A woven bag defect detection method and system based on image pattern recognition

Abstract

The application relates to the technical field of image pattern recognition, in particular to a woven bag defect detection method and system based on image pattern recognition. The method comprises the following steps: collecting a woven bag surface image and material reflection characteristic data; analyzing image illumination distribution, definition and contrast, and combining the reflection characteristics to identify material changes to obtain an image quality index; calculating a defect score for a potential defect area according to the index; receiving an operator's adjustment intention for a first threshold value, combining the image quality index and historical defect data to generate a suggested defect detection threshold value range prompt containing a second threshold value; performing first and second determinations on the defect score with the first threshold value and the second threshold value respectively, and if the cumulative number of defects found in the second determination but not in the first determination reaches a preset warning value, outputting a warning information. The technical problems of low detection accuracy, high missed detection rate and false alarm rate caused by image quality degradation, improper detection threshold setting and complex artifact interference in the prior art are solved.

Description

Technical Field

[0001] This application relates to the field of image pattern recognition technology, and more specifically, to a method and system for detecting defects in woven bags based on image pattern recognition. Background Technology

[0002] On high-speed woven bag production lines, manual quality inspection is inefficient and prone to omissions. Existing image recognition systems are also susceptible to misjudgments due to changes in lighting and surface contamination against complex textures. In actual operation, industrial camera lenses are exposed to plastic fibers, dust, and oil mist for extended periods, gradually forming uneven coatings that cause localized overly bright spots, blurred dark areas, decreased contrast, and degraded clarity. While the system automatically extends exposure time to compensate for the overall brightness reduction, the high-speed movement of the woven bags causes motion blur, further weakening or even blending minute defect outlines into the background. Due to the combined effects of uneven lighting, decreased contrast, and motion blur, parameters and feature sets established during development based on clear images are no longer suitable for the current input, making it difficult to distinguish normal textures from real defects, leading to a surge in false alarms. To reduce false alarms, operators often increase thresholds such as the "minimum continuous pixel area," which lowers the alarm frequency but causes real defects such as small broken threads, missed weave points, and minor tears to be directly ignored by the system, increasing the risk of missed detections. Meanwhile, after replacing the raw materials with a new batch, the reflective properties of the materials, combined with the lens scattering effect, produce artifacts such as irregular bright bands and dark lines. These artifacts, which resemble real defects, further exacerbate false positives and false negatives. This ultimately creates a complex dilemma: poor image quality, artificially imposed thresholds to mask small defects, and material changes introducing artifacts. These multiple factors intertwine, causing the system's original discrimination logic to fail. Both false positive and false negative rates are high, and these issues cannot be resolved simply by adjusting parameters or retraining.

[0003] To address this problem, existing technologies urgently need improvement. Summary of the Invention

[0004] This application discloses a method and system for detecting defects in woven bags based on image pattern recognition, aiming to solve the technical problems in existing woven bag defect detection methods, such as decreased detection accuracy, high false alarm rate and high missed detection rate caused by image quality degradation, improper detection threshold setting and complex artifact interference.

[0005] The technical solution of this application is as follows: In a first aspect, this application discloses a method for detecting defects in woven bags based on image pattern recognition, comprising the following steps: Acquire image data of woven bags and data on the reflective properties of the woven bag material; The image data of woven bags is analyzed to evaluate the illumination distribution, detail sharpness and overall contrast of the images. Combined with the reflection characteristic data, artifacts caused by the superposition of material property changes and optical system degradation are identified in the images, so as to obtain the image quality index characterizing the imaging quality of woven bag image data. Based on image quality indicators, the relative importance of different defect features in defect determination is adjusted, and based on the adjusted relative importance, the defect score is calculated for potential defect areas in woven bag image data. Receive the adjustment intention regarding the first defect detection threshold, and based on image quality indicators and historical defect data, predict the impact of the adjustment intention on the missed detection of minor defects, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold; Based on the defect score and the obtained first defect detection threshold, a first defect determination is performed on the potential defect area. At the same time, based on the defect score and the second defect detection threshold, a second defect determination is performed on the potential defect area. When the cumulative number of potential defects identified by the second defect determination that are determined to be non-defects by the first defect determination reaches a preset warning value, an early warning message is generated.

[0006] Further, it receives an adjustment intention regarding a first defect detection threshold, and based on image quality metrics and historical defect data, predicts the impact of the adjustment intention on the missed detection of minor defects, generating a suggested defect detection threshold range prompt that includes a second defect detection threshold, including: While acquiring woven bag image data, micro-pattern images of micro-pattern reference targets are continuously collected. The micro-pattern reference targets are set at fixed positions within the field of view of the industrial camera and are independently illuminated. The sharpness of the micro-pattern images is analyzed to obtain the micro-feature resolution index. Based on historical defect data, we obtain statistical analysis results of recently identified defects, and based on the statistical analysis results, we obtain initial prediction results of the impact of adjustment intentions on the missed detection of minor defects. When the micro-feature resolution index meets the preset resolution decline condition, a correction factor is applied to the initial prediction result to increase the predicted false negative risk. The second defect detection threshold is determined based on the increased false negative risk, and a suggested defect detection threshold range containing the second defect detection threshold is generated. Otherwise, the second defect detection threshold is determined based on the initial prediction result, and a suggested defect detection threshold range containing the second defect detection threshold is generated.

[0007] Furthermore, early warning information is generated, including: Receive potential defect information identified by secondary defect determination and determined as non-defect by primary defect determination, and record it as ignoring potential defect information; The cumulative number of ignored potential defect information, defect severity, and operator historical response behavior are obtained. Defect severity is determined by defect score and / or defect area, and historical response behavior includes operator response time and / or confirmation operation records for historical warning information. Based on the cumulative number, the severity of defects, and historical response behavior, the initial warning intensity and presentation method of the warning information are determined, and the warning information is output when the cumulative number reaches the preset warning value. After the warning information is output, if no effective response to the warning information is received within the preset time period and the cumulative number continues to increase, the warning intensity of the warning information is increased and / or the presentation of the warning information is changed. The initial warning intensity of subsequent warning information is updated based on the statistical results of response time in historical response behaviors.

[0008] Furthermore, based on image quality indicators, the relative importance of different defect features in defect determination is adjusted, and based on the adjusted relative importance, defect scores are calculated for potential defect areas in the woven bag image data, including: Identify at least two image degradation factors that coexist in the image quality index to obtain a multiple degradation state; For each defect feature, the reliability attenuation coefficient corresponding to each defect feature is calculated based on multiple degradation states and in combination with preset degradation interaction rules. Based on the reliability attenuation coefficient, a feature adjustment factor is obtained through nonlinear mapping, and the feature adjustment factor is used to nonlinearly adjust the relative importance of each defect feature in defect determination. Based on the relative importance after nonlinear adjustment, the defect score is calculated for potential defect areas in woven bag image data.

[0009] Further, it receives an adjustment intention regarding a first defect detection threshold, and based on image quality metrics and historical defect data, predicts the impact of the adjustment intention on the missed detection of minor defects, generating a suggested defect detection threshold range prompt that includes a second defect detection threshold, including: Record operator adjustment behaviors, including adjustment values, adjustment frequency, and adoption of system suggestions; Based on the adjusted behavior, calculate the operator's trust index and operational habit preference for the suggested defect detection threshold range; By combining image quality metrics and historical defect data, the impact of adjustment intentions on the missed detection of minor defects is predicted to obtain the initial risk of missed detection, and an initial suggested defect detection threshold range is generated based on the initial risk of missed detection. Based on the trust index and operational habits, the initial suggested defect detection threshold range is adjusted, and a second defect detection threshold is determined from the adjusted suggested defect detection threshold range, generating a suggested defect detection threshold range prompt. The recommended defect detection threshold range is displayed using an optimized presentation method.

[0010] Furthermore, the suggested defect detection threshold range is displayed using an optimized presentation method, including: The operator analyzes the historical response times of at least two different presentation methods and selects the presentation method with the shorter response time as the initial presentation method for subsequent suggested defect detection threshold range prompts. If no effective response is made to the suggested defect detection threshold range prompt within the preset suggested time and / or the second defect detection threshold is not adopted, and the adjustment intention continues to occur, the presentation intensity of the suggested defect detection threshold range prompt will be increased and / or switched to a higher intensity presentation mode.

[0011] Furthermore, based on the adjusted behavior, the operator's trust index and operational habit preferences regarding the suggested defect detection threshold range are calculated, including: Continuously monitor the stability of the data stream reflecting operator adjustments; When insufficient data volume and / or outliers are detected in the adjustment behavior data stream, the existing reliable adjustment behavior data is slightly perturbed to generate virtual operator adjustment behavior data to expand the adjustment behavior dataset. Real-time filtering is performed on the adjustment behavior data to identify occasional large adjustment behaviors, and these occasional large adjustment behaviors are marked as low-weight data to determine the recent stable adjustment behavior data. When calculating the trust index, recent stable adjustment behavior data is used, and a time decay factor is introduced into the recent stable adjustment behavior data. Based on the recent stable adjustment behavior data, the adoption mark and / or deviation of each adjustment behavior on the suggested defect detection threshold range are determined. The adoption mark and / or deviation are weighted and normalized according to the time decay factor to obtain the trust index. When calculating operational habit preferences, pattern recognition is performed on the operator's adjustment behavior in specific production situations, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences.

[0012] Furthermore, the adoption indicators and / or deviations are weighted and normalized according to a time decay factor to obtain a trust index, including: Continuously monitor the operating status of the production line, including the material batch of woven bags, equipment health status, and / or environmental parameters; When the operating status is detected to meet the preset change conditions, identify the historical adjustment behavior data corresponding to the current production situation; A difference analysis is performed on historical adjustment behavior data and recent stable adjustment behavior data to determine whether the adjustment behavior pattern corresponding to the recent stable adjustment behavior data is related to changes in the operational status. The difference analysis includes at least comparing the statistical characteristics of the adoption identifier and / or deviation degree corresponding to the historical adjustment behavior data and the recent stable adjustment behavior data to obtain the difference analysis results. When the results of the difference analysis indicate that the adjustment behavior pattern is related to the change in the operating status, a situational adaptability weight is introduced for the recently stable adjustment behavior data in the trust index calculation. The adoption markers and / or deviations are jointly weighted by the time decay factor and the context adaptability weight, and then weighted, summarized and normalized to obtain the context adaptability trust index.

[0013] Furthermore, pattern recognition is performed on operators' adjustment behaviors in specific production contexts, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences, including: Pattern recognition is performed on the operator's adjustment behavior in a specific production context to obtain the adjustment behavior pattern, wherein the adjustment behavior pattern includes the sequence characteristics of the adoption identifier and / or the distribution characteristics of the deviation degree; Obtain typical adjustment behavior patterns of other operators in specific production situations; The differences between the modified behavior pattern and the typical modified behavior pattern are compared to determine whether there is a significant deviation, and the comparison results are obtained. When the results of the difference comparison show that there is a significant deviation, the adjusted behavior pattern will be judged as an abnormal behavior pattern. A weighting adjustment mechanism is applied to abnormal behavior patterns to reduce their influence weight in the calculation of context-dependent operational habit preferences. Consistency assessment of the adjusted behavior patterns after weighting was performed to obtain context-dependent operational habit preferences.

[0014] Secondly, this application also discloses a defect detection system for woven bags based on image pattern recognition, comprising: The image data acquisition module is used to acquire image data of woven bags and reflective property data of woven bag materials; The image quality assessment module is used to analyze woven bag image data to evaluate the image's illumination distribution, detail sharpness, and overall contrast. It also combines reflectance data to identify artifacts in the image caused by the superposition of material property changes and optical system degradation, thereby obtaining image quality indicators that characterize the imaging quality of woven bag image data. The defect feature weight adjustment module is used to adjust the relative importance of different defect features in defect determination according to image quality indicators, and calculate the defect score for potential defect areas in woven bag image data based on the adjusted relative importance. The threshold suggestion module is used to receive the adjustment intention of the first defect detection threshold, and predict the impact of the adjustment intention on the missed detection of minor defects based on the image quality index and historical defect data, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold. The detection and early warning module is used to perform a first defect judgment on the potential defect area based on the defect score and the obtained first defect detection threshold. At the same time, it performs a second defect judgment on the potential defect area based on the defect score and the second defect detection threshold. When the cumulative number of potential defects identified by the second defect judgment and determined as non-defects by the first defect judgment reaches a preset warning value, an early warning message is generated.

[0015] Beneficial effects: The image pattern recognition-based defect detection method for woven bags disclosed in this application effectively solves the technical problems of decreased detection accuracy, high false alarm rate and high missed detection rate caused by image quality degradation, improper detection threshold setting and complex artifact interference in the prior art by introducing a number of innovative mechanisms. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating a defect detection method for woven bags based on image pattern recognition provided in this application.

[0017] Figure 2 This application provides a structural block diagram of a woven bag defect detection system based on image pattern recognition.

[0018] In the diagram: 1. Image data acquisition module; 2. Image quality assessment module; 3. Defect feature weight adjustment module; 4. Threshold suggestion module; 5. Detection and early warning module. Detailed Implementation

[0019] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0020] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this application, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0021] Reference Figure 1 This application proposes a defect detection method for woven bags based on image pattern recognition, including: S1000: Acquire image data of woven bags and reflective property data of woven bag materials; S2000: Analyzes woven bag image data to evaluate the image's illumination distribution, detail sharpness, and overall contrast. Combined with reflectance data, it identifies artifacts in the image caused by the superposition of material property changes and optical system degradation, thereby obtaining image quality indicators that characterize the imaging quality of woven bag image data. S3000: Based on the image quality index, adjust the relative importance of different defect features in defect determination, and calculate the defect score for potential defect areas in woven bag image data based on the adjusted relative importance. S4000: Receives an adjustment intention regarding the first defect detection threshold, and predicts the impact of the adjustment intention on the missed detection of minor defects based on image quality indicators and historical defect data, and generates a suggested defect detection threshold range prompt containing the second defect detection threshold; S5000: Perform a first defect determination on the potential defect area based on the defect score and the obtained first defect detection threshold. At the same time, perform a second defect determination on the potential defect area based on the defect score and the second defect detection threshold. When the cumulative number of potential defects identified by the second defect determination that are determined to be non-defects by the first defect determination reaches a preset warning value, an early warning message is generated.

[0022] Among them, woven bag image data refers to the visual information of the woven bag surface acquired by imaging equipment such as industrial cameras, such as grayscale or color images. The reflectivity data of the woven bag material refers to the reflection, absorption, and scattering characteristics of the woven bag material to different wavelengths of light, used to understand light and shadow changes in the image and assist in the identification of artifacts. Image quality indicators are a set of parameters that quantify the imaging quality of woven bag image data, comprehensively reflecting illumination uniformity, detail resolution, and contrast level. Defect score is a numerical value that quantifies the severity of potential defect areas, and its calculation considers multiple defect features and their relative importance. The first defect detection threshold is the initial defect judgment standard set by the operator; the second defect detection threshold is a more refined threshold intelligently recommended by the system based on image quality and historical data, used to assist in judgment. The preset warning value is the threshold at which the system triggers an early warning when the cumulative number of potential defects identified in the secondary defect judgment but judged as non-defects in the primary defect judgment reaches a certain level.

[0023] The method described in this application is typically deployed in the quality inspection stage of a woven bag production line. It uses an industrial camera to capture images of woven bags in real time, and a high-performance computing unit performs image processing and analysis. The system interacts with the production line control system to obtain production status information and provides inspection results and early warning information to the operator.

[0024] In practical implementation, the first step is to acquire image data of the woven bags and the reflectance characteristics data of the woven bag material. The woven bag image data is acquired in real time by an industrial camera, such as a high-resolution CCD or CMOS camera; the camera can be fixed above the production line to continuously photograph the passing woven bags. The reflectance characteristics data of the woven bag material can be obtained through offline experimental measurements, such as using a spectrometer to analyze the reflectance spectra of different batches of woven bag material under different lighting conditions, and storing the data in a database for system retrieval.

[0025] Next, the woven bag image data is analyzed to evaluate the image's illumination distribution, detail sharpness, and overall contrast. Combined with reflectance data, artifacts caused by the superposition of material property changes and optical system degradation are identified, thus obtaining image quality indicators characterizing the imaging quality of the woven bag image data. For example, illumination distribution can be evaluated by calculating the average brightness difference in different regions of the image; detail sharpness can be quantified by edge detection algorithms or Fourier transform analysis of high-frequency components of the image; overall contrast can be obtained by the standard deviation of pixel gray values ​​or histogram distribution. When identifying artifacts, the reflectance characteristics of the current image are compared with preset normal material reflectance characteristics. Combined with an optical system degradation model (e.g., a model of the impact of lens contamination on the image based on historical data), image segmentation and feature matching algorithms are used to identify abnormal regions that do not conform to normal texture patterns. Finally, the evaluation results are integrated to form a multi-dimensional image quality index vector.

[0026] Subsequently, based on image quality indicators, the relative importance of different defect features in defect determination is adjusted. Based on the adjusted relative importance, defect scores are calculated for potential defect areas in the woven bag image data. For example, when image quality indicators indicate motion blur, the system can reduce the weight of fine edge features and increase the weight of large-area color block anomalies or texture interruptions, because motion blur has a greater impact on fine edges. Defect features may include the size, shape, color, and degree of texture variation of the defect; the defect score can be calculated using a weighted summation method, that is, the quantified value of each defect feature is multiplied by its relative importance weight and then summed to obtain the total score.

[0027] Furthermore, the system receives an adjustment intention regarding the first defect detection threshold and, based on image quality metrics and historical defect data, predicts the impact of the adjustment intention on the missed detection of minor defects, generating a suggested defect detection threshold range that includes a second defect detection threshold. For example, an operator might input a higher first defect detection threshold through a human-machine interface to reduce false alarms; after receiving the adjustment intention, the system combines the current image quality metrics (e.g., minor defects are more likely to be missed when image clarity is low) with historical defect data (e.g., types of minor defects that were easily missed under similar image quality conditions in the past), uses a machine learning model to predict the risk of missed minor defects, and generates a suggested defect detection threshold range accordingly, which includes a recommended second defect detection threshold, to minimize the missed detection rate of minor defects while ensuring detection efficiency.

[0028] Finally, a first defect determination is performed on the potential defect area based on the defect score and the obtained first defect detection threshold. Simultaneously, a second defect determination is performed on the potential defect area based on the defect score and the second defect detection threshold. When the cumulative number of potential defects identified by the second defect determination but judged as non-defects by the first defect determination reaches a preset warning value, an early warning message is generated. The first defect determination is based on the first defect detection threshold; for example, a defect score higher than the first defect detection threshold is determined to be a defect. The second defect determination is based on the second defect detection threshold, which is usually lower or more refined than the first defect detection threshold, aiming to capture minute defects that might be missed by the first defect determination. The system continuously monitors the cumulative number of potential defects "judged as non-defects by the first defect determination," and when it reaches the preset warning value, an early warning message is generated, alerting the operator to the risk of missed detection and suggesting a reassessment of the detection threshold or a check of the system status.

[0029] In another embodiment of this application, S4000 is further proposed to include: S4001: While acquiring woven bag image data, continuously acquire micro-pattern images of micro-pattern reference targets. The micro-pattern reference targets are set at fixed positions within the field of view of the industrial camera and are independently illuminated. The sharpness of the micro-pattern images is analyzed to obtain the micro-feature resolution index. S4002: Based on historical defect data, obtain statistical analysis results of recently identified defects, and based on the statistical analysis results, obtain initial prediction results of the impact of adjustment intentions on the missed detection of minor defects; S4003: When the micro-feature resolution index meets the preset resolution decline condition, apply a correction factor to the initial prediction result to increase the predicted false negative risk, determine the second defect detection threshold based on the increased false negative risk, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold; otherwise, determine the second defect detection threshold based on the initial prediction result, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold.

[0030] Specifically, while acquiring woven bag image data, micro-pattern images of a micro-pattern reference target are continuously acquired. The micro-pattern reference target is a physical calibrator with a known fine structure, such as a high-resolution line, dot matrix, or checkerboard pattern. It is positioned at a fixed location within the industrial camera's field of view and is independently illuminated to ensure its imaging conditions are unaffected by the surface characteristics of the woven bag or fluctuations in ambient light. Independent illumination isolates the imaging conditions of the micro-pattern reference target, allowing the sharpness analysis of the micro-pattern images to more accurately reflect the optical performance of the industrial camera itself. Sharpness analysis of the micro-pattern images quantifies the industrial camera's actual ability to resolve minute features. Sharpness analysis can employ methods such as frequency domain analysis based on Fourier transform, edge sharpness detection, modulation transfer function (MTF) calculation, or point spread function (PSF) evaluation to obtain a micro-feature resolution index. The micro-feature resolution index is a numerical indicator used to characterize the camera's ability to capture minute details under current operating conditions; a lower index value generally indicates poorer resolution.

[0031] Furthermore, statistical analysis results of recently identified defects are obtained based on historical defect data. Historical defect data includes information such as the type, quantity, severity, corresponding image quality indicators, and detection thresholds of defects detected by the system over a past period. A correlation model between defect characteristics and the risk of missed detection is established through statistical analysis, and then an initial prediction result is obtained based on the statistical analysis results regarding the impact of adjustment intentions on the missed detection of minor defects. The initial prediction result is used to characterize the risk of missed detection of minor defects that may be caused by the operator adjusting the first defect detection threshold under ideal or conventional optical conditions.

[0032] Based on this, when the micro-feature resolution index meets a preset resolution degradation condition, a correction factor is applied to the initial prediction result to increase the predicted false negative risk. The preset resolution degradation condition is, for example, that the micro-feature resolution index is below a preset threshold, indicating a significant degradation in the optical performance of the industrial camera, which may lead to an underestimation of the risk of missing minute defects. The correction factor is a coefficient greater than 1 and can be dynamically adjusted according to the degree of resolution degradation; for example, the more severe the resolution degradation, the larger the correction factor. Subsequently, a second defect detection threshold is determined based on the increased false negative risk, and a suggested defect detection threshold range including the second defect detection threshold is generated. Otherwise, when the micro-feature resolution index does not meet the preset resolution degradation condition, indicating normal camera optical performance, the second defect detection threshold is directly determined based on the initial prediction result, and a suggested defect detection threshold range including the second defect detection threshold is generated.

[0033] By introducing real-time monitoring of the micro-feature resolution capability of industrial cameras, this application takes into account the degradation of the optical system itself when predicting the risk of missing small defects.

[0034] In some preferred embodiments, the following specific example illustrates the situation: Imagine an industrial camera on a woven bag production line, used to capture images of the bags. A micro-pattern reference target, printed with a standard resolution test image, is positioned at a fixed location within the camera's field of view. While acquiring woven bag images, the system simultaneously captures micro-pattern images of the reference target every 10 minutes. The system then performs sharpness analysis on these micro-pattern images, such as calculating the average edge sharpness, to obtain a micro-feature resolution index.

[0035] For example, initially, the micro-feature resolution index is 0.85. After the operator adjusts the first defect detection threshold, the system predicts, based on historical defect data and current image quality indicators, that this adjustment will increase the risk of missing micro-defects by 5%, generating an initial prediction result. After a period of operation, due to slight wear on the camera lens, the micro-feature resolution index drops to 0.70. The system detects that this index meets the preset resolution decrease condition (e.g., the preset threshold is 0.75). When the operator adjusts the first defect detection threshold again, the system first obtains the initial prediction result based on historical defect data (e.g., a 5% increase in the risk of missing defects), and then applies a correction factor (e.g., 1.2) to increase the predicted increase in the risk of missing defects to 5%. 1.2 = 6%; Based on the increased risk of missed detection after the adjustment, a more conservative second defect detection threshold is determined, for example, the threshold is adjusted from 0.6 to 0.55, and a suggested defect detection threshold range including the second defect detection threshold of 0.55 is generated. Conversely, if the micro-feature resolution index remains at 0.80 and the preset resolution decline condition is not met, the system directly determines the second defect detection threshold based on the initial prediction result (e.g., a 5% increase in the risk of missed detection), for example, it remains at 0.6, and a corresponding suggested defect detection threshold range is generated.

[0036] In another embodiment of this application, a method for generating early warning information is further proposed, comprising: S5100: Receives potential defect information identified by the secondary defect determination and determined as non-defect by the primary defect determination, which is denoted as ignoring potential defect information; S5200: Acquire the cumulative number of ignored potential defect information, defect severity, and operator historical response behavior, wherein defect severity is determined by defect score and / or defect area, and historical response behavior includes operator response time and / or confirmation operation records for historical warning information; S5300: Determines the initial warning intensity and presentation method of the warning information based on the cumulative number, the severity of the defect, and historical response behavior, and outputs the warning information when the cumulative number reaches the preset warning value; S5400: After outputting a warning message, if no effective response to the warning message is received within a preset time period and the cumulative number continues to increase, the warning intensity of the warning message is increased and / or the presentation of the warning message is changed. S5500: Update the initial warning intensity of subsequent warning information based on the response time statistics in historical response behaviors.

[0037] Specifically, receiving ignored potential defect information refers to the system continuously monitoring potential defects identified by secondary defect assessment but judged as non-defects by primary defect assessment. This ignored potential defect information is used to characterize potential missed detection risks in primary defect assessment. The system acquires the cumulative number of ignored potential defect information, defect severity, and operator historical response behavior to form a basis for early warning decisions: defect severity is quantified by defect score and / or defect area; for example, a higher defect score or a larger defect area indicates a more severe defect. Operator historical response behavior includes the operator's response time and / or confirmation operation records for historical early warning information, used to characterize the operator's level of attention to early warning information and processing efficiency.

[0038] Furthermore, the system determines the initial warning intensity and presentation method of the warning information based on the cumulative number, the severity of the defect, and historical response behavior, and outputs the warning information when the cumulative number reaches a preset warning value, thus achieving adaptive warning. For example, for operators with higher defect severity or longer historical response times, the system can set a higher initial warning intensity (such as a more prominent visual cue or a louder audible alarm) and a more direct presentation method (such as a forced pop-up).

[0039] Furthermore, after outputting the warning information, if no effective response to the warning information is received within a preset time period and the cumulative number continues to increase, the system increases the warning intensity of the warning information and / or changes the presentation method of the warning information; as a preferred implementation, the system updates the initial warning intensity of subsequent warning information based on the response time statistics in historical response behaviors, so as to achieve self-learning and optimization.

[0040] In some preferred embodiments, it is assumed that on a woven bag production line, the system inspects products through primary defect detection and secondary defect detection. In one instance, secondary defect detection identifies 10 minor scratches in a batch of products. However, these scratches have low defect scores and do not reach the threshold of primary defect detection. Therefore, they are judged as non-defects by primary defect detection, and this minor scratch information is recorded as ignoring potential defect information. The system obtains the cumulative number of these 10 scratches, their relatively low defect severity, and queries the historical response time of operator A to similar minor defects, which is usually relatively long. Based on this, a medium-intensity initial warning is determined, such as displaying a flashing yellow icon on the operation interface, accompanied by a slight prompting sound, and notifying the operator in the form of a pop-up window. If operator A does not acknowledge the warning within a preset 30 seconds and subsequently accumulates 5 more similar ignored potential defect information, the system automatically increases the warning intensity, such as changing the yellow icon to red, increasing the volume of the prompting sound, and forcing the pop-up window to be at the top. At the same time, the system updates the initial warning intensity of subsequent warnings based on the statistical results of operator A's response time to historical warning information. For example, if operator A frequently delays response, a slightly higher intensity initial warning will be used the next time a similar situation occurs.

[0041] In another embodiment of this application, S3000 specifically includes: S3100: Identify at least two image degradation factors that exist simultaneously in the image quality index to obtain a multiple degradation state; S3200: For each defect feature, calculate the reliability attenuation coefficient corresponding to each defect feature based on multiple degradation states and in combination with preset degradation interaction rules. S3300: Based on the reliability attenuation coefficient, a feature adjustment factor is obtained through nonlinear mapping, and the feature adjustment factor is used to nonlinearly adjust the relative importance of each defect feature in defect determination. S3400: Based on the relative importance after nonlinear adjustment, the defect score is calculated for potential defect areas in woven bag image data.

[0042] Specifically, identifying at least two image degradation factors simultaneously present in an image quality index to obtain a multiple degradation state refers to the system's comprehensive analysis of image quality indicators to identify multiple imaging quality degradation factors present in the image simultaneously. For example, an image quality index may simultaneously indicate two degradation factors: "low contrast" and "local blurring." A multiple degradation state is a combination pattern of these degradation factors, used to more completely characterize the true quality of the image.

[0043] Specifically, for each defect feature, based on multiple degradation states and pre-defined degradation interaction rules, a reliability attenuation coefficient is calculated for each defect feature. This coefficient refers to the difference in detectability of different defect types (e.g., stains, holes, uneven texture) under different multiple degradation states. The degradation interaction rules describe the mutual influence of degradation factors and their combined effect on the visibility / identifiability of defect features. For example, when low lighting and high noise coexist, the reliability attenuation of feature extraction for fine texture defects may be more significant than that of a single degradation. The reliability attenuation coefficient is used to quantify the degree of reliability reduction of a defect feature relative to the ideal situation under a specific multiple degradation state.

[0044] In practical applications, feature adjustment factors are obtained through nonlinear mapping based on the reliability decay coefficient. These feature adjustment factors are then used to nonlinearly adjust the relative importance of each defect feature in defect determination. This means that the reliability decay coefficient is taken as input and converted into a feature adjustment factor through a nonlinear function or lookup table. This process characterizes the relationship of "small adjustment for slight degradation and significant adjustment for severe degradation" and applies it to the relative importance weight of the original defect features. This reduces the relative importance of defect features with lower reliability and increases or maintains the relative importance of defect features with higher reliability.

[0045] Furthermore, based on the nonlinearly adjusted relative importance, defect scores are calculated for potential defect regions in woven bag image data. The defect score is calculated by combining the defect feature values ​​extracted from the image with the updated relative importance of the defect features, and then calculating a comprehensive defect score for each potential defect region. This makes the defect score explicitly consider the impact of the current image quality on the detectability of different defect features.

[0046] In some preferred embodiments, it is assumed that the image data acquired by the industrial camera on the woven bag production line simultaneously exhibits two degradation factors: "low lighting" and "slight blur." These two factors together constitute a specific multiple degradation state. For the "stain" defect feature, its color features may not be obvious under low lighting, but its edge features are still discernible under slight blur. For the "uneven weave" defect feature, its texture features may become difficult to distinguish under slight blur, but its impact is relatively small under low lighting.

[0047] At this point, the system will calculate the reliability decay coefficients of the "stain" and "uneven texture" features under the current "low lighting + slight blur" multiple degradation state, based on the preset degradation interaction rules. For example, the reliability decay coefficient of the "stain" feature may be 0.8 (indicating a 20% decrease in reliability), while the reliability decay coefficient of the "uneven texture" feature may be 0.6 (indicating a 40% decrease in reliability).

[0048] Subsequently, these reliability attenuation coefficients are converted into feature adjustment factors through a nonlinear mapping. For example, an attenuation coefficient of 0.8 may be mapped to an adjustment factor of 0.9, while an attenuation coefficient of 0.6 may be mapped to an adjustment factor of 0.7. These adjustment factors will be applied, respectively, to the original relative importance of the "stains" and "uneven texture" features in defect determination.

[0049] Assuming the original relative importance of the "stain" feature is 0.4, and the original relative importance of the "uneven texture" feature is 0.3, after nonlinear adjustment, the relative importance of the "stain" feature becomes 0.4. 0.9 = 0.36, while the relative importance of the "uneven weave" feature becomes 0.3. 0.7 = 0.21.

[0050] Finally, the system calculates defect scores for potential defect areas in the woven bag image data based on these non-linearly adjusted relative importances. In this way, the system can more accurately reflect the contribution of different defect features to the final defect determination under the current image degradation conditions, thereby avoiding misjudgments or missed detections due to image quality issues and improving the accuracy and reliability of detection.

[0051] In another embodiment of this application, S4000 further includes: S4100: Records operator adjustment behavior, including adjustment values, adjustment frequency, and adoption of system suggestions; S4200: Based on the adjustment behavior, calculate the operator's trust index and operating habit preference for the suggested defect detection threshold range; S4300: Combining image quality metrics and historical defect data, predicts the impact of adjustment intentions on the missed detection of minor defects to obtain an initial risk of missed detection, and generates an initial suggested defect detection threshold range based on the initial risk of missed detection. S4400: Based on the trust index and operational habits, adjust the initial suggested defect detection threshold range, determine the second defect detection threshold from the adjusted suggested defect detection threshold range, and generate a suggested defect detection threshold range prompt; S4500: Displays suggested defect detection threshold ranges using an optimized presentation method.

[0052] Specifically, recording operator adjustment behavior means that the system continuously monitors and stores various adjustments made by operators to the first defect detection threshold, including the specific value of each adjustment, the frequency of the adjustment, and whether the operator adopts the previously suggested defect detection threshold range suggested by the system, in order to form a basic dataset for characterizing operator behavior patterns.

[0053] Specifically, based on the adjustment behavior, the system calculates the operator's trust index and operational habit preference for the suggested defect detection threshold range. This means that the system quantifies the operator's acceptance of the system's suggestions and their tendency to adjust the threshold in different situations based on the adjustment behavior data. The trust index reflects the operator's reliance on the system's recommended threshold. For example, if the operator frequently adopts the system's suggestions, the trust index is higher. The operational habit preference reveals how the operator tends to adjust the threshold under specific production conditions. For example, some operators may tend to set stricter thresholds to reduce missed detections, while others may tend to set more lenient thresholds to reduce false alarms.

[0054] In practical applications, combining image quality indicators and historical defect data, the system predicts the impact of adjustment intentions on the missed detection of minor defects to obtain the initial risk of missed detection. Based on the initial risk of missed detection, it generates a preliminary suggested defect detection threshold range. This means that the system first completes a technical risk assessment based on image quality indicators and historical defect data to obtain the risk of missed detection of minor defects that may be caused by the operator's current adjustment intentions, and outputs a preliminary suggested defect detection threshold range obtained solely from technical data analysis.

[0055] Furthermore, based on the trust index and operational habits, the initial suggested defect detection threshold range is adjusted, and a second defect detection threshold is determined from the adjusted range. This generates a suggested defect detection threshold range prompt. This means that the initial suggested defect detection threshold range is modified by incorporating operator-specific factors, making the recommendations more aligned with the operator's actual operating habits and psychological expectations while meeting detection risk constraints. For example, if the operator has a high trust index, the system may be more inclined to recommend a threshold closer to the operator's historical adoption behavior within a safe range. If the operator has a clear preference in a specific situation, and that preference has proven effective, the system fine-tunes the suggested range based on that preference, thereby determining the second defect detection threshold from the adjusted range and generating a suggested defect detection threshold range prompt. The second defect detection threshold is always used for internal risk monitoring in secondary defect judgment. "Adoption" only affects the setting of the first defect detection threshold and production line alarm preferences, and does not affect the system's secondary judgment logic.

[0056] In addition, the optimized presentation method for displaying suggested defect detection threshold range prompts means that the system selects a more accessible and effective display format based on the operator's response habits or the current production environment. This can be achieved through methods such as color, flashing, sound prompts, or highlighting in specific interface areas to improve the visibility and adoption efficiency of suggested defect detection threshold range prompts.

[0057] The solution proposed in this application incorporates operator-specific factors into the threshold recommendation process by recording and analyzing operator adjustment behaviors, thereby alleviating the "human-machine incoordination" problem caused by traditional threshold recommendation systems that rely solely on image quality indicators and historical defect data.

[0058] In some preferred embodiments, it is assumed that on a woven bag production line, the system continuously monitors and records operator A's adjustment behavior regarding a first defect detection threshold. For example, operator A adopted the system's suggested threshold range 80% of the time over the past week, and when faced with a specific type of image blur (as indicated by image quality indicators), he always tended to slightly lower the threshold by 0.05 to improve sensitivity. Based on these records, the system calculates operator A's trust index in the system's recommendations as 0.8 and identifies his operational habit preference of "slightly lowering the threshold" in image blurry situations.

[0059] When the system receives an intention from operator A to adjust the first defect detection threshold, it first combines the current image quality index (e.g., a slight decrease in detail sharpness is detected) and historical defect data to predict the initial risk of missed detection that this adjustment may cause, and generates an initial suggested defect detection threshold range, such as [0.45, 0.55].

[0060] Subsequently, the system will utilize operator A's trust index of 0.8 and their operational habits in blurred image scenarios. Due to the high trust level, the system will tend to make fine adjustments based on the initial suggestions. Considering operator A's tendency to lower the threshold when the image is blurred, the system may adjust the suggested range to [0.43, 0.53] and determine the second defect detection threshold from it, for example, 0.48.

[0061] Ultimately, based on operator A's historical response data, the system will select an optimized presentation method. For example, it might highlight the suggested defect detection threshold range [0.43, 0.53] in green on the main interface, accompanied by a brief text prompt: "Considering the current decrease in image clarity and your operating habits, the suggested defect detection threshold range has been slightly adjusted." This personalized suggestion, tailored to the operator's habits, will greatly increase the likelihood of the operator adopting the suggestion, thereby managing the defect detection threshold more effectively and reducing the missed detection of minor defects.

[0062] In another embodiment of this application, step S4500 is further proposed to include: S4510: Analyze the historical response times of the operator for at least two different presentation methods, and select the presentation method with the shorter response time as the initial presentation method for subsequent suggested defect detection threshold range prompts based on the historical response times; S4520: When there is no effective response to the suggested defect detection threshold range prompt within the preset suggested time and / or the second defect detection threshold is not adopted, and the adjustment intention continues to occur, increase the presentation intensity of the suggested defect detection threshold range prompt and / or switch to a higher intensity presentation mode.

[0063] Specifically, "historical response time of operators to at least two different presentation methods" refers to the time that the system continuously records from the appearance of the prompt to the operator's effective response (such as confirmation, modification, or adoption) when faced with different forms of suggested defect detection threshold range prompts (e.g., pop-ups, flashing, voice prompts, color changes, etc.); "presentation method with shorter response time" refers to the presentation method with the shortest average response time of operators in historical data, which the system uses as the default initial presentation method for subsequent suggested defect detection threshold range prompts to improve the efficiency of prompt delivery.

[0064] The "preset suggestion time" refers to the time window set by the system for operators to respond to suggested defect detection threshold range prompts, such as 5 seconds, 10 seconds, or dynamically adjusted according to production rhythm; "effective response" means that the operator has made a clear interaction with the suggested defect detection threshold range prompts, such as clicking the "confirm" button, manually adjusting the threshold and saving, etc.; "second defect detection threshold not adopted" means that the operator ignores the suggestion after receiving it, or manually sets a first defect detection threshold that is significantly different from the second defect detection threshold; "adjustment intentions continue to occur" means that the operator tries to adjust the first defect detection threshold multiple times in a short period of time, indicating that they have doubts or dissatisfaction with the current threshold setting. When no effective response and / or no adoption of the second defect detection threshold occurs within the preset suggestion time, and the adjustment intentions continue to occur, the system will "increase the presentation intensity of the suggested defect detection threshold range prompts" and / or "switch to a higher intensity presentation mode," such as changing the prompt box from a normal color to flashing red, or adding voice broadcast on the basis of text prompts, in order to attract the operator's attention through stronger visual or auditory stimulation and avoid potential missed detection risks.

[0065] As a specific implementation, suppose on a woven bag production line, the system needs to suggest a defect detection threshold range to the operator. The system first calculates the operator's average response time over a period of time to "pop-up prompts," "flashing border prompts," and "voice prompts." If the "flashing border prompt" is the shortest, for example, 3 seconds, while the other two are 5 seconds and 8 seconds respectively, the system automatically selects the "flashing border prompt" as the initial presentation method for subsequent suggested defect detection threshold range prompts. Furthermore, if after the system issues a "flashing border prompt," and the operator does not click the confirmation button or manually adjust the threshold within a preset 10-second suggestion time, and the system detects that the operator attempts to manually adjust the first defect detection threshold twice consecutively within the next 30 seconds (indicating a continued adjustment intention), the system immediately increases the intensity of the prompt, for example, by increasing the flashing frequency of the "flashing border prompt," accompanied by an urgent "beep" voice prompt, or even changing the background color of the entire display area to red, to more strongly urge the operator to respond effectively to the suggested defect detection threshold range prompt, thereby avoiding the potential for missed minor defects due to failure to promptly adopt the second defect detection threshold.

[0066] In another embodiment of this application, S4200 further includes: S4210: Continuously monitor the stability of the data stream reflecting operator adjustment behaviors; S4220: When insufficient data volume and / or outliers are detected in the adjustment behavior data stream, the existing reliable adjustment behavior data is slightly perturbed to generate virtual operator adjustment behavior data to expand the adjustment behavior dataset; S4230: Perform real-time filtering on the adjustment behavior data, identify occasional large adjustment behaviors, mark occasional large adjustment behaviors as low-weight data, and determine the recent stable adjustment behavior data; S4240: When calculating the trust index, recent stable adjustment behavior data is used, and a time decay factor is introduced into the recent stable adjustment behavior data. Based on the recent stable adjustment behavior data, the adoption mark and / or deviation of each adjustment behavior on the suggested defect detection threshold range are determined. The adoption mark and / or deviation are weighted and normalized according to the time decay factor to obtain the trust index. S4250: When calculating operational habit preferences, pattern recognition is performed on the operator's adjustment behavior in a specific production context, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences.

[0067] Specifically, continuously monitoring the stability of the operator's adjustment behavior data stream means that the system tracks in real time the operator's adjustment value, adjustment frequency, and adoption of system suggestions when adjusting the first defect detection threshold, and evaluates the volatility and completeness of the data from a time dimension, so as to promptly detect insufficient data, missing data or abnormal jumps, and provide a basis for subsequent data processing.

[0068] Specifically, when the amount of data in the adjustment behavior data stream is insufficient and / or outliers are detected, the existing reliable adjustment behavior data is slightly perturbed to generate virtual operator adjustment behavior data to expand the adjustment behavior dataset. This means that when historical adjustment behavior data is insufficient to support robust statistical analysis or there are outliers that significantly deviate from the normal pattern, the system uses the adjustment behavior data that has been determined to be reliable as a benchmark, and generates simulated samples by introducing random noise or small offsets to increase the amount of data and distribution coverage, thereby improving the robustness and stability of subsequent calculations.

[0069] In practical applications, real-time filtering of adjustment behavior data is performed to identify occasional large-scale adjustments. These occasional large-scale adjustments are then marked as low-weight data to determine recently stable adjustment behavior data. This involves the system smoothing the data sequence using methods such as moving averages and median filtering, and combining this with thresholds or statistical criteria to identify transient, large-scale deviations in adjustment behavior. The identified occasional large-scale adjustments are assigned low weights to reduce their contribution in subsequent calculations, thereby avoiding masking the operator's long-term stable habits and determining recently stable adjustment behavior data.

[0070] When calculating the trust index, recent stable adjustment behavior data is used, and a time decay factor is introduced into this data. Based on this data, the adoption indicator and / or deviation of each adjustment behavior relative to the suggested defect detection threshold range are determined. The adoption indicator and / or deviation are weighted, summarized, and normalized according to the time decay factor to obtain the trust index. The adoption indicator represents whether the operator has adopted the system's suggested second defect detection threshold, and the deviation quantifies the difference between the operator's actual adjustment value and the system's suggested value. The time decay factor gives higher weight to recent behaviors and lower weight to earlier behaviors, allowing the trust index to dynamically reflect the operator's current level of trust in the system's suggestions. The weighted summarization and normalization process integrates the weighted adoption indicator and / or deviation into a unified trust index.

[0071] When calculating operational habit preferences, pattern recognition is performed on operators' adjustment behaviors in specific production scenarios, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences. This means that the system uses different production scenarios (e.g., different material batches, different equipment states, different environmental parameters, etc.) as grouping conditions to perform pattern recognition on operators' adjustment trajectories, extracts the patterns of their tendency to raise or lower thresholds and magnitudes in specific scenarios, and evaluates the consistency using indicators such as frequency and stability, thereby forming a context-dependent operational habit preference model to support more accurate prediction of adjustment intentions in similar scenarios.

[0072] In some preferred embodiments, suppose a woven bag production line needs to adjust a first defect detection threshold during operation. The system first continuously monitors the operator's historical adjustment behavior data stream, including the value of each adjustment, the frequency of adjustment, and whether the system's previous suggestions were adopted. If the system finds that the amount of operator adjustment data in the past week is insufficient, or if there is an abnormally large adjustment (e.g., suddenly adjusting the threshold from 50 to 1000), the system will activate a data processing mechanism: for insufficient data, virtual adjustment behavior data is generated based on stable adjustment data over the past month through small-range random perturbations to expand the adjustment behavior dataset; for occasional large adjustment behaviors, they are identified and marked as low-weight data to ensure that abnormal behavior does not have an excessive impact on the calculation of trust index and operating habit preferences.

[0073] When calculating the operator's trust index, the system uses recent (e.g., the last two weeks) stable adjustment behavior data; for each adjustment behavior, an adoption indicator and deviation degree are determined, and a time decay factor is introduced, for example, the behavior of the most recent day has a weight of 1.0, two days ago has a weight of 0.9, and so on; then the weighted adoption indicator and deviation degree are weighted, summarized and normalized to obtain a trust index between 0 and 1, for example, 0.85, which indicates that the operator has a high degree of trust in the system's suggestions.

[0074] When calculating operational habit preferences, the system identifies operator adjustment behavior patterns in specific production scenarios. For example, when producing woven bags of a specific material, operators typically fine-tune the threshold by 5 units based on the system's suggestion. The system analyzes the operator's adjustment behavior in past production runs of woven bags of that material, identifies the "specific material - fine-tune - 5" pattern, and assesses its consistency. When this pattern consistently occurs, it is established as the operator's operational habit preference in that scenario. Ultimately, the system uses a trust index and context-dependent operational habit preferences to adjust the initially suggested defect detection threshold range, thereby generating suggested defect detection threshold ranges that better reflect the operator's actual situation and preferences.

[0075] In another embodiment of this application, a trust index is obtained by weighting and normalizing the adoption identifier and / or deviation according to a time decay factor, including: S4231: Continuously monitor the operating status of the production line, including the material batch of woven bags, equipment health status and / or environmental parameters; S4232: When the operating status is detected to meet the preset change conditions, identify the historical adjustment behavior data corresponding to the current production situation; S4233: Perform a difference analysis on historical adjustment behavior data and recent stable adjustment behavior data to determine whether the adjustment behavior pattern corresponding to the recent stable adjustment behavior data is related to changes in the operating status. The difference analysis shall at least include comparing the statistical characteristics of the adoption identifier and / or deviation degree corresponding to the historical adjustment behavior data and the recent stable adjustment behavior data to obtain the difference analysis results. S4234: When the results of the difference analysis indicate that the adjustment behavior pattern is related to the change in the operating status, a situational adaptability weight is introduced for the recently stable adjustment behavior data in the trust index calculation. S4235: The adoption identifier and / or deviation are jointly weighted by the time decay factor and the context adaptability weight, and then weighted and normalized to obtain the context adaptability trust index.

[0076] Specifically, continuous monitoring of the production line's operational status refers to the system collecting various parameters related to the production line's operation in real time or near real time to construct a profile of the current production situation. These parameters include, but are not limited to, material batch information for woven bags (such as supplier, production date, material composition, etc.), equipment health status (such as wear and tear of industrial cameras, light source brightness decay, and smooth operation of mechanical components, etc.), and environmental parameters (such as workshop temperature, humidity, and light intensity, etc.).

[0077] When the system detects that the operating status meets the preset change conditions, such as when the material batch is changed, the health index of key equipment components is lower than the preset threshold, or the ambient temperature fluctuates beyond the safe range, the system will trigger the situation recognition mechanism and identify historical adjustment behavior data that is similar to or corresponds to the current production situation from the historical database, so as to provide a reference for the operator's behavior in the current situation.

[0078] The purpose of conducting a difference analysis between historical adjustment behavior data and recent stable adjustment behavior data is to assess whether there are significant differences between recent operator behavior patterns and historical behavior patterns under the current production situation. Difference analysis can employ various statistical methods, such as comparing statistical characteristics like the distribution of adoption indicators (e.g., adoption rate, rejection rate), the mean, variance, or distribution shape of deviations. This comparison helps determine whether the operator behavior patterns reflected in the recent stable adjustment behavior data are still applicable to the current production situation.

[0079] When the discrepancy analysis results indicate a correlation between adjustment behavior patterns and changes in operational status, it means that recently stable adjustment behavior data may be affected by contextual changes and cannot fully represent the operator's trust level in the current context. To improve the accuracy of the trust index, a contextual adaptability weight is introduced into the calculation of the trust index for recently stable adjustment behavior data. This contextual adaptability weight can be dynamically adjusted based on the discrepancy analysis results; for example, the greater the discrepancy, the lower the contextual adaptability weight, and vice versa, or higher weight can be assigned to data under specific contexts according to preset rules.

[0080] Finally, the adoption markers and / or deviations are jointly weighted according to the time decay factor and the context adaptability weight, and then weighted and normalized to obtain the context adaptability trust index. This trust index reflects both the time proximity of operator behavior and the particularity of the current production situation, thus more accurately and flexibly reflecting the operator's true trust level under different production conditions.

[0081] The solution proposed in this application solves the problem of insufficient accuracy of the trust index in dynamic production scenarios by continuously monitoring the operating status of the production line and using it as a key factor in adjusting the trust index calculation.

[0082] In another embodiment of this application, S4250 further includes: S4251: Perform pattern recognition on the operator's adjustment behavior in a specific production context to obtain the adjustment behavior pattern, wherein the adjustment behavior pattern includes the sequence characteristics of the adoption identifier and / or the distribution characteristics of the deviation degree; S4252: Obtain typical adjustment behavior patterns of other operators in a specific production context; S4253: Compare the differences between the adjusted behavior pattern and the typical adjusted behavior pattern to determine whether there is a significant deviation and obtain the difference comparison results; S4254: When the results of the difference comparison show that there is a significant deviation, the adjusted behavior pattern will be judged as an abnormal behavior pattern; A weighting adjustment mechanism is applied to abnormal behavior patterns to reduce their influence weight in the calculation of context-dependent operational habit preferences. S4255: Conduct a consistency assessment of the adjusted behavior patterns after weighting to obtain context-dependent operational habit preferences.

[0083] Specifically, pattern recognition of operator adjustment behavior in specific production scenarios refers to analyzing the adjustment behavior of operators in specific production scenarios (e.g., specific material batches, equipment status, or environmental conditions) in response to suggested defect detection threshold ranges, using data mining or machine learning algorithms, to extract reusable behavioral patterns. The adjustment behavior patterns include the sequential features of adoption identifiers (e.g., the number and order of consecutive adoption or rejection of system suggestions) and the distribution features of deviation (e.g., the range and frequency of differences between the adjustment value and the system suggestion value), used to characterize the operator's behavioral patterns in different scenarios.

[0084] Among them, obtaining typical adjustment behavior patterns of other operators in a specific production situation refers to collecting and analyzing historical adjustment behavior data of multiple experienced or "standard" operators in the same or similar production situations, and forming one or more benchmark behavior patterns through aggregation statistics or clustering; the typical adjustment behavior patterns serve as a comparison reference, representing operating habits that are generally accepted or proven effective in that situation.

[0085] In practical applications, comparing the differences between the adjusted behavior pattern and the typical adjusted behavior pattern means using statistical methods or pattern matching algorithms to quantify the differences between the two in order to determine whether there is a significant deviation; for example, comparing the similarity of adoption identifier sequences and the statistical distance of the deviation distribution (such as KL divergence or Euclidean distance) to identify the uniqueness or abnormality of the current operator's behavior.

[0086] When the difference comparison results show a significant deviation, the adjusted behavior pattern will be judged as an abnormal behavior pattern, indicating that it is statistically significantly different from the behavior patterns of most operators or standard operators, which may correspond to misoperation, fatigue, lack of experience or distrust of the system.

[0087] Furthermore, applying a weight adjustment mechanism to abnormal behavior patterns means reducing the influence weight of abnormal behavior patterns when calculating context-dependent operational habit preferences; for example, assigning lower weighting coefficients to abnormal behavior patterns, or directly excluding some abnormal data in the calculation, in order to reduce their interference with the final result and improve the robustness of the preference model.

[0088] Therefore, conducting a consistency assessment of the adjusted behavior patterns after weighting refers to comprehensively analyzing the remaining more representative adjusted behavior patterns after weakening the impact of abnormal behavior, assessing their consistency and stability, and thus obtaining more accurate and reliable context-dependent operational habit preferences.

[0089] The solution proposed in this application reduces the impact of abnormal or inconsistent behaviors in complex production scenarios on the accuracy of calculating context-dependent operational habit preferences through a chain of "pattern recognition - typical comparison - anomaly detection - weight adjustment - consistency assessment".

[0090] In some preferred embodiments, a specific example is given below. Suppose that on a woven bag production line, the system detects that operator A, in a specific batch of materials (scenario X), increases the system-recommended second defect detection threshold by 20% three times consecutively. Historical data shows that in this scenario, most operators typically only increase it by 5% to 10%, or leave it unchanged.

[0091] First, the system will perform pattern recognition on operator A's adjustment behavior in scenario X, and obtain the sequence characteristics of the adoption indicator (continuously rejecting system suggestions and significantly increasing the threshold) and the distribution characteristics of the deviation (the increase is much greater than the average).

[0092] Next, the system will obtain the typical adjustment behavior patterns of other operators (e.g., operators B, C, and D) in scenario X. The typical adjustment behavior patterns show that they tend to slightly increase the threshold or keep it unchanged in this scenario.

[0093] Subsequently, the system compared the adjustment behavior pattern of operator A with these typical adjustment behavior patterns, and the comparison results showed that there was a significant deviation.

[0094] Therefore, the system classifies operator A's adjustment behavior pattern as an abnormal behavior pattern.

[0095] To avoid such abnormal behavior having too much impact on the calculation of operational habit preferences, the system applies a weight adjustment mechanism, for example, reducing the weight of operator A's three abnormal adjustment behaviors in the calculation of context-dependent operational habit preferences by 50%.

[0096] Ultimately, the system performs a consistency evaluation on the weighted adjustment behavior patterns to obtain more stable and accurate context-dependent operational habit preferences. For example, even if operator A exhibits a few abnormal behaviors, if their behavior in other contexts or over a longer period is consistent with the norm, the final calculation result can still reflect their overall and more reliable adjustment tendencies, thereby supporting the system in generating more reasonable and effective suggestions for defect detection threshold ranges.

[0097] Reference Figure 2 The specific embodiments of this application also disclose a defect detection system for woven bags based on image pattern recognition, including: Image data acquisition module 1 is used to acquire image data of woven bags and reflective property data of woven bag materials; Image quality assessment module 2 is used to analyze woven bag image data to evaluate the image's illumination distribution, detail sharpness, and overall contrast. It also combines reflection characteristic data to identify artifacts caused by the superposition of material property changes and optical system degradation, thereby obtaining image quality indicators that characterize the imaging quality of woven bag image data. The defect feature weight adjustment module 3 is used to adjust the relative importance of different defect features in defect determination according to the image quality index, and calculate the defect score for the potential defect area of ​​the woven bag image data based on the adjusted relative importance. The threshold suggestion module 4 is used to receive the adjustment intention of the first defect detection threshold, and predict the impact of the adjustment intention on the missed detection of minor defects based on the image quality index and historical defect data, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold. The detection and early warning module 5 is used to perform a first defect judgment on the potential defect area based on the defect score and the obtained first defect detection threshold. At the same time, it performs a second defect judgment on the potential defect area based on the defect score and the second defect detection threshold. When the cumulative number of potential defects identified by the second defect judgment and determined as non-defects by the first defect judgment reaches a preset warning value, an early warning message is generated.

[0098] The system proposed in this application aims to provide a more intelligent and robust solution for detecting defects in woven bags. By integrating multiple functional modules, it achieves accurate identification and early warning of defects in woven bags.

[0099] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A method for detecting defects in woven bags based on image pattern recognition, characterized in that, include: Acquire image data of woven bags and data on the reflective properties of the woven bag material; The woven bag image data is analyzed to evaluate the image's illumination distribution, detail sharpness, and overall contrast. Combined with the reflection characteristic data, artifacts caused by the superposition of material property changes and optical system degradation are identified in the image, thereby obtaining image quality indicators that characterize the imaging quality of the woven bag image data. Based on the image quality index, the relative importance of different defect features in defect determination is adjusted, and based on the adjusted relative importance, the defect score is calculated for the potential defect area in the woven bag image data. Receive an adjustment intention regarding a first defect detection threshold, and based on the image quality index and historical defect data, predict the impact of the adjustment intention on the missed detection of minor defects, and generate a suggested defect detection threshold range prompt including a second defect detection threshold; Based on the defect score and the obtained first defect detection threshold, a first defect determination is performed on the potential defect area. At the same time, based on the defect score and the second defect detection threshold, a second defect determination is performed on the potential defect area. When the cumulative number of potential defects identified by the second defect determination that are determined to be non-defects by the first defect determination reaches a preset warning value, an early warning message is generated.

2. The method for detecting defects in woven bags according to claim 1, characterized in that, Receive an adjustment intention regarding a first defect detection threshold, and based on the image quality index and historical defect data, predict the impact of the adjustment intention on the missed detection of minor defects, and generate a suggested defect detection threshold range prompt including a second defect detection threshold, including: While acquiring the woven bag image data, the micro-pattern image of the micro-pattern reference target is continuously acquired. The micro-pattern reference target is set at a fixed position in the field of view of the industrial camera and is independently illuminated. The sharpness analysis of the micro-pattern image is performed to obtain the micro-feature resolution index. Based on the historical defect data, a statistical analysis of recently identified defects is obtained, and based on the statistical analysis results, an initial prediction of the impact of the adjustment intention on the missed detection of minor defects is obtained. When the micro-feature resolution index meets the preset resolution decline condition, a correction factor is applied to the initial prediction result to increase the predicted false negative risk, and a second defect detection threshold is determined based on the increased false negative risk, and a suggested defect detection threshold range prompt containing the second defect detection threshold is generated; otherwise, a second defect detection threshold is determined based on the initial prediction result, and a suggested defect detection threshold range prompt containing the second defect detection threshold is generated.

3. The method for detecting defects in woven bags according to claim 1, characterized in that, Generate early warning information, including: The potential defect information identified by the secondary defect determination and determined as non-defect by the primary defect determination is denoted as the ignored potential defect information. The cumulative number of ignored potential defect information, defect severity, and operator historical response behavior are obtained, wherein the defect severity is determined by the defect score and / or the area of ​​the defect region, and the historical response behavior includes the operator's response time and / or confirmation operation record to historical warning information; Based on the cumulative quantity, the severity of the defect, and the historical response behavior, the initial warning intensity and presentation method of the warning information are determined, and the warning information is output when the cumulative quantity reaches a preset warning value. After the warning information is output, if no effective response to the warning information is received within a preset time period and the cumulative number continues to increase, the warning intensity of the warning information is increased and / or the presentation method of the warning information is changed. Based on the response time statistics in the historical response behavior, the initial warning intensity of subsequent warning information is updated.

4. The method for detecting defects in woven bags according to claim 1, characterized in that, Based on the image quality index, the relative importance of different defect features in defect determination is adjusted, and based on the adjusted relative importance, defect scores are calculated for potential defect areas in the woven bag image data, including: Identify at least two image degradation factors that coexist in the image quality index to obtain a multiple degradation state; For each defect feature, the reliability attenuation coefficient corresponding to each defect feature is calculated based on the multiple degradation states and the preset degradation interaction rules. Based on the reliability attenuation coefficient, a feature adjustment factor is obtained through nonlinear mapping, and the relative importance of each defect feature in defect determination is nonlinearly adjusted using the feature adjustment factor. Based on the relative importance after nonlinear adjustment, the defect score is calculated for the potential defect areas in the woven bag image data.

5. The method for detecting defects in woven bags according to claim 1, characterized in that, Receive an adjustment intention regarding a first defect detection threshold, and based on the image quality index and historical defect data, predict the impact of the adjustment intention on the missed detection of minor defects, and generate a suggested defect detection threshold range prompt including a second defect detection threshold, including: Record the operator's adjustment behavior, including the adjustment value, adjustment frequency, and adoption of system suggestions; Based on the aforementioned adjustment behavior, calculate the operator's trust index and operational habit preference for the suggested defect detection threshold range; By combining the image quality index and the historical defect data, the impact of the adjustment intention on the missed detection of minor defects is predicted to obtain the initial missed detection risk, and an initial suggested defect detection threshold range is generated based on the initial missed detection risk. Based on the trust index and the operating habit preference, the preliminary suggested defect detection threshold range is adjusted, and the second defect detection threshold is determined from the adjusted suggested defect detection threshold range to generate a suggested defect detection threshold range prompt. The recommended defect detection threshold range is displayed using an optimized presentation method.

6. The method for detecting defects in woven bags according to claim 5, characterized in that, The suggested defect detection threshold range is displayed using an optimized presentation method, including: The operator calculates the historical response time of at least two different presentation methods and selects the presentation method with the shorter response time as the initial presentation method for subsequent suggested defect detection threshold range prompts based on the historical response time. If no effective response is made to the suggested defect detection threshold range prompt within the preset suggested time and / or the second defect detection threshold is not adopted, and the adjustment intention continues to occur, the presentation intensity of the suggested defect detection threshold range prompt is increased and / or switched to a higher intensity presentation mode.

7. The method for detecting defects in woven bags according to claim 5, characterized in that, Based on the aforementioned adjustment behavior, calculate the operator's trust index and operational habit preferences regarding the suggested defect detection threshold range, including: Continuously monitor the stability of the data stream reflecting operator adjustments; When the amount of data in the adjustment behavior data stream is insufficient and / or anomalies are detected, the existing reliable adjustment behavior data is slightly perturbed to generate virtual operator adjustment behavior data to expand the adjustment behavior dataset. The adjustment behavior data is filtered in real time to identify occasional large adjustment behaviors, and these occasional large adjustment behaviors are marked as low-weight data to determine the recent stable adjustment behavior data. When calculating the trust index, recent stable adjustment behavior data is used, and a time decay factor is introduced into the recent stable adjustment behavior data. Based on the recent stable adjustment behavior data, the adoption mark and / or deviation of each adjustment behavior to the suggested defect detection threshold range are determined. The adoption mark and / or deviation are weighted and normalized according to the time decay factor to obtain the trust index. When calculating the operational habit preferences, pattern recognition is performed on the operator's adjustment behavior in a specific production context, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences.

8. The method for detecting defects in woven bags according to claim 7, characterized in that, The adoption identifier and / or deviation are weighted, summarized, and normalized according to the time decay factor to obtain the trust index, including: Continuously monitor the operating status of the production line, including the material batch of woven bags, equipment health status, and / or environmental parameters; When the operating state is detected to meet the preset change conditions, historical adjustment behavior data corresponding to the current production situation is identified. A difference analysis is performed on the historical adjustment behavior data and the recent stable adjustment behavior data to determine whether the adjustment behavior pattern corresponding to the recent stable adjustment behavior data is related to the change in operating status. The difference analysis includes at least comparing the statistical characteristics of the adoption identifier and / or deviation degree corresponding to the historical adjustment behavior data and the recent stable adjustment behavior data to obtain the difference analysis results. When the results of the difference analysis indicate that the adjustment behavior pattern is related to changes in the operating status, a contextual adaptation weight is introduced into the recently stable adjustment behavior data in the trust index calculation. The adoption identifier and / or deviation are jointly weighted according to the time decay factor and the context adaptability weight, and then weighted, summarized and normalized to obtain the context adaptability trust index.

9. The method for detecting defects in woven bags according to claim 7, characterized in that, Pattern recognition is performed on operators' adjustment behaviors in specific production contexts, and the consistency of the identified adjustment behavior patterns is evaluated to obtain context-dependent operational habit preferences, including: Pattern recognition is performed on the operator's adjustment behavior in a specific production context to obtain the adjustment behavior pattern, wherein the adjustment behavior pattern includes the sequence characteristics of adoption identifiers and / or the distribution characteristics of deviation. Obtain typical adjustment behavior patterns of other operators in specific production situations; The difference between the adjusted behavior pattern and the typical adjusted behavior pattern is compared to determine whether there is a significant deviation, and the difference comparison result is obtained. When the difference comparison results show a significant deviation, the adjustment behavior pattern is determined to be an abnormal behavior pattern. A weight adjustment mechanism is applied to the abnormal behavior patterns to reduce their influence weight in the calculation of context-dependent operational habit preferences. Consistency assessment of the adjusted behavior patterns after weighting was performed to obtain context-dependent operational habit preferences.

10. A defect detection system for woven bags based on image pattern recognition, characterized in that, include: The image data acquisition module is used to acquire image data of woven bags and reflective property data of woven bag materials; The image quality assessment module is used to analyze the woven bag image data to evaluate the image's illumination distribution, detail sharpness, and overall contrast. It also combines the reflection characteristic data to identify artifacts in the image caused by the superposition of material property changes and optical system degradation, thereby obtaining image quality indicators that characterize the imaging quality of the woven bag image data. The defect feature weight adjustment module is used to adjust the relative importance of different defect features in defect determination according to the image quality index, and calculate the defect score for potential defect areas in the woven bag image data based on the adjusted relative importance. The threshold suggestion module is used to receive the adjustment intention of the first defect detection threshold, and predict the impact of the adjustment intention on the missed detection of minor defects based on the image quality index and historical defect data, and generate a suggested defect detection threshold range prompt containing the second defect detection threshold. The detection and early warning module is used to perform a first defect determination on the potential defect area based on the defect score and the obtained first defect detection threshold, and at the same time, perform a second defect determination on the potential defect area based on the defect score and the second defect detection threshold. When the cumulative number of potential defects identified by the second defect determination that are determined to be non-defects by the first defect determination reaches a preset warning value, an early warning message is generated.

Images