CN-122023817-A - Paper tube breakage defect detection method based on shape characteristics

Abstract

The invention discloses a paper tube breakage defect detection method based on shape characteristics, which relates to the technical field of image recognition analysis and comprises the steps of receiving and identifying a multiple-disc case through a breakage monitoring database; according to the invention, the position of the current defect can be indicated by constructing a paper tube breakage point state wave diagram through a compound disc history case, the internal decision feature of the model is accurately associated with an external real result through constructing a detection decision-real state mapping relation table, whether the problem is caused by poor data quality or distortion of the model feature can be identified through analyzing double space mapping consistency, quantitative monitoring and early warning of system performance attenuation are realized through identifying the feature and predicting threshold failure time, the fragile condition of a system decision boundary is clarified through evaluating the influence of environment and batch, the newly discovered defect mode is brought into a high risk mode library, and thus the spiral rise of detection accuracy and stability is realized, and intelligent autonomous operation is realized.

Inventors

• WEI XIUJIN

Assignees

• 浙江郡琳新材料科技有限公司

Dates

Publication Date

20260512

Application Date

20260121

Claims (8)

1. 1. The paper tube breakage defect detection method based on the shape characteristics is characterized by comprising the following steps of: The method comprises the steps that firstly, a multiple-disc case is received and marked through a damage monitoring database, wherein the multiple-disc case comprises a missing detection defect case, a false alarm defect case and an edge detection case with the confidence coefficient lower than a preset threshold value, and paper tube shape characteristic extraction is carried out to obtain a target sample detection set; Acquiring original mode data in the paper tube detection process based on a target sample detection set, acquiring shape characteristics of each monitoring point of the paper tube, establishing a multi-disc approval point on the paper tube, analyzing paper tube breakage points according to the multi-disc approval point, and constructing a paper tube breakage point state wave diagram to form an evolution sequence of the paper tube changing along with time; step three, acquiring detection decision features in the detection process, including real-time classification confidence coefficient of each monitoring point during detection and weight distribution in the feature extraction process, performing space-time alignment and association analysis on the decision features and the paper tube breakage point state wave diagrams, and constructing a detection decision-real state mapping relation table; Analyzing the mapping consistency of the defect area in the original data space and the feature vector space based on a detection decision-real state mapping relation table, performing quality assessment on the image data, identifying the change value of the feature characterization and the failure time set by the threshold value, assessing the influence of the environment and the image data quality, and generating a shape feature judgment result to generate a shape feature judgment result; And fifthly, optimizing the execution flow of the damage detection according to the shape characteristic judgment result, generating a system capacity defect diagnosis report, adding a complex case with a similar characteristic mode into a high risk mode library, and adjusting the early warning triggering condition of the real-time detection system.
2. 2. The method for detecting paper tube breakage defect based on shape characteristics according to claim 1, wherein the method for extracting the shape characteristics of the paper tube to obtain a target sample detection set comprises the following steps: S100, automatically acquiring a case to be multiplexed marked by an online detection system from a damage monitoring database, and automatically associating and packaging original data triplets of the case to be multiplexed, wherein the triplets comprise high-resolution image/point cloud data, all levels of feature images and confidence degrees output by a detection model, and corresponding production time stamps and environment parameters; S101, aiming at different types of complex cases, differential feature extraction emphasis is adopted, for missed detection cases, three-dimensional point cloud density distribution features are extracted, for false positive cases, the disturbance shape features causing misjudgment are mainly analyzed, the essential difference of the disturbance shape features and real damage on multi-scale textures or three-dimensional depth continuity is extracted, for low-confidence edge cases, the parts of the disturbance shape features, which are similar to the perfect and typical damage features, are extracted, boundary feature vectors in a high-dimensional feature space are formed, and the extracted two-dimensional contour shape features and three-dimensional point cloud features are subjected to early fusion treatment; S102, carrying out structural storage on the extracted depth identification features and the joint shape description according to defect types, false alarm reasons, edge modes and the like to form a sample database, and marking damage and normal label types for each sample in the target sample detection set.
3. 3. The method for detecting the paper tube breakage defect based on the shape characteristics according to claim 1, wherein the method is characterized in that based on a target sample detection set, raw mode data in the paper tube detection process is acquired, the shape characteristics of each monitoring point of the paper tube are acquired, and a multi-disc calibration point is established on the paper tube, and the specific process is as follows: S200, extracting shape features of point cloud data and image data according to original mode data in the paper tube detection process; S201, uniformly selecting multiple disc calibration points in the length direction of a paper tube, wherein the multiple disc calibration points comprise end face points, center points and plane side face points at two ends of the paper tube, clustering point cloud data by using a clustering algorithm, clustering points with similar characteristics into one type, and then selecting the most representative point from each type as the multiple disc calibration point; s202, integrating the extracted shape characteristic data and the multi-disc datum point data into a unified data management set for structured storage.
4. 4. The method for detecting paper tube breakage defect based on shape characteristics according to claim 1, wherein the paper tube breakage point analysis is performed according to the disc duplication approval point, a paper tube breakage point state wave diagram is constructed, and an evolution sequence of paper tube change along with time is formed, wherein the specific process is as follows: S300, observing feature changes of damaged points and shape feature changes of associated multiple disc approval points along with the time, and calculating the distance between each damaged point and each multiple disc approval point at different time points based on a distance measurement method to serve as the spatial position of the multiple disc approval points; s301, a state fluctuation diagram is a two-dimensional space-time matrix diagram taking time as a horizontal axis, taking the space position of a multi-disc approval point as a vertical axis, and taking color or representing abnormality, wherein each wave band on the vertical axis represents the change of the health condition of the multi-disc approval point along with the time; S302, automatically extracting key time sequence characteristics from a state fluctuation diagram, analyzing the evolution rate of a damaged point, aligning the damaged point data detected at different time points with the associated multi-disc approval point data, arranging the data according to time sequence to form an evolution sequence which changes along with time, and marking the evolution rate of the damaged point.
5. 5. The method for detecting the paper tube breakage defect based on the shape characteristics according to claim 1, wherein the method comprises the steps of performing space-time alignment and association analysis on decision characteristics and a paper tube breakage point state wave diagram, and constructing a detection decision-real state mapping relation table, wherein the specific process is as follows: S400, acquiring internal response characteristics of the content of the detection model, which are set in the real-time detection process, including weight distribution in the characteristic extraction process, acquiring real-time classification confidence and decision time stamps, acquiring detection decision characteristics, and fusing the detection decision characteristics with space-time information of a paper tube breakage point state wave diagram; s401, carrying out space-time accurate alignment on the detection decision feature and the space-time information of the paper tube breakage point state wave diagram, synchronizing a time axis and matching space coordinates to generate a space-time alignment time sequence, and analyzing the coincidence degree between the detection decision feature and the paper tube breakage point to be used as the relevance between the decision feature and the breakage point; S402, according to the time sequence of time-space alignment and the relevance between the decision feature and the damaged point, the time sequence is stored as a relationship table which can be inquired and analyzed in a structuring way, and a mapping relationship is established.
6. 6. The method for detecting the paper tube breakage defect based on the shape characteristics according to claim 1 is characterized by analyzing the mapping consistency of a defect area in an original data space and a feature vector space based on a detection decision-real state mapping relation table, and performing quality evaluation on image data, and specifically comprises the following steps of. S500, calculating coincidence coefficients among different defect areas in an original data space, calculating similarity coefficients among feature vectors of the different defect areas in a feature vector space, and analyzing the correlation between the coincidence coefficients and the similarity coefficients to obtain feature consistency coefficients; s501, analyzing structural definition, local contrast and illumination uniformity based on original image data to obtain a definition evaluation coefficient of the image data; S502, performing image quality evaluation according to the feature consistency coefficient, the definition evaluation coefficient and the defect degree to obtain an image comprehensive evaluation value, and classifying the image into three grades of high quality, medium quality and low quality.
7. 7. The method for detecting paper tube breakage defect based on shape feature according to claim 1, wherein the method for detecting paper tube breakage defect based on shape feature is characterized by identifying a change value of feature characterization and a failure time set by a threshold value, evaluating influence of environment and image data quality, and generating a shape feature judgment result, and specifically comprises the following steps: S600, selecting an initial paper tube in a stable state in a time period, calculating the average value of each characteristic in the stage as a reference value, and calculating the difference value between each characteristic value and the reference value for each subsequent time point to obtain a characteristic representation change value; S601, continuously monitoring a characteristic change value according to an initial threshold value set by a detection model, and recording a time point when the characteristic change value exceeds the threshold value for the first time, wherein the time point is the failure time set by the threshold value; S602, grouping the data according to different levels of environmental parameters, drawing a distribution scatter diagram of features under different environmental parameters for each grouped environmental parameter, marking decision boundary points, analyzing the change among the groups by using a method to obtain the most sensitive environmental factor, and generating a shape feature judgment result, wherein the shape feature judgment result comprises an image comprehensive evaluation value, a feature characterization change value, a threshold value set failure time and the most sensitive environmental factor.
8. 8. The method for detecting the breakage defect of the paper tube based on the shape characteristics according to claim 1, wherein a system capacity defect report is generated, and the deviation between the current detection dead zone and the detection model is marked, and the method specifically comprises the following steps: s700, adding the characteristic mode and the associated decision path of the complex case into a high risk mode library for early warning enhancement in real-time detection; s701, starting an incremental learning task, and performing fine adjustment on a feature extractor or a classification model by using a multi-disc approval point and evolution sequence data thereof to optimize a decision boundary; S702, updating a dynamic threshold rule, taking environmental parameters and production batches as input, and adaptively adjusting tolerance and confidence threshold of shape feature judgment.

Description

Paper tube breakage defect detection method based on shape characteristics Technical Field The invention relates to the technical field of image recognition analysis, in particular to a paper tube breakage defect detection method based on shape characteristics. Background With the development of machine vision technology, an automatic detection method based on image processing appears, a paper tube is used as an important winding core material in the industries of textile, packaging, paper making and the like, the surface quality of the paper tube directly influences the winding flatness and the production efficiency of downstream products, Currently, detection of paper tube breakage in the industrial field mainly relies on manual visual inspection or machine vision methods based on simple rules, but these techniques have significant drawbacks: The prior art scheme is mainly divided into two types, namely, a traditional image processing-based method is adopted, paper tube outlines are extracted through edge detection and threshold segmentation, and then geometric features of circularity and rectangularity are calculated to be compared with a standard template, so that the method is simple in calculation, extremely sensitive to illumination change and background interference, difficult to accurately distinguish real breakage and interference, and high in false alarm rate; Secondly, based on a deep learning method, the end-to-end defect classification is directly carried out on paper tube images by using a convolutional neural network model, the method improves the identification capability to a certain extent, the performance is seriously dependent on a large number of balanced training samples with accurate labeling, the defect samples in actual production are rare and have changeable forms, when missed detection or false alarm occurs, the problem that the rapid positioning is a data problem, a characteristic problem or a model problem is difficult, a detection system is usually trained based on a static data set, parameters are fixed after deployment, and the method cannot adapt to dynamic factors of fluctuation of production line speed and variation of ambient illumination, so that the performance gradually drifts and declines in long-term operation; in view of the above technical drawbacks, a solution is now proposed. Disclosure of Invention The invention aims to enable a system to actively make up a capacity short board of the system by focusing on the multi-disc case row for targeted learning, and trace the root cause of misjudgment by decision-state mapping analysis, so that the detection system is changed into a detection system which can continuously learn from errors of the system and dynamically adjust an early warning strategy from a static tool needing manual repeated debugging. In order to achieve the purpose, the invention adopts the following technical scheme that the paper tube breakage defect detection method based on the shape characteristics comprises the following steps: The method comprises the steps that firstly, a multiple-disc case is received and marked through a damage monitoring database, wherein the multiple-disc case comprises a missing detection defect case, a false alarm defect case and an edge detection case with the confidence coefficient lower than a preset threshold value, and paper tube shape characteristic extraction is carried out to obtain a target sample detection set; Acquiring original mode data in the paper tube detection process based on a target sample detection set, acquiring shape characteristics of each monitoring point of the paper tube, establishing a multi-disc approval point on the paper tube, analyzing paper tube breakage points according to the multi-disc approval point, and constructing a paper tube breakage point state wave diagram to form an evolution sequence of the paper tube changing along with time; step three, acquiring detection decision features in the detection process, including real-time classification confidence coefficient of each monitoring point during detection and weight distribution in the feature extraction process, performing space-time alignment and association analysis on the decision features and the paper tube breakage point state wave diagrams, and constructing a detection decision-real state mapping relation table; Analyzing the mapping consistency of the defect area in the original data space and the feature vector space based on a detection decision-real state mapping relation table, performing quality assessment on the image data, identifying the change value of the feature characterization and the failure time set by the threshold value, assessing the influence of the environment and the image data quality, and generating a shape feature judgment result to generate a shape feature judgment result; And fifthly, optimizing the execution flow of the damage detection according to the shape characteristic judgment result, gener