information; and using the adjusted inspection resource allocation information to perform visual quality inspection on the products on the production line to obtain product quality identification results. The purpose of this invention is to solve the problem of low reliability in product quality identification in existing technologies, which makes it difficult to effectively identify product quality.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and specifically to a real-time visual pattern recognition method and system for industrial inspection. Background Technology

[0002] Industrial welding processes often generate intense arc light, metal spatter, and welding fumes, which visually manifest as areas of instantaneous high brightness, random bright spots, and diffuse occlusion. This leads to frequent misjudgments or omissions in existing visual pattern recognition methods. Furthermore, long-term operation of production lines causes wear and deformation of the equipment's mechanical structure, and subtle differences in raw materials from different batches can alter the reflection characteristics of the arc light, the visual morphology of the molten pool, and the distribution patterns of spatter during welding. Because existing visual pattern recognition methods and their associated image processing parameters are optimized for initial calibration states and specific material properties, they lack the ability to adapt to gradual, nonlinear environmental or process drifts. This necessitates periodic manual calibration and parameter adjustments by technicians, which is time-consuming, labor-intensive, and difficult to perform in real-time.

[0003] Meanwhile, production lines need to flexibly switch between producing products of various specifications and materials according to market demand, resulting in significant differences in welding process parameters and consequently, substantial changes in visual characteristics. While existing identification methods can preset multiple sets of methods and parameters for different product types, the switching mechanism has limitations. It requires detailed data collection and method training for all possible product types beforehand, and non-standardized welding phenomena may occur during the transition phase of the switching process. Therefore, existing methods have low reliability and struggle to effectively identify product quality. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time visual pattern recognition method and system for industrial inspection, which solves the problem that the existing technology has low reliability in identifying product quality, making it difficult to effectively identify product quality.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a real-time visual pattern recognition method for industrial inspection, comprising:

[0006] The system acquires visual inspection data collected from products on the production line over a period of time, along with equipment process parameters of the welding equipment and inspection resource allocation information during product acquisition. The visual inspection data includes product image information, product batch information, production environment information, and equipment status information of the visual acquisition equipment. The inspection resource allocation information includes the allocation of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy.

[0007] Based on the visual inspection data and equipment process parameters, a risk assessment coefficient is determined to characterize the current operating status of the production line.

[0008] The risk assessment coefficient is used to adjust the detection resource allocation information to obtain the adjusted detection resource allocation information.

[0009] Using the adjusted detection resource allocation information, visual quality inspection is performed on the products on the production line to obtain the product quality identification results.

[0010] Furthermore, the step of determining the risk assessment coefficient used to characterize the current production line operating status based on the visual inspection data and equipment process parameters includes:

[0011] The visual inspection data and equipment process parameters are respectively denoised to obtain denoised visual inspection data and equipment process parameters.

[0012] Based on the denoised visual inspection data and equipment process parameters, determine the data weight of the visual inspection data and the parameter weight of the equipment process parameters;

[0013] Based on the denoised visual inspection data, data weights, equipment process parameters, and parameter weights, a risk assessment coefficient is determined to characterize the current operating status of the production line.

[0014] Further, the steps of denoising the visual inspection data and equipment process parameters to obtain denoised visual inspection data and equipment process parameters include:

[0015] Based on the visual detection data, noise offset signal and brightness interference signal are determined;

[0016] Based on the noise offset signal and the brightness interference signal, determine the offset time point of the noise offset signal and the interference time point of the brightness interference signal;

[0017] The offset time point and the interference time point are compared to obtain the comparison result of the time points;

[0018] Using the comparison results at the aforementioned time points, the visual inspection data and equipment process parameters are denoised to obtain denoised visual inspection data and equipment process parameters.

[0019] Further, the step of determining the noise offset signal and the brightness interference signal based on the visual detection data includes:

[0020] Based on the visual detection data, environmental audio information and molten pool image information are determined;

[0021] The ambient audio information is subjected to spectral analysis to obtain the original noise-shifted signal;

[0022] The brightness change rate of the molten pool image information is analyzed to obtain the original brightness interference signal;

[0023] The original noise offset signal and the original brightness interference signal are respectively subjected to signal correction processing to obtain the noise offset signal and the brightness interference signal.

[0024] Further, the step of performing brightness change rate analysis on the molten pool image information to obtain the original brightness interference signal includes:

[0025] Based on the molten pool image information, determine consecutive image frames of the molten pool edge region;

[0026] The brightness change rate is analyzed on the continuous image frames to obtain the original brightness interference signal.

[0027] Furthermore, the step of adjusting the detection resource allocation information using the risk assessment coefficient to obtain the adjusted detection resource allocation information includes:

[0028] The risk level is determined based on the aforementioned risk assessment coefficient;

[0029] Using the risk level, an adjustment scheme is matched from the preset resource allocation model to obtain a resource allocation adjustment scheme;

[0030] The resource allocation adjustment scheme is used to adjust the detection resource allocation information to obtain the adjusted detection resource allocation information.

[0031] Furthermore, the step of matching adjustment schemes from a preset resource allocation model using the aforementioned risk level to obtain a resource allocation adjustment scheme includes:

[0032] Using the risk level, adjustment schemes are matched from the preset resource allocation model to obtain all candidate adjustment schemes and the matching value of each candidate adjustment scheme;

[0033] By using the matching value of each candidate adjustment scheme, all candidate adjustment schemes are screened to obtain resource allocation adjustment schemes.

[0034] Furthermore, the step of using the adjusted detection resource allocation information to perform visual quality inspection on the products on the production line and obtaining the product quality identification results includes:

[0035] Using the adjusted detection resource allocation information, initial visual quality inspection is performed on the products on the production line to obtain the consumption coefficient of detection resources;

[0036] Using the consumption coefficient of the detection resources, the adjusted detection resource allocation information is corrected to obtain the corrected detection resource allocation information;

[0037] Using the corrected detection resource allocation information, visual quality inspection is performed on the products on the production line to obtain the product quality identification results.

[0038] Further, the step of correcting the adjusted detection resource allocation information using the consumption coefficient of the detection resources to obtain the corrected detection resource allocation information includes:

[0039] The consumption coefficient of the detected resource is compared with a preset consumption threshold to obtain a consumption comparison value;

[0040] Using the consumption comparison value, the adjusted detection resource allocation information is corrected to obtain the corrected detection resource allocation information.

[0041] The present invention also provides a real-time visual pattern recognition system for industrial inspection, the system comprising:

[0042] The data acquisition module is used to acquire visual inspection data collected from products on the production line over a period of time, equipment process parameters of welding equipment, and inspection resource allocation information during product acquisition. The visual inspection data includes product image information, product batch information, production environment information, and equipment status information of the visual acquisition equipment. The inspection resource allocation information includes the allocation information of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy.

[0043] The coefficient determination module is used to determine the risk assessment coefficient that characterizes the current operating status of the production line based on the visual inspection data and equipment process parameters.

[0044] The information adjustment module is used to adjust the detection resource allocation information using the risk assessment coefficient to obtain the adjusted detection resource allocation information.

[0045] The quality inspection module is used to perform visual quality inspection on products on the production line using the adjusted inspection resource allocation information, and obtain the product quality identification results.

[0046] Compared with existing technologies, the real-time visual pattern recognition method and system for industrial inspection of the present invention have the following advantages:

[0047] This invention acquires visual inspection data, equipment process parameters, and inspection resource allocation information, determines a risk assessment coefficient based on this information, adjusts the inspection resource allocation using this coefficient, and finally uses the adjusted resource allocation for quality inspection. Simultaneously, by dynamically evaluating the production line's operating status and adjusting inspection resources in real time, it significantly improves the real-time performance, accuracy, and robustness of visual quality inspection, achieving more efficient and intelligent monitoring and quality control of industrial production lines, thereby effectively identifying product quality. Attached Figure Description

[0048] To more clearly illustrate the specific embodiments of the present invention, the accompanying drawings used in the specific embodiments will be briefly described below. In all the drawings, the elements or parts are not necessarily drawn to scale.

[0049] Figure 1 This is a flowchart of a real-time visual pattern recognition method for industrial inspection according to the present invention.

[0050] Figure 2 This is a structural block diagram of a real-time visual pattern recognition system for industrial inspection according to the present invention.

[0051] In the diagram: 210, Data Acquisition Module; 220, Coefficient Determination Module; 230, Information Adjustment Module; 240, Quality Inspection Module.

[0052] The implementation and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0053] The following drawings disclose several embodiments of the present invention. For clarity, many practical details will be described in the following description. However, it should be understood that these practical details are not intended to limit the invention. That is, in some embodiments of the invention, these practical details are not essential. Furthermore, for the sake of simplicity, some conventional structures and components will be shown in the drawings in a simple schematic manner.

[0054] It should be noted that all directional indications (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indication will also change accordingly.

[0055] Furthermore, in this invention, the use of terms such as "first" and "second" is for descriptive purposes only and does not specifically refer to any order or sequence, nor is it intended to limit the invention. They are merely used to distinguish components or operations described using the same technical terms, and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but only if they are feasible for those skilled in the art. If a combination of technical solutions is contradictory or impossible to implement, such a combination should be considered nonexistent and not within the scope of protection claimed by this invention.

[0056] To further understand the content, features, and effects of this invention, the following embodiments are provided, and detailed descriptions are given below in conjunction with the accompanying drawings:

[0057] Please see Figure 1 This invention provides a real-time visual pattern recognition method for industrial inspection, comprising the following steps:

[0058] S100. Acquire visual inspection data collected from products on the production line over a period of time, along with the equipment process parameters of the welding equipment and the resource allocation information for product inspection during data collection. The visual inspection data includes product image information, product batch information, production environment information, and the equipment status information of the visual acquisition equipment. The resource allocation information includes the allocation of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy. Specifically, visual inspection data refers to all relevant data obtained from products on the production line through visual acquisition equipment over a period of time, encompassing product image information, product batch information, production environment information, and the equipment status information of the visual acquisition equipment. This data forms the basis for evaluating the production line's operating status and product quality. The equipment process parameters of the welding equipment refer to various parameters used to control the operating status of the welding equipment during the welding process, such as welding current, welding voltage, welding speed, and shielding gas flow rate. These parameters directly affect welding quality and visual characteristics. The resource allocation information refers to the computing resources, storage resources, processing frequency resources, and accuracy resources allocated by the visual inspection system for image computing, image storage, image processing, and image processing accuracy. Reasonable resource allocation is crucial for ensuring efficient system operation. Specifically, visual inspection data can be acquired in various ways. For example, high-speed industrial cameras can be used to continuously photograph products on the production line, and the captured image information can be stored. Simultaneously, sensors can be used to acquire information such as temperature, humidity, and lighting conditions in the production environment, and the operating status information of the equipment, such as working mode and fault codes, can be obtained by reading the internal logs or status interfaces of the visual acquisition device. Product batch information can be obtained by scanning barcodes or QR codes on the products. The acquisition of welding equipment process parameters can be achieved through data interface communication with the welding equipment controller. For example, parameters such as current, voltage, and wire feed speed in the welding equipment's PLC (Programmable Logic Controller) can be read in real time via the OPCUA or Modbus protocol. The acquisition of inspection resource allocation information can be achieved by querying the resource scheduling module or configuration file within the visual inspection system. For example, the initial image processing capability, image storage space, image processing frequency, and image processing accuracy can be manually configured, and this configuration information is read and used as the initial inspection resource allocation information.

[0059] S200. Based on the visual inspection data and equipment process parameters, determine a risk assessment coefficient to characterize the current operating status of the production line. The risk assessment coefficient is a quantitative index determined comprehensively based on the visual inspection data and equipment process parameters, used to characterize the degree of risk in the current production line's operating status. This coefficient can reflect potential problems in the production line regarding quality, efficiency, or safety. Specifically, the risk assessment coefficient can be determined using various algorithms. For example, a rule-based expert system can be used, with a set of preset rules. When the visual inspection data or equipment process parameters meet specific conditions, a corresponding risk level is triggered, and the risk assessment coefficient is calculated. For example, when continuous abnormal defects (such as porosity and cracks) appear in the product image or the welding current fluctuation exceeds a preset threshold, a higher risk assessment coefficient is calculated according to preset rules.

[0060] S300. Using the risk assessment coefficient, the detection resource allocation information is adjusted to obtain adjusted detection resource allocation information. The adjusted detection resource allocation information refers to the result of optimizing the original detection resource allocation information under the guidance of the risk assessment coefficient, so that the detection resources can more effectively cope with the current operating status of the production line. Specifically, the adjustment of the detection resource allocation information can be achieved through various strategies. For example, a threshold-based adjustment strategy can be used. When the risk assessment coefficient exceeds a preset threshold, the image computing power and image processing frequency are increased according to a preset adjustment scheme to improve the real-time performance and accuracy of the detection.

[0061] S400. Using the adjusted detection resource allocation information, perform visual quality inspection on the products on the production line to obtain the product quality identification result. The product quality identification result refers to the conclusions drawn after analyzing the product using visual quality inspection methods regarding whether the product meets quality standards and what defects exist. Specifically, visual quality inspection can be achieved through various image processing and pattern recognition technologies. For example, traditional machine vision methods based on feature extraction and classification, such as edge detection, shape matching, and texture analysis, can be used, combined with classifiers such as support vector machines or decision trees, to identify and classify defects in product images.

[0062] This invention effectively avoids performance degradation caused by a lack of adaptability by introducing a risk assessment coefficient and dynamically adjusting the allocation of inspection resources. Specifically, it dynamically assesses potential risks by comprehensively analyzing visual inspection data and equipment process parameters based on the real-time operating status of the production line. The introduction of the risk assessment coefficient enables timely and quantitative responses to abnormal situations on the production line (such as environmental interference, equipment drift, and process changes), thereby automatically adjusting the allocation of inspection resources in real time. For example, when the risk assessment coefficient is high, image computing power and image processing frequency can be intelligently increased to handle more complex image processing tasks or improve the real-time performance of defect detection; when the risk assessment coefficient is low, resource consumption can be appropriately reduced to optimize system operating efficiency. Through dynamic adjustment, computing resources can be utilized more efficiently while ensuring detection accuracy, avoiding performance bottlenecks or resource waste caused by improper resource allocation. Furthermore, by applying the adjusted inspection resource allocation information to the final visual quality inspection stage, it is ensured that the visual pattern recognition system can operate with optimal configuration under different production conditions, thereby improving the accuracy and reliability of product quality identification. This significantly improves the robustness and intelligence of industrial vision inspection systems, enabling them to better adapt to the increasing complexity and uncertainty in modern industrial production.

[0063] In some embodiments of this application described above, the step of determining a risk assessment coefficient to characterize the current production line operating status based on the visual inspection data and equipment process parameters includes:

[0064] The visual inspection data and equipment process parameters are denoised to obtain denoised visual inspection data and equipment process parameters. This step involves applying various signal processing or data cleaning techniques to eliminate or reduce random noise, outliers, or irrelevant interference in the data, thereby improving the purity and reliability of the data. The purpose is to ensure that the subsequent calculation of risk assessment coefficients is based on high-quality data, avoiding negative impacts from noise on the assessment results. For example, low-pass filtering, median filtering, wavelet denoising, and Kalman filtering can be used to process time series data, or outlier detection algorithms can be used to identify and remove outlier data points.

[0065] Based on the denoised visual inspection data and equipment process parameters, the data weights of the visual inspection data and the parameter weights of the equipment process parameters are determined. This step assigns corresponding weights to different types of data based on their characteristics, importance, and contribution to the risk assessment of the production line's operating status. The aim is to more accurately reflect the indicative role of different data sources in risk assessment when calculating risk assessment coefficients, ensuring that more critical and reliable data carries greater weight in the assessment. Weights can be determined through expert experience, historical data analysis, machine learning model training (e.g., determining feature importance through regression analysis or decision tree models), or adaptive adjustment algorithms. For example, higher weights can be assigned to critical visual defect features or equipment overload parameters.

[0066] Based on the denoised visual inspection data, data weights, equipment process parameters, and parameter weights, a risk assessment coefficient is determined to characterize the current operating status of the production line. This step involves combining the denoised and weighted data, and calculating a numerical value that quantifies the current operational risk of the production line using a pre-defined mathematical model or algorithm. For example, methods such as weighted averaging, fuzzy logic reasoning, neural network models, or support vector machines can be used to transform the processed input into a single risk assessment coefficient. The magnitude of this coefficient directly reflects the current risk level faced by the production line, providing a quantitative basis for subsequent adjustments to inspection resources.

[0067] Specifically, on industrial production lines, visual inspection data may be affected by fluctuations in workshop lighting or noise from the camera sensor itself, while the equipment process parameters (such as current and voltage) of welding equipment may experience instantaneous fluctuations due to power instability or measurement errors. First, to denoise this data, a medium-range filter can be applied to the image brightness information in the visual inspection data to eliminate salt-and-pepper noise; simultaneously, a Kalman filter can be applied to the time-series data of the equipment process parameters to smooth measurement noise and predict true values. After processing, cleaner and more stable denoised visual inspection data and equipment process parameters are obtained. Second, when determining data and parameter weights, considering that product image information is crucial for defect identification, a higher weight (e.g., 0.7) can be assigned to product image information in the visual inspection data, while a relatively lower weight (e.g., 0.3) can be assigned to production environment information. Similarly, for the current and voltage parameters of the welding equipment, since they directly affect welding quality, a higher parameter weight (e.g., 0.8) can be assigned, while a lower weight (e.g., 0.2) can be assigned to auxiliary parameters such as equipment temperature. Finally, the denoised data and determined weights are input into a pre-defined risk assessment model, such as a weighted summation model or a trained neural network model, to output the final risk assessment coefficient. For example, if the model outputs a high risk assessment coefficient, it indicates that the current production line has a high operational risk, requiring timely adjustments to the allocation of detection resources to strengthen monitoring. This ensures the accuracy and reliability of risk assessment, providing strong support for the stable operation of the production line.

[0068] This invention effectively filters out potential interference information in visual inspection data and equipment process parameters by introducing noise reduction processing, thereby ensuring the accuracy and stability of the data used for risk assessment. Furthermore, by assigning data weights and parameter weights to different types of data, the calculation of risk assessment coefficients fully considers the contribution and reliability differences of different data sources to the risk of production line operation. Because of the preprocessing and weighting of the data, the final risk assessment coefficients more realistically and accurately represent the risk level of the current production line operation, avoiding assessment bias caused by problems with the quality of the original data.

[0069] In some embodiments of this application described above, the steps of denoising the visual inspection data and the equipment process parameters to obtain denoised visual inspection data and equipment process parameters include:

[0070] Based on the visual inspection data, noise offset signals and brightness interference signals are identified. Specifically, identifying noise offset signals and brightness interference signals is used to identify two main types of noise affecting the quality of visual inspection data. Noise offset signals typically refer to baseline drift or random fluctuations in image or sensor data caused by environmental factors (such as mechanical vibration and electromagnetic interference); brightness interference signals mainly refer to image brightness anomalies caused by changes in lighting conditions (such as ambient light flicker and changes in welding arc light). By distinguishing between these two signals, a more targeted basis can be provided for subsequent denoising processing.

[0071] Based on the noise offset signal and the brightness interference signal, the offset time point of the noise offset signal and the interference time point of the brightness interference signal are determined. Determining these time points is crucial for accurately locating the time window in which the noise occurs. The offset time point indicates the start and end times of the noise offset signal, while the interference time point indicates the start and end times of the brightness interference signal. Precise time point location helps to process only the data segments affected by noise during denoising, avoiding unnecessary modifications to normal data.

[0072] The offset time point and the interference time point are compared to obtain the time point comparison results. Specifically, the purpose is to analyze the temporal overlap or sequential relationship of different noise sources. For example, if two noises occur within the same time period, a combined denoising strategy is required; if they occur at different time periods, they can be processed separately. The comparison results provide important temporal information for the formulation of denoising strategies.

[0073] Using the comparison results at the aforementioned time points, the visual inspection data and equipment process parameters are denoised to obtain denoised visual inspection data and equipment process parameters. Specifically, the denoising process is based on the accurate identification and analysis of noise types and their occurrence times. For example, for data affected by noise offset signals within a specific offset time point, filtering and baseline correction methods can be used; for data affected by brightness interference signals within a specific interference time point, brightness normalization and image enhancement methods can be used. Denoising based on time point comparison ensures the accuracy and effectiveness of denoising.

[0074] Specifically, on a welding production line, product image information and production environment information are continuously collected, while the equipment process parameters of the welding equipment are recorded. During a certain time period, a large piece of equipment near the production line starts up, generating mechanical vibrations that cause periodic image shifts in the visual inspection data, identified as noise shift signals. Simultaneously, fluctuations in arc brightness during welding cause irregular changes in the brightness of the molten pool image area, identified as brightness interference signals. Specifically, based on the collected visual inspection data, the noise shift signal is determined by analyzing motion vectors in the image sequence or baseline drift of sensor readings. Simultaneously, brightness interference signals are determined by analyzing the image's brightness histogram or the pixel intensity change rate in specific areas. For example, the noise shift signal may persist continuously during the time period T1 to T2, while the brightness interference signal may frequently appear during the time period T3 to T4. Comparing the shift time points and interference time points reveals partial overlap between T1 to T2 and T3 to T4. Based on this comparison, the system performs targeted denoising processing on the visual inspection data and equipment process parameters. For example, during the time period T1 to T2, image stabilization algorithms or Kalman filtering are applied to the visual inspection data to correct image offset; during the time period T3 to T4, adaptive brightness normalization is performed on the molten pool image information. For overlapping time periods, a combined denoising strategy can be used, such as performing brightness normalization first and then image stabilization. Through refined denoising processing, high-quality denoised visual inspection data and equipment process parameters are finally obtained, ensuring the accuracy of subsequent risk assessment.

[0075] This invention first identifies noise offset signals and brightness interference signals based on visual inspection data, thereby decomposing complex noise problems into identifiable specific noise types. Subsequently, by determining the offset and interference time points of the noise signals, it achieves precise capture of the noise occurrence sequence. Given that different noise sources may affect data at different times or in different ways, comparing the offset and interference time points reveals the combined or independent effects of noise, providing crucial information for developing refined denoising strategies. Due to the in-depth analysis of noise types and timing, subsequent denoising processing can be specifically applied to affected data segments, effectively addressing the potential for blind and inefficient denoising processes.

[0076] In some embodiments of this application described above, the step of determining the noise offset signal and the brightness interference signal based on the visual detection data includes:

[0077] Based on the visual inspection data, environmental audio information and molten pool image information are determined. The environmental audio information refers to audio data captured by acoustic sensors or microphones built into the visual acquisition device in the production environment during visual acquisition of the product. The audio data may include noise caused by equipment operation and environmental vibrations. The molten pool image information refers to image data containing the molten metal area captured by the visual acquisition device during welding or similar high-temperature processing. Changes in the brightness of the molten pool are often closely related to factors such as the stability of the welding process and the occurrence of defects.

[0078] The environmental audio information is subjected to spectral analysis to obtain the original noise-shifted signal. This step converts the time-domain audio signal into a frequency-domain representation, thereby identifying noise components within a specific frequency range. For example, methods such as Fourier transform can be used to analyze the frequency distribution and energy intensity of the audio signal to identify the original noise-shifted signal, whether periodic or non-periodic.

[0079] The brightness change rate is analyzed on the molten pool image information to obtain the original brightness interference signal. This step is used to detect the drastic change in brightness over time in the image. For example, the rate of brightness change can be quantified by calculating the average brightness or local brightness difference of the molten pool region between two consecutive image frames. A brightness change rate that is too high or too low may indicate an anomaly in the production process, thus obtaining the original brightness interference signal.

[0080] The original noise offset signal and the original brightness interference signal are respectively subjected to signal correction processing to obtain the noise offset signal and the brightness interference signal. This step removes possible measurement errors, environmental interference, or irrelevant components from the original signals, making them more accurately reflect the actual noise offset and brightness interference. For example, techniques such as filtering, baseline correction, and thresholding can be used to refine the original signals to improve the accuracy of subsequent analysis, ultimately obtaining the noise offset signal and brightness interference signal for denoising.

[0081] This invention effectively identifies original noise offset signals and original brightness interference signals from different sources by separating environmental audio information and melt pool image information from visual inspection data and performing targeted spectral analysis and brightness change rate analysis on them respectively. This allows for more accurate capture of potential interference factors in the production environment that may affect visual inspection results. By performing signal correction processing on the original signals, the purity and accuracy of the identified noise offset signals and brightness interference signals are further improved, providing a reliable basis for subsequently determining the offset time point of the noise offset signal and the interference time point of the brightness interference signal, thereby ensuring the effectiveness of the denoising process.

[0082] In some embodiments of this application described above, the step of performing brightness change rate analysis on the molten pool image information to obtain the original brightness interference signal includes:

[0083] Based on the molten pool image information, continuous image frames of the molten pool edge region are determined. The molten pool image information typically refers to images or video sequences of the molten metal pool acquired by visual acquisition equipment during welding or similar industrial production processes. The molten pool edge region refers to the boundary area between the molten metal and the surrounding solid materials or gaseous medium. Variations in the shape, size, and brightness of this region are often closely related to welding quality, process stability, and external environmental interference. Determining continuous image frames of the molten pool edge region involves identifying and extracting the boundary portion of the molten pool from the acquired molten pool image information and organizing it continuously across multiple frames in a time series.

[0084] The brightness change rate is analyzed on the consecutive image frames to obtain the original brightness interference signal. This step involves calculating the degree of change in pixel brightness over time in the edge region of the melt pool in these consecutive image frames. For example, image processing algorithms, such as edge detection and region segmentation, can be used to accurately locate the melt pool edge. Subsequently, the average brightness, maximum brightness, or brightness of a specific pixel in the edge region is compared between different time frames to calculate the brightness change rate. The brightness change rate can be expressed as an increase or decrease in brightness per unit time, or a more complex statistical indicator.

[0085] This invention analyzes the rate of brightness change by focusing on continuous image frames at the edge of the molten pool. This allows for more accurate detection of brightness interference caused by factors such as external light source fluctuations, welding arc instability, or sensor noise. The edge of the molten pool is more sensitive to interference, and its brightness changes often directly reflect the presence and intensity of the interference. By analyzing the rate of brightness change across consecutive frames, normal process fluctuations can be effectively distinguished from abnormal brightness interference, thus generating a more accurate original signal of the brightness interference. This avoids potential misjudgments that might result from a coarse analysis of the entire molten pool image, improving the accuracy of interference signal identification.

[0086] In some embodiments of this application described above, the step of adjusting the detection resource allocation information using the risk assessment coefficient to obtain adjusted detection resource allocation information includes:

[0087] Based on the risk assessment coefficient, the risk level is determined. This step involves comparing the calculated risk assessment coefficient with a preset risk threshold to classify the current production line's operating status into different risk levels. For example, the risk level can be divided into multiple tiers such as low risk, medium risk, and high risk, each corresponding to a different range of risk assessment coefficients. The risk assessment coefficient is a comprehensive quantitative indicator of the current production line's operating status, and its value directly reflects the likelihood of potential production anomalies or quality problems.

[0088] Using the risk level, adjustment schemes are matched from a preset resource allocation model to obtain a resource allocation adjustment scheme. This step refers to determining the risk level of the current production line, and then searching for and selecting the most suitable adjustment strategy from a pre-established resource allocation model based on that risk level. This preset resource allocation model can be a database or rule set containing multiple resource allocation strategies, each strategy being associated with a specific risk level, aiming to optimize the allocation of detection resources such as image computing power, image storage space, image processing frequency, and image processing accuracy.

[0089] The resource allocation adjustment scheme is used to adjust the detection resource allocation information, resulting in adjusted detection resource allocation information. This step refers to modifying and optimizing the current detection resource allocation information according to the matched specific adjustment scheme. For example, when the risk level is high, the adjustment scheme may indicate increasing the image processing frequency and accuracy to ensure more stringent quality detection; when the risk level is low, the allocation of certain resources may be appropriately reduced to improve resource utilization efficiency. The aim is to make the allocation of detection resources more reasonable and efficient to adapt to production needs under different risk levels.

[0090] Specifically, on an industrial production line, the risk assessment coefficient calculated using visual inspection data and equipment process parameters is 0.75. Based on preset risk thresholds (e.g., 0-0.3 for low risk, 0.3-0.6 for medium risk, and 0.6-1.0 for high risk), a risk assessment coefficient of 0.75 is determined to be a high-risk level. Subsequently, this high-risk level is used to match a corresponding adjustment scheme from a preset resource allocation model. This model might preset a strategy for high-risk levels such as increasing the image processing frequency by 20%, improving image processing accuracy by 15%, and increasing image storage space by 10%. Based on this resource allocation adjustment scheme, the current inspection resource allocation information will be adjusted; for example, the original image processing frequency might be adjusted from 100 frames / second to 120 frames / second, the image processing accuracy from 90% to 103.5%, and the image storage space from 1TB to 1.1TB. Thus, the adjusted inspection resource allocation information will be used for subsequent visual quality inspection to meet the higher requirements for inspection capabilities under high-risk conditions, ensuring reliable identification of product quality.

[0091] This invention effectively addresses the limitations of directly adjusting a single risk assessment coefficient by introducing risk level determination and an adjustment scheme matching mechanism based on a preset resource allocation model. Specifically, firstly, the risk assessment coefficient is transformed into a more operational risk level, enabling a more detailed classification of the production line's operating status. Secondly, through a preset resource allocation model, different risk levels are associated with corresponding resource allocation adjustment schemes, thereby achieving intelligent and automated resource allocation strategies. Due to the hierarchical matching and strategic adjustment methods, the allocation of inspection resources can be dynamically and accurately optimized according to the actual risk situation, avoiding blind adjustments or excessive resource consumption, and ensuring the effectiveness and efficiency of quality inspection under different production conditions.

[0092] In some embodiments of this application described above, the step of matching adjustment schemes from a preset resource allocation model using the risk level to obtain a resource allocation adjustment scheme includes:

[0093] Using the aforementioned risk level, adjustment schemes are matched from a preset resource allocation model to obtain all candidate adjustment schemes and a matching value for each candidate adjustment scheme. Specifically, all candidate adjustment schemes refer to the set of all potential resource allocation adjustment schemes identified within a certain matching range that meet certain matching conditions when matching from the preset resource allocation model according to the risk level. Candidate schemes may differ in allocation strategies regarding image computing power, image storage space, image processing frequency, and image processing accuracy. The matching value for each candidate adjustment scheme is a quantitative indicator measuring the degree of suitability or superiority of each candidate adjustment scheme with the current risk level. This matching value can be calculated based on various factors, such as the degree of risk mitigation, resource utilization efficiency, impact on production efficiency, and implementation costs. For example, the matching value can be a comprehensive score or a measure of the distance between the scheme and the risk level.

[0094] By utilizing the matching value of each candidate adjustment scheme, all candidate adjustment schemes are screened to obtain resource allocation adjustment schemes. Scheme screening refers to the process of selecting the optimal or most suitable resource allocation adjustment scheme for the current production line operating state from all candidate adjustment schemes according to preset screening rules or optimization objectives. Screening rules may include, but are not limited to: selecting the scheme with the highest matching value, selecting the scheme that meets a specific performance threshold, or making a decision based on other auxiliary indicators (such as historical performance data and expert experience) among multiple schemes with similar matching values.

[0095] Specifically, after determining the current production line's risk level to be moderately high based on risk assessment coefficients, a query and matching process is performed within a pre-defined resource allocation model based on this risk level. At this point, the model may identify multiple candidate adjustment schemes related to the moderately high risk level. For example, candidate adjustment scheme A might suggest increasing image computing power by 20%, image processing frequency by 15%, and image storage space by 10GB, while maintaining image processing accuracy unchanged; candidate adjustment scheme B might suggest increasing image computing power by 15%, image processing frequency by 20%, and image storage space by 5GB, while improving image processing accuracy by 5%; candidate adjustment scheme C might suggest increasing image computing power by 25%, image processing frequency by 10%, and image storage space by 15GB, while maintaining image processing accuracy unchanged. For these candidate schemes, their respective matching values ​​are calculated based on pre-defined evaluation criteria (e.g., comprehensively considering resource consumption, impact on production efficiency, and historical risk mitigation effects). Assuming the calculation results are: a matching value of 0.85 for scheme A, 0.92 for scheme B, and 0.80 for scheme C. Subsequently, these matching values ​​are used for scheme selection. If the filtering rule selects the option with the highest matching value, then option B will be selected as the final resource allocation adjustment plan. If the filtering rule selects the option with a matching value higher than 0.90 and the lowest resource consumption, then option B will also be selected. In this way, even if multiple potential adjustment plans exist, quantitative evaluation and intelligent filtering can be used to select the resource allocation adjustment plan that best suits the current production line operating status and optimization goals, thereby achieving more precise and effective resource management.

[0096] This invention avoids the limitations of direct matching by first acquiring all candidate adjustment schemes and their matching values, and then screening these candidate schemes. Specifically, when matching from a preset resource allocation model using risk levels, it no longer simply seeks a single matching result, but comprehensively identifies all potentially feasible resource allocation adjustment schemes and assigns a quantified matching value to each scheme. The introduction of the matching value concept clearly characterizes the merits of each candidate scheme. Based on this, by screening these candidate schemes, the adjustment scheme that best meets the actual needs of the current production line, most effectively addresses risks, and has the highest resource utilization efficiency can be selected from multiple possibilities according to preset optimization goals or strategies. This avoids suboptimal selections that may result from simple matching, ensuring that the finally selected resource allocation adjustment scheme has higher adaptability and robustness.

[0097] In some embodiments of this application described above, the step of using the adjusted detection resource allocation information to perform visual quality inspection on products on the production line and obtain product quality identification results includes:

[0098] Using the adjusted inspection resource allocation information, initial visual quality inspection is performed on products on the production line to obtain the consumption coefficient of inspection resources. Specifically, initial quality inspection refers to a pre-inspection or trial inspection of products using the preliminarily adjusted inspection resource allocation information before formal comprehensive quality inspection. Its purpose is to monitor and obtain the actual resource consumption in the current inspection process in real time, such as the actual usage of image computing power, image storage space, image processing frequency, and image processing accuracy. From this, the consumption coefficient of inspection resources can be obtained, which characterizes the actual consumption ratio or absolute consumption value of each inspection resource relative to its allocated amount in the initial inspection stage.

[0099] Using the consumption coefficient of the aforementioned detection resources, the adjusted detection resource allocation information is corrected to obtain corrected detection resource allocation information. This step refers to a secondary correction of the previously adjusted detection resource allocation information based on the risk assessment coefficient, according to the actual obtained consumption coefficient of the detection resources. For example, if the initial detection finds that the consumption of a certain resource is too high, the allocation of that resource can be appropriately increased; if the consumption is too low, the allocation can be appropriately reduced to avoid resource waste. In practical applications, the correction process can employ various algorithms, such as dynamic optimization based on proportional adjustment, threshold judgment, or machine learning models. The aim is to make the allocation of detection resources closer to actual needs, improving resource utilization and detection effectiveness.

[0100] Using the corrected detection resource allocation information, visual quality inspection is performed on products on the production line to obtain product quality identification results. The corrected detection resource allocation information is a more accurate and optimized resource allocation scheme after initial detection feedback and correction processing. Using this corrected detection resource allocation information, final, more accurate, and efficient visual quality inspection can be performed on products on the production line, thereby obtaining reliable product quality identification results.

[0101] Specifically, in product inspection, based on a risk assessment coefficient, image computing power is initially allocated to 80% and image storage space to 70%. During initial quality inspection, actual image computing power consumption reaches 90%, while image storage space consumption is only 50%. At this point, the consumption coefficient of inspection resources is calculated; for example, the consumption coefficient of image computing power is 1.125 (90% / 80%), and the consumption coefficient of image storage space is 0.714 (50% / 70%). Based on the consumption coefficient, the adjusted inspection resource allocation information is corrected. Specifically, the allocation of image computing power may be corrected to 90%, while the allocation of image storage space may be corrected to 50%. Finally, using the corrected inspection resource allocation information (e.g., 90% image computing power, 50% image storage space), visual quality inspection of products on the production line is performed, resulting in more accurate and efficient product quality identification results. This ensures the flexibility and adaptability of resource allocation, avoiding resource waste or performance bottlenecks that may result from fixed allocation.

[0102] This invention effectively avoids the problem of adjusted testing resource allocation information not perfectly matching actual testing needs by introducing feedback from initial quality inspection and resource consumption coefficients. Specifically, an initial quality inspection is first performed, which allows for real-time acquisition of the actual consumption of testing resources and generates a consumption coefficient. This consumption coefficient serves as a real-time feedback signal, used to correct the adjusted testing resource allocation information. Because of this dynamic correction based on actual consumption, the allocation of testing resources can more accurately adapt to the actual testing needs of the current production line, avoiding blind or lagging resource allocation. This ensures that subsequent formal quality inspections are conducted under optimized resource configuration, thereby improving the accuracy and efficiency of the inspection.

[0103] In some embodiments of this application described above, the step of correcting the adjusted detection resource allocation information using the consumption coefficient of the detection resources to obtain corrected detection resource allocation information includes:

[0104] The consumption coefficient of the detection resources is compared with a preset consumption threshold to obtain a consumption comparison value. Specifically, the consumption coefficient of the detection resources refers to the quantitative indicators of the actual image computing power, image storage space, image processing frequency, and image processing accuracy consumed during the initial visual quality inspection of products on the production line. This coefficient reflects the actual resource requirements of the current inspection task. The preset consumption threshold can be understood as a benchmark value pre-set by the system to measure the rationality of the detection resource consumption. This threshold can be set based on historical data, rules of thumb, or the performance requirements of a specific production line, aiming to provide a reference standard for resource consumption. Comparing the consumption coefficient of the detection resources with the preset consumption threshold aims to assess the rationality of the current resource consumption. For example, when the consumption coefficient is higher than the threshold, it may indicate insufficient resource allocation or that the complexity of the inspection task exceeds expectations; when the consumption coefficient is lower than the threshold, it may indicate redundant resource allocation. The resulting consumption comparison value provides a quantitative decision-making basis for subsequent resource correction processing.

[0105] Using the consumption comparison value, the adjusted detection resource allocation information is corrected to obtain the corrected detection resource allocation information.

[0106] Specifically, after initial visual quality inspection of products on the production line, the current image computing power consumption coefficient is 1.2, while the preset image computing power consumption threshold is 1.0. Comparing the consumption coefficient of 1.2 with the threshold of 1.0 yields a consumption comparison value indicating a 20% overconsumption of current image computing power. Based on this consumption comparison value, the adjusted detection resource allocation information will be corrected. For example, the resources allocated to image computing power may be appropriately reduced, or the priority of image processing tasks may be adjusted to optimize resource utilization. Conversely, if the consumption coefficient is 0.8, lower than the threshold of 1.0, the consumption comparison value indicates resource redundancy. Excess resources may be allocated to other more demanding detection stages, or appropriate resource reclamation may be implemented to ensure dynamic balance and efficient utilization of resource allocation.

[0107] This invention introduces a preset consumption threshold, providing a clear reference benchmark for the consumption coefficient of detected resources. By comparing the actual consumption coefficient with this threshold, the rationality or degree of deviation of the current resource consumption can be quantified, thereby generating a consumption comparison value with guiding significance. This consumption comparison value serves as the direct basis for correction processing, enabling the adjusted detected resource allocation information to be corrected more accurately and intelligently, avoiding blind adjustments and ensuring dynamic optimization and efficient utilization of resource allocation.

[0108] For a real-time visual pattern recognition method for industrial inspection based on any of the above embodiments, please refer to [link to relevant documentation]. Figure 2The present invention also provides a real-time visual pattern recognition system for industrial inspection, the system comprising a data acquisition module 210, a coefficient determination module 220, an information adjustment module 230, and a quality inspection module 240.

[0109] The data acquisition module 210 is used to acquire visual inspection data collected from products on the production line over a period of time, equipment process parameters of welding equipment, and inspection resource allocation information when products are collected. The visual inspection data includes product image information, product batch information, production environment information, and equipment status information of the visual acquisition equipment. The inspection resource allocation information includes the allocation information of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy.

[0110] The coefficient determination module 220 is used to determine the risk assessment coefficient that characterizes the current operating status of the production line based on the visual inspection data and equipment process parameters.

[0111] The information adjustment module 230 is used to adjust the detection resource allocation information using the risk assessment coefficient to obtain the adjusted detection resource allocation information.

[0112] The quality inspection module 240 is used to perform visual quality inspection on the products on the production line using the adjusted inspection resource allocation information, and obtain the product quality identification result.

[0113] In this embodiment, the data acquisition module 210 perceives the production line status in real time, the coefficient determination module 220 dynamically assesses potential risks, and the information adjustment module 230 intelligently adjusts the allocation of detection resources. During the detection process, the quality inspection module 240 dynamically adjusts parameters such as the parallelism of image processing, the batch size of model inference, and the caching strategy of image storage based on the adjusted detection resource allocation information. This ensures that quality inspection can be performed with optimal performance under the current risk level, and the product quality identification results can be output. Through an adaptive resource management strategy, the system can flexibly adjust image computing power, storage space, processing frequency, and accuracy according to the real-time operating status of the production line. This ensures efficient use of computing resources while maintaining detection accuracy, avoiding performance bottlenecks or resource waste caused by improper resource allocation. Consequently, it can automatically and in real-time respond to abnormal situations on the production line, significantly improving the robustness and intelligence level of the industrial vision inspection system.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions 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 invention, and they should all be covered within the scope of the present invention specification.

Claims

1. A real-time visual pattern recognition method for industrial inspection, characterized in that, include: The system acquires visual inspection data collected from products on the production line over a period of time, along with equipment process parameters of the welding equipment and inspection resource allocation information during product acquisition. The visual inspection data includes product image information, product batch information, production environment information, and equipment status information of the visual acquisition equipment. The inspection resource allocation information includes the allocation of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy. Based on the visual inspection data and equipment process parameters, a risk assessment coefficient is determined to characterize the current operating status of the production line. The risk assessment coefficient is used to adjust the detection resource allocation information to obtain the adjusted detection resource allocation information. Using the adjusted detection resource allocation information, visual quality inspection is performed on the products on the production line to obtain the product quality identification results.

2. A real-time visual pattern recognition method for industrial inspection according to claim 1, characterized in that, The steps for determining the risk assessment coefficient used to characterize the current production line operating status based on the visual inspection data and equipment process parameters include: The visual inspection data and equipment process parameters are respectively denoised to obtain denoised visual inspection data and equipment process parameters. Based on the denoised visual inspection data and equipment process parameters, determine the data weight of the visual inspection data and the parameter weight of the equipment process parameters; Based on the denoised visual inspection data, data weights, equipment process parameters, and parameter weights, a risk assessment coefficient is determined to characterize the current operating status of the production line.

3. A real-time visual pattern recognition method for industrial inspection according to claim 2, characterized in that, The steps of denoising the visual inspection data and equipment process parameters to obtain denoised visual inspection data and equipment process parameters include: Based on the visual detection data, noise offset signal and brightness interference signal are determined; Based on the noise offset signal and the brightness interference signal, determine the offset time point of the noise offset signal and the interference time point of the brightness interference signal; The offset time point and the interference time point are compared to obtain the comparison result of the time points; Using the comparison results at the aforementioned time points, the visual inspection data and equipment process parameters are denoised to obtain denoised visual inspection data and equipment process parameters.

4. The real-time visual pattern recognition method for industrial inspection according to claim 3, characterized in that, The steps for determining the noise offset signal and the brightness interference signal based on the visual detection data include: Based on the visual detection data, environmental audio information and molten pool image information are determined; The ambient audio information is subjected to spectral analysis to obtain the original noise-shifted signal; The brightness change rate of the molten pool image information is analyzed to obtain the original brightness interference signal; The original noise offset signal and the original brightness interference signal are respectively subjected to signal correction processing to obtain the noise offset signal and the brightness interference signal.

5. A real-time visual pattern recognition method for industrial inspection according to claim 4, characterized in that, The steps of performing brightness change rate analysis on the molten pool image information to obtain the original brightness interference signal include: Based on the molten pool image information, determine consecutive image frames of the molten pool edge region; The brightness change rate is analyzed on the continuous image frames to obtain the original brightness interference signal.

6. The real-time visual pattern recognition method for industrial inspection as claimed in claim 1 wherein, The steps for adjusting the detection resource allocation information using the risk assessment coefficient to obtain the adjusted detection resource allocation information include: The risk level is determined based on the aforementioned risk assessment coefficient; Using the risk level, an adjustment scheme is matched from the preset resource allocation model to obtain a resource allocation adjustment scheme; The resource allocation adjustment scheme is used to adjust the detection resource allocation information to obtain the adjusted detection resource allocation information.

7. A real-time visual pattern recognition method for industrial inspection according to claim 6, characterized in that, The steps for obtaining a resource allocation adjustment plan by matching adjustment schemes from a preset resource allocation model using the aforementioned risk level include: Using the risk level, adjustment schemes are matched from the preset resource allocation model to obtain all candidate adjustment schemes and the matching value of each candidate adjustment scheme; By using the matching value of each candidate adjustment scheme, all candidate adjustment schemes are screened to obtain resource allocation adjustment schemes.

8. The real-time visual pattern recognition method for industrial inspection according to claim 1, characterized in that, The steps for performing visual quality inspection on products on the production line using the adjusted detection resource allocation information to obtain product quality identification results include: Using the adjusted detection resource allocation information, initial visual quality inspection is performed on the products on the production line to obtain the consumption coefficient of detection resources; Using the consumption coefficient of the detection resources, the adjusted detection resource allocation information is corrected to obtain the corrected detection resource allocation information; Using the corrected detection resource allocation information, visual quality inspection is performed on the products on the production line to obtain the product quality identification results.

9. A real-time visual pattern recognition method for industrial inspection according to claim 8, characterized in that, The steps for correcting the adjusted detection resource allocation information using the consumption coefficient of the detection resources to obtain the corrected detection resource allocation information include: The consumption coefficient of the detected resource is compared with a preset consumption threshold to obtain a consumption comparison value; Using the consumption comparison value, the adjusted detection resource allocation information is corrected to obtain the corrected detection resource allocation information.

10. A real-time visual pattern recognition system for industrial inspection, characterized in that, The system includes: The data acquisition module is used to acquire visual inspection data collected from products on the production line over a period of time, equipment process parameters of welding equipment, and inspection resource allocation information during product acquisition. The visual inspection data includes product image information, product batch information, production environment information, and equipment status information of the visual acquisition equipment. The inspection resource allocation information includes the allocation information of inspection resources to image computing power, image storage space, image processing frequency, and image processing accuracy. The coefficient determination module is used to determine the risk assessment coefficient that characterizes the current operating status of the production line based on the visual inspection data and equipment process parameters. The information adjustment module is used to adjust the detection resource allocation information using the risk assessment coefficient to obtain the adjusted detection resource allocation information. The quality inspection module is used to perform visual quality inspection on products on the production line using the adjusted inspection resource allocation information, and obtain the product quality identification results.

Images