Explosion-proof membrane detection method and system based on machine vision

Abstract

The invention relates to the technical field of explosion-proof membrane detection, and discloses an explosion-proof membrane detection method and system based on machine vision, and the method comprises the steps: obtaining data including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation, actual installation deviation value and the like through an acquisition system, carrying out smooth preprocessing, and then carrying out surface scratch detection; feature extraction is carried out, a result is decomposed into feature groups, a feature matrix is constructed according to the feature groups, and abnormal features deviating from normal are marked; analyzing a correlation influence value between scratch distribution and intensity fluctuation; if the value exceeds a preset correlation threshold value, carrying out region segmentation on the scratch data to identify tiny scratch features, extracting local stress concentration features in intensity fluctuation, and adjusting feature weights according to the local stress concentration features; and in combination with the new weight set and the actual installation deviation value, calculating a quality evaluation score of the micro region, and comparing the quality evaluation score with a preset threshold value to judge a final quality detection result of the explosion-proof membrane. According to the invention, key index data are fused, and comprehensive automatic evaluation of the quality of the explosion-proof membrane is realized.

Description

Technical Field

[0001] This invention relates to the field of explosion-proof film testing technology, and in particular to an explosion-proof film testing method and system based on machine vision. Background Technology

[0002] In the field of modern industrial manufacturing and safety, explosion-proof film is a key protective material widely used in construction, automobiles and industrial equipment. Its quality is directly related to personal safety and equipment stability. The quality inspection of explosion-proof film involves multiple aspects such as surface defects, optical performance, mechanical strength and installation quality, and the quality of each aspect is inextricably linked.

[0003] In one existing technology, when it is necessary to identify surface defect features, traditional algorithms such as edge detection and threshold segmentation are used to locate and extract the defect features of the explosion-proof film. These algorithms can automatically learn and identify complex surface defect features, offering high detection accuracy and adaptability. However, when it is necessary to identify optical performance, optical inspection equipment is used. In other words, the inspection at each stage is conducted independently and measured separately. This mainly results in a lack of a scientific and comprehensive scoring system, leading to fragmented inspection results that make it difficult to form an intuitive judgment on the overall performance of the explosion-proof film.

[0004] Therefore, in order to solve the problem of weak correlation between various testing links, how to construct a comprehensive scoring system based on multi-link testing data, and how to integrate indicators from various dimensions through scientific methods to ultimately achieve a comprehensive evaluation of the quality of explosion-proof films have become key issues that urgently need to be addressed. Summary of the Invention

[0005] This invention provides a machine vision-based method and system for detecting explosion-proof films, which solves the problem of independent testing of various components in current explosion-proof film testing technology and enables a comprehensive evaluation of the quality of explosion-proof films.

[0006] In a first aspect, to solve the above-mentioned technical problems, the present invention provides a machine vision-based method for detecting explosion-proof films, comprising: An initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation value, is obtained through a pre-established explosion-proof film detection and acquisition system. The initial multidimensional data set is processed to obtain a smoothed data set; Feature extraction is performed on the smoothed dataset to obtain a decomposed feature data set; a feature matrix is ​​constructed based on the decomposed feature data set, and abnormal feature annotation is performed on feature values ​​that deviate from the normal range to obtain an annotated feature matrix. Feature weights are assigned based on the labeled feature matrix, and a preliminary analysis is conducted on the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence values ​​of features in each dimension. When the correlation influence value is determined to exceed the preset correlation threshold, the surface scratch distribution data is segmented into regions to obtain micro-scratch features, and the local stress concentration features are extracted from the edge tensile strength fluctuation data. The feature weights of each dimension are adjusted according to the micro-scratch characteristics and the local stress concentration characteristics to obtain an adjusted feature weight set. The quality assessment score of the micro-area is then calculated by combining the adjusted feature weight set with the actual installation deviation value. The quality test result of the explosion-proof film is determined based on the quality assessment score and the preset score threshold.

[0007] Preferably, the step of processing the initial multi-dimensional data set to obtain a smoothed data set includes: The initial multi-dimensional data set is processed to unify its format, resulting in a standard data set. Based on the standard dataset, missing values ​​are interpolated to obtain the complete dataset after filling in the missing values. The outliers in the complete dataset are cleaned to obtain a cleaned, standardized dataset. Based on the specified data set, a smoothing process is performed using a sliding window average to obtain a smoothed initial data set. The smoothed initial data set is normalized to obtain a unified data group; Numerical data transformation is performed on the unified data set to obtain the final smoothed data set.

[0008] Preferably, the step of extracting features from the smoothed data set to obtain the decomposed feature data set includes: Defect feature points are separated from the surface scratch distribution data in the smoothed dataset to obtain the separated scratch feature groups. When the feature points in the scratch feature group exceed the preset feature threshold range, the scratch feature group is labeled to obtain the labeled classification data group. The edge tensile strength fluctuation data in the classified data group are parameterized to obtain the decomposed feature data group.

[0009] Preferably, the step of constructing a feature matrix based on the decomposed feature data set, and annotating feature values ​​that deviate from the normal range to obtain an annotated feature matrix includes: The decomposed feature data groups are aligned to obtain an aligned data set. Based on the aligned dataset, feature integration is performed to obtain an integrated feature group; The integrated feature groups are used to construct a matrix to obtain the constructed matrix data; When any feature value in the matrix data exceeds a preset feature range, the data corresponding to that feature value is marked to obtain a labeled feature matrix.

[0010] Preferably, the step of assigning feature weights based on the labeled feature matrix and conducting a preliminary analysis of the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence values ​​of each dimension of features includes: Based on the labeled feature matrix, surface scratch distribution and edge tensile strength fluctuation data are extracted and integrated to obtain a distribution dataset; The importance value of each feature is obtained by comparing the distributed dataset with the feature weights in the historical database. When the importance value exceeds a preset standard value threshold, the data corresponding to the importance value will be marked as key association points to obtain the compared association dataset; Based on the aforementioned associated dataset, the scratch distribution and edge tensile strength fluctuation data are processed to obtain the integrated associated impact value.

[0011] Preferably, when the correlation influence value is determined to exceed a preset correlation threshold, the surface scratch distribution data is segmented into regions to obtain micro-scratch features, and local stress concentration features are extracted from the edge tensile strength fluctuation data, including: When the correlation influence value exceeds the preset correlation threshold, the surface scratch distribution data is segmented to obtain the segmented scratch area dataset. The micro-scratches are extracted from the scratch region dataset. Local stress concentration features are obtained by extracting edge intensity fluctuation data; When the correlation value between the micro-scratch feature and the local stress concentration feature is higher than a preset correlation value threshold, the micro-scratch feature and the local stress concentration feature are integrated to obtain an abnormal feature group.

[0012] Preferably, the step of adjusting the feature weights of each dimension according to the micro-scratch features and the local stress concentration features to obtain the adjusted feature weight set includes: Obtain the local stress concentration features and stress distribution features in the abnormal feature group; Based on the comparison of the density and distribution range of the micro-scratches with the abnormal feature groups, the corresponding scratch density deviation dataset is obtained; The feature weight ratios of the scratch density deviation dataset, local stress concentration features, and stress distribution features are adjusted to obtain the adjusted feature weight set.

[0013] Preferably, the quality assessment score of the micro-area is calculated by combining the adjusted feature weight set and the actual installation deviation value, including: Based on the adjusted feature weight set and the actual installation deviation value, a preliminary calculation of the regional quality performance is performed to obtain a preliminary quality performance dataset. The preliminary quality performance dataset is subjected to data fusion processing to obtain a quality assessment score.

[0014] Preferably, determining the quality test result of the explosion-proof film based on the quality assessment score and a preset score threshold includes: When the quality assessment score is determined to be lower than a preset score threshold, the explosion-proof film is deemed to be of substandard quality. When the quality assessment score is determined to be higher than or equal to a preset score threshold, the explosion-proof film is deemed to be of qualified quality.

[0015] Secondly, the present invention provides a machine vision-based explosion-proof film detection system, comprising: The data acquisition module is used to acquire an initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values, through a pre-established explosion-proof film detection and acquisition system. The data processing module is used to process the initial multi-dimensional data set to obtain a smoothed data set; The feature extraction module is used to extract features based on the smoothed data set to obtain the decomposed feature data set; The feature annotation module is used to construct a feature matrix based on the decomposed feature data group, and to annotate the feature values ​​that deviate from the normal range to obtain the annotated feature matrix. The preliminary analysis module is used to assign feature weights based on the labeled feature matrix and to perform a preliminary analysis on the correlation between surface scratch distribution and edge tensile strength fluctuation, so as to obtain the correlation influence value of each dimension feature. The data extraction module is used to perform region segmentation on the surface scratch distribution data to obtain micro-scratch features when the correlation influence value is determined to exceed the preset correlation threshold, and to extract local stress concentration features from the edge tensile strength fluctuation data. The score evaluation module is used to adjust the feature weights of each dimension according to the micro-scratch characteristics and the local stress concentration characteristics to obtain an adjusted feature weight set, and to calculate the quality evaluation score of the micro-area by combining the adjusted feature weight set and the actual installation deviation value. The test result module is used to determine the quality test result of the explosion-proof film based on the quality assessment score and the preset score threshold.

[0016] Thirdly, the present invention also provides an electronic device, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the machine vision-based explosion-proof film detection method described in any one of the above.

[0017] Fourthly, the present invention also provides a computer-readable storage medium comprising a stored computer program, wherein, when the computer program is executed, it controls the device containing the computer-readable storage medium to perform the machine vision-based explosion-proof film detection method described above.

[0018] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention constructs a comprehensive scoring system based on multi-stage testing data by acquiring multi-dimensional data, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values. This makes the test results no longer fragmented, but integrates various performance indicators to form a comprehensive quality assessment. This invention can conduct a holistic and systematic evaluation of the quality of explosion-proof films, providing more valuable conclusions for quality control and decision-making.

[0019] (2) After processing the initial multi-dimensional dataset to obtain a smoothed dataset, this invention not only performs feature extraction and annotation, but also analyzes the correlation between surface scratch distribution and edge tensile strength fluctuation. When the correlation influence value exceeds a preset threshold, the surface scratch distribution data is further segmented into regions to obtain micro-scratch features, and local stress concentration features are extracted. These meticulous operations help to more accurately identify defects and potential risks in the explosion-proof film. Compared with traditional methods, this invention can more accurately detect quality problems in the explosion-proof film, especially some micro-scratches and stress concentrations that may be overlooked, thereby improving the overall detection accuracy and making it more conducive to ensuring the safety and reliability of the explosion-proof film.

[0020] (3) Considering that different features may have different degrees of influence on the quality of explosion-proof film under different circumstances, this invention analyzes the correlation influence value to obtain the correlation influence value of each dimension feature. When the correlation threshold is exceeded, the weights of each dimension feature are adjusted to obtain the adjusted feature weight set. This dynamic adjustment of feature weights makes the weight allocation more reasonable, can better reflect the importance of each feature in the actual quality assessment, avoids the deviation that may be caused by fixed weight allocation, improves the accuracy and reliability of the quality assessment results, and enables the assessment results to more realistically reflect the actual quality status of the explosion-proof film. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the explosion-proof film detection method based on machine vision provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of the explosion-proof film detection system based on machine vision provided in the second embodiment of the present invention. Detailed Implementation

[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] To address the aforementioned issues, the following specific embodiments will provide a detailed description and explanation of a machine vision-based explosion-proof film detection method provided in this application.

[0024] Reference Figure 1 The first embodiment of the present invention provides a machine vision-based method for detecting explosion-proof films, comprising the following steps: S11, an initial multi-dimensional data set including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation value is obtained through a pre-established explosion-proof film detection and acquisition system; S12, perform data processing on the initial multi-dimensional data set to obtain a smoothed data set; S13, Perform feature extraction based on the smoothed data set to obtain the decomposed feature data group; S14, construct a feature matrix based on the decomposed feature data group, and label the feature values ​​that deviate from the normal range as abnormal features to obtain the labeled feature matrix; S15, Based on the labeled feature matrix, feature weights are assigned, and a preliminary analysis is performed on the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence values ​​of each dimension feature. S16, when it is determined that the correlation influence value exceeds the preset correlation threshold, the surface scratch distribution data is divided into regions to obtain micro scratch features, and the local stress concentration features are extracted from the edge tensile strength fluctuation data. S17. Adjust the feature weights of each dimension according to the micro-scratch features and the local stress concentration features to obtain the adjusted feature weight set, and calculate the quality assessment score of the micro-area by combining the adjusted feature weight set and the actual installation deviation value. S18, determine the quality test result of the explosion-proof film based on the quality assessment score and the preset score threshold.

[0025] In step S11, the explosion-proof film detection and acquisition system needs to be established in advance to obtain the surface scratch distribution, optical transmittance deviation, edge tensile strength fluctuation data and actual installation deviation value of the explosion-proof film in a small area, and combine them to obtain an initial multi-dimensional data set to ensure the integrity of the data.

[0026] It should be noted that, to obtain the surface scratch distribution data, the surface scratches and their regional distribution attributes can be detected using a high-resolution imaging device to perform a micron-level scan of the explosion-proof film surface. The device has a resolution of 1000 pixels per square millimeter, capable of capturing minute scratches as small as 0.05 millimeters in width. During scanning, the system assigns unique location coordinates to each region, such as region A1 corresponding to coordinates (10, 20), and records the scratch length and depth data to obtain the surface scratch distribution data.

[0027] Specifically, for testing the optical transmittance deviation, a spectral analyzer can be used to detect the target area. The normal transmittance standard is 90%, and the preset deviation threshold is ±2%. When the transmittance of a certain area is only 87%, it is below the threshold, and the system will automatically mark it as an abnormal area; when the transmittance of a certain area is 91%, it meets the normal transmittance standard. This method quantifies optical performance to ensure that the explosion-proof film will not affect the visual effect or safety performance due to light transmission problems during use. The preset deviation threshold is set at ±2%, which is based on historical data statistics and industry standards.

[0028] Furthermore, the detection of abnormal edge tensile strength fluctuation data marking areas can be performed using tensile testing equipment to analyze the strength fluctuations in the edge region. If the standard tensile strength threshold is 50 Newtons, and the test value of a certain area is only 40 Newtons, it is recorded as a potential risk point; if the test value of a certain area is 55 Newtons, the edge tensile strength of the current area meets the requirements and no marking is needed. This detection method can identify insufficient edge strength in advance, avoiding the risk of detachment or breakage due to insufficient tension after installation.

[0029] Next, the actual installation deviation value can be analyzed using image processing to determine the flatness of the bonding area. A high-precision camera is used to capture images of the bonding area, and a deviation distribution map is generated. If a flatness deviation of 0.3 mm is found in a certain area, exceeding the standard value of 0.1 mm, it is marked and recorded as the installation deviation value. This analysis helps identify potential air bubbles or wrinkles during installation, ensuring the bonding effect of the explosion-proof film.

[0030] In another feasible approach, the above testing process can be further optimized by adjusting equipment parameters or threshold standards to accommodate different types of explosion-proof membranes. For example, for thinner membrane materials, the tensile strength threshold can be appropriately lowered to 45 Newtons to avoid misjudgment.

[0031] In step S12, the initial multi-dimensional data set is processed to obtain a smoothed data set, including: standardizing the format of the initial multi-dimensional data set to obtain a standard data set; interpolating missing values ​​based on the standard data set to obtain a complete data set with imputed values; cleaning outliers in the complete data set to obtain a cleaned, normalized data set; smoothing the normalized data set using a sliding window average to obtain a smoothed initial data set; normalizing the smoothed initial data set to obtain a unified data group; and converting the unified data group into numerical data to obtain the final smoothed data set.

[0032] It should be noted that, based on the source attributes of the initial multi-dimensional data set, the original records from different sources are transformed to achieve format unification, resulting in a uniform initial data set. Data sources include production equipment records, installation site feedback, and laboratory test results. These data vary in format, such as text, tables, or image annotations. Data transformation is used to integrate heterogeneous data into standard structured tables. For example, all records are unified into a table format containing fields such as "Detection Area," "Detection Item," and "Numerical Result." Suppose that in the data records of a certain production batch, surface scratch data is mainly described in text, while transmittance data is stored as a percentage. Through transformation, key information from the text descriptions is extracted into numerical values, and format unification processing is performed to obtain a uniform standard data set. This method ensures data consistency in subsequent processes.

[0033] It should be noted that linear interpolation is used for missing values ​​in the standard dataset. Specifically, reasonable extrapolation is made based on the value range of adjacent data points, and the missing values ​​are filled in according to the extrapolation results. In the light transmittance testing of explosion-proof films, data for a certain area is missing, while the light transmittance of adjacent areas is 88% and 92% respectively. Through linear interpolation, the light transmittance of the missing area can be extrapolated to be approximately 90%, thus obtaining a complete dataset. This method avoids the impact of missing data on subsequent analysis, especially in continuous testing projects, maintaining the smoothness and logic of the data.

[0034] Specifically, for the complete dataset, data cleaning involves comparing records and removing duplicates; for example, if multiple test records for the same area are used, only the latest result is retained. Simultaneously, field validation removes non-compliant content, including invalid text descriptions or incorrectly formatted values. If a record has an empty test area field or a value exceeding a reasonable range (e.g., transmittance displayed as 150%), it is directly deleted or marked as invalid, ultimately forming a cleaned and standardized dataset. This cleaning process ensures the accuracy and usability of the data.

[0035] Furthermore, a sliding window can be used to average the adjacent values ​​of each dimension's data points, resulting in a smoothed initial data set. When processing the transmittance data of explosion-proof film, assuming the transmittance records for a certain area are 85%, 87%, and 90%, taking the average of three adjacent data points through a sliding window can smooth the data to a value of approximately 87%. This method reduces abrupt changes in the data, providing a more stable foundation for subsequent analysis.

[0036] Subsequently, normalization processing of the smoothed initial dataset adjusts the data of each dimension to a preset range, scaling all values ​​to the range of 0 to 1. The tensile strength data of the explosion-proof film ranges from 40 Newtons to 60 Newtons. Through normalization, 40 Newtons can be mapped to 0, 60 Newtons to 1, and intermediate values ​​such as 50 Newtons to 0.5.

[0037] Finally, in the data type conversion stage, text-based data can be converted to numerical data to ensure consistency. In the explosion-proof film inspection record, if a field is recorded as "high intensity," it can be converted to the value 5 using preset rules; if the field content is "unknown," it is marked as abnormal and deleted, resulting in the final smoothed data set. This conversion and marking mechanism helps ensure the uniformity of data format, laying the foundation for subsequent verification.

[0038] In step S13, feature extraction is performed based on the smoothed data set to obtain a decomposed feature data group, including: separating defect feature points from the surface scratch distribution data in the smoothed data set to obtain a separated scratch feature group; when the feature points in the scratch feature group exceed a preset feature threshold range, feature annotation is performed on the scratch feature group to obtain an annotated classification data group; and parameter splitting is performed on the edge tensile strength fluctuation data in the classification data group to obtain a decomposed feature data group.

[0039] It should be noted that in the data analysis process for explosion-proof film testing, based on the processing of smoothed datasets, key variables are first extracted using principal component analysis. For surface scratch distribution data, the system automatically applies algorithms to reduce the dimensionality of scratch density and depth. Assuming that the scratch density data of a certain batch of film material is 3.2 scratches per square centimeter and the average depth is 0.1 mm, the algorithm projects these multidimensional data into the principal component space, extracts the top two principal components with a contribution rate of 85% as key variables, and separates the scratch feature groups.

[0040] Specifically, for the extracted scratch feature groups, the system uses a clustering algorithm to classify and label the defects, dividing the scratches into three categories: minor, moderate, and severe.

[0041] Scratches with a depth of less than 0.05 mm are labeled as minor, while those with a depth of more than 0.15 mm are labeled as severe, resulting in a categorized data set. Specifically, minor scratches are categorized by depth threshold: 0-0.05 mm; moderate scratches by depth threshold: 0.05-0.15 mm; and severe scratches by depth threshold: more than 0.15 mm.

[0042] Finally, for the edge tensile strength fluctuation data, the system decomposes it into multiple sub-indicators. When the tensile strength of a certain membrane material is 55.3 Newtons, the algorithm automatically splits it into two sub-indicators: transverse strength of 30.2 Newtons and longitudinal strength of 25.1 Newtons. The fluctuation rate is further analyzed, with a transverse fluctuation rate of 2.4% and a longitudinal fluctuation rate of 1.9%, generating the decomposed feature data set.

[0043] In step S14, constructing a feature matrix based on the decomposed feature data group and annotating feature values ​​that deviate from the normal range to obtain an annotated feature matrix includes: aligning the decomposed feature data group to obtain an aligned data set; integrating features based on the aligned data set to obtain an integrated feature group; constructing a matrix from the integrated feature group to obtain constructed matrix data; and marking the data corresponding to the feature value when any feature value in the matrix data exceeds a preset feature range to obtain an annotated feature matrix.

[0044] It's important to note that timeline synchronization is performed on the decomposed feature data sets to ensure consistency in timestamps across different dimensions. During the inspection process, surface scratch data and edge tensile strength data are collected from different devices. When the timestamp for the scratch data is recorded as 10:00:05, while the timestamp for the strength data is recorded as 10:00:07, resulting in a 2-second discrepancy, data alignment is used to adjust these data to the same time reference, for example, aligning them all to 10:00:05, resulting in an aligned dataset. This synchronization method ensures the matching of multi-dimensional data across time dimensions, laying the foundation for subsequent integration.

[0045] Specifically, the synchronized dataset undergoes feature integration, unifying and transforming multi-dimensional features to obtain an integrated feature set. For example, if the scratch depth feature value of the explosion-proof film is 0.18 mm, while the preset scratch depth threshold is 0.15 mm, this feature value is marked as a feature to be processed because it exceeds the threshold. Similarly, if the optical transmittance deviation feature value is 3.2%, and the threshold is 3.0%, it will also be marked as a feature to be processed. This marking method helps to quickly identify feature data that requires further attention, ensuring that the integrated feature set meets structured requirements and provides a clear data framework for subsequent analysis.

[0046] Furthermore, the integrated feature set matrix is ​​converted into a structured feature matrix. Assuming the feature set contains three dimensions—scratch depth, transmittance deviation, and peak tensile strength—if any batch of transmittance deviation data is missing, the matrix construction process will indicate incomplete data and request the addition or removal of relevant data, ultimately generating a complete matrix. This approach ensures the comprehensiveness of the matrix data, providing a reliable foundation for subsequent anomaly detection.

[0047] Finally, for the constructed matrix data, anomaly annotation is used to determine whether feature values ​​exceed preset ranges. When the normal range for scratch depth is 0.05 to 0.15 mm, and a batch of membranes has a scratch depth of 0.20 mm, it is marked as an anomalous feature because it exceeds the range. Similarly, if the range for peak tensile strength is set to 45 to 55 Newtons, and a batch of membranes has an edge tensile strength of 60 Newtons, it will also be marked as an anomalous feature. This annotation method produces an annotated feature matrix. The resulting feature matrix clearly reflects the location and type of anomalous data, facilitating subsequent targeted processing and improving detection accuracy.

[0048] In step S15, the process of allocating feature weights based on the labeled feature matrix and conducting a preliminary analysis of the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence value of each dimension feature includes: extracting and integrating surface scratch distribution and edge tensile strength fluctuation data based on the labeled feature matrix to obtain a distribution dataset; comparing the distribution dataset with the feature weights in the historical database to obtain the importance value of each feature; when the importance value exceeds a preset standard value threshold, marking the data corresponding to the importance value as key correlation points to obtain a compared correlation dataset; and organizing the scratch distribution and edge tensile strength fluctuation data based on the correlation dataset to obtain the integrated correlation influence value.

[0049] It's important to note that the key to extracting surface scratch and edge intensity distribution data from the labeled feature matrix lies in ensuring consistency between the distribution characteristics and tensile fluctuations. In one inspection, the surface scratch distribution data showed that the scratch depth was concentrated at 0.12 mm in certain areas, while the edge intensity distribution exhibited significant fluctuations in intensity values ​​at specific locations. Through data extraction, these distribution characteristics can be compared with the tensile fluctuation data, revealing an overlap between the areas of concentrated scratches and areas of significant tensile fluctuations. These overlaps can then be extracted and integrated to obtain the distribution dataset.

[0050] Specifically, based on historical data and by applying weighted assignments, the feature weights in the distributed dataset are compared, paying attention to the correlation between surface scratches and edge strength. Historical data shows that for every 0.05 mm increase in surface scratch depth, edge strength decreases by an average of 2 Newtons; this correlation is quantified as feature weights. By comparing the scratch depth and strength values ​​in the current dataset, the importance value of the scratch feature is 0.75, while that of the strength feature is 0.65. When the preset importance threshold is 0.6, both meet the criteria for labeling, and these two features require focused attention in subsequent analysis. This weighting method highlights key features and improves the targeting of the analysis.

[0051] Furthermore, to more thoroughly compare the surface scratch distribution and tensile fluctuation intensity distribution, a correlation analysis between the two is necessary. If the scratch distribution shows that the scratch depth of a batch of membranes fluctuates between 0.08 and 0.14 mm, while the tensile fluctuation intensity ranges between 40 and 50 Newtons, and the preset fluctuation range threshold is an intensity fluctuation not exceeding 8 Newtons, then the current batch exceeds the threshold and is marked as a key correlation point. This correlated dataset is then compiled and saved. This marking method helps to quickly locate abnormal areas that may affect membrane performance, providing a reference for quality control.

[0052] Subsequently, the scratch depth and tensile fluctuation intensity distribution in the correlated datasets were organized and analyzed through data integration to obtain the integrated correlation influence value. The method for obtaining the correlation influence value specifically includes the following steps: First, the surface scratch distribution data and edge tensile strength fluctuation data are standardized and preprocessed to eliminate dimensional differences. The surface scratch distribution data is converted into a scratch density vector per square centimeter, and the edge tensile strength fluctuation data is converted into an intensity fluctuation vector in Newton units. Correlation influence value. The Pearson correlation coefficient method is used to calculate the specific formula as follows:

[0053] in, Indicates the first Surface scratch density values ​​for each region Indicates the first The edge tensile strength fluctuation value of each region This represents the average surface scratch density across all regions. This represents the average value of the fluctuation in tensile strength at the edges of all regions.

[0054] During the calculation, scratch density and intensity fluctuation values ​​are simultaneously acquired for each small area (e.g., 1 cm²), forming data pairs. For example, a sample of explosion-proof film is divided into 9 areas, with a scratch density vector of [2.1, 1.8, 3.2, 2.5, 2.9, 1.6, 3.1, 2.4, 2.7] scratches / cm² and an intensity fluctuation vector of [48.5, 50.2, 45.8, 47.3, 46.1, 51.2, 45.3, 47.9, 46.7] Newtons. The correlation influence value calculated by the formula is approximately -0.72 (ranging from -1 to 1), and the absolute value is greater than the preset correlation threshold of 0.7, thus triggering the in-depth analysis in step S16. To eliminate random errors, a sliding window method (window size 3×3 area unit) is used to smooth the data, that is, the moving average of the data within the window is calculated and then the Pearson formula is reapplied.

[0055] It should be noted that the construction of the historical database and the definition and acquisition of feature weights specifically include the following steps. First, the historical database is formed through long-term accumulated explosion-proof film testing data, including multi-dimensional data such as optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation, and actual installation deviation values. Each sample in the database contains complete quality inspection result labels, i.e., qualified and unqualified labels, for subsequent weight analysis.

[0056] The definition of feature weights is based on the training results of the machine learning model on historical data. Specifically, a random forest algorithm is used, with the quality inspection results of the explosion-proof film as the target variable, and optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation, and actual installation deviation value as feature variables for model training. After training, the weight value of each feature is defined by the feature importance score output by the model, which reflects the degree of contribution of each feature to the quality judgment.

[0057] The specific method for calculating the importance values ​​of each feature is as follows: Based on the trained random forest model, the average reduction in impurity of each feature across all decision trees is calculated using the following formula:

[0058] in Representation of features The importance value ranges from [0,1], with a larger value indicating a greater contribution of the feature to the quality assessment; This represents the total number of decision trees in a random forest, for example... =100, set based on the size of the historical database; Representation of features In the The reduction in impurity resulting from splitting a decision tree is typically calculated based on Gini impurity or information gain, such as the formula for Gini impurity reduction: - After the importance values ​​are calculated, they are converted into weights through normalization. First, the importance values ​​of all features are summed, and then the individual importance value of each feature is divided by the sum to ensure that the total weight is 1.

[0059] For example, based on the Gini impurity formula, if the total reduction in impurity of the characteristic optical transmittance deviation in 100 trees is 8.5, then its importance value is 8.5 / 100=0.085, which is converted to a weight of 0.2 after normalization. If the sum of the importance values ​​of all features is 0.4, then its weight is 0.085 / 0.4=0.2125, which is rounded to 0.21.

[0060] The historical database is updated quarterly, merging newly added explosion-proof film test data (both qualified and unqualified samples) with the existing data to retrain the random forest model. Before the update, the new data undergoes consistency verification, such as aligning the format and units with historical data to ensure data quality. After the weights are recalculated, if the feature weights change by more than 10%, such as the optical transmittance deviation weight changing from 0.2 to 0.18, the version change is recorded, and the new weights are applied synchronously in the testing process.

[0061] In step S16, when it is determined that the correlation influence value exceeds a preset correlation threshold, the surface scratch distribution data is segmented to obtain micro-scratch features, and local stress concentration features are extracted from the edge tensile strength fluctuation data. This includes: when the correlation influence value exceeds the preset correlation threshold, the surface scratch distribution data is segmented to obtain a segmented scratch region dataset; micro-scratch features are extracted from the scratch region dataset; local stress concentration features are extracted from the edge strength fluctuation data; when the correlation value between the micro-scratch features and the local stress concentration features is higher than a preset correlation value threshold, the micro-scratch features and the local stress concentration features are integrated to obtain an abnormal feature group.

[0062] If, after processing, the correlation value between scratch depth and tensile fluctuation is 0.82, which is higher than the preset business target requirement of 0.7, then it is marked. When the correlation value exceeds the preset correlation threshold, the surface scratch distribution data needs to be segmented to obtain a segmented scratch area dataset.

[0063] It should be noted that in one inspection, the surface scratch distribution of a batch of explosion-proof films showed that some areas had a high scratch density, with more than 5 scratches per square centimeter, while other areas were sparser, with only 1-2 scratches. By segmenting the surface scratch distribution data, high-density and low-density areas could be separated, resulting in independent scratch region datasets. Further checks were conducted to ensure that the segmentation of these regions met the requirements, such as the clarity of the boundaries of high-density areas, to ensure the accuracy of subsequent analysis. This segmentation method helps simplify the complex scratch distribution into analyzable units, laying the foundation for subsequent feature extraction.

[0064] Specifically, feature extraction is used to extract the distribution features of minute scratches from the segmented scratch region dataset, with a focus on the completeness of anomaly feature extraction. Within high-density scratch regions, there are some minute scratches, barely perceptible to the naked eye, with a length of only 0.05 mm. Feature extraction allows for the complete recording of the distribution location and quantity of these minute scratches, yielding minute scratch features. This refined processing highlights hidden quality issues, providing detailed data support for subsequent comparisons.

[0065] Furthermore, by comparing the refined scratch feature set with the edge intensity fluctuation data, local stress concentration features are extracted. In high-density scratch areas, when the edge intensity fluctuation amplitude increases from the normal value of 3 Newtons to 6 Newtons, the trend of the intensity fluctuation amplitude changes significantly with the scratch, exhibiting stress concentration characteristics, which are then output as local stress concentration features. This comparison method helps identify the potential link between scratch distribution and stress anomalies.

[0066] Finally, the scratch feature set and stress distribution feature set are organized and calculated through data integration to obtain the correlation value. When the correlation value is higher than the preset correlation threshold, such as a correlation value of 0.85 and a correlation value threshold of 0.7, the two sets of features need to be fused to obtain the integrated abnormal feature set.

[0067] The correlation value calculation uses the cosine similarity formula. First, the data in the scratch feature set of the small region is converted into a numerical vector S; simultaneously, the data in the stress feature set of the same region is converted into a vector T of the same dimension. The specific calculation process includes three steps: First, the sum of the corresponding products of the values ​​in each dimension of vectors S and T is calculated, for example, the scratch density value multiplied by the stress peak value, plus the depth value multiplied by the stress gradient, etc., forming the numerator value; Second, the square root of the sum of the squares of all values ​​in S and T is calculated, and the square root of the sum of the squares of all values ​​in T is multiplied to obtain the denominator value; finally, the correlation value is equal to the numerator value divided by the denominator value, and the value ranges from -1 to +1.

[0068] Furthermore, when the calculation result for a specific small area, such as 0.85, exceeds a preset threshold of 0.7, the feature fusion mechanism is triggered. The scratch feature set and the stress distribution feature set are then fused using a weighted linear method to obtain an integrated abnormal feature set.

[0069] For example, firstly, the density values ​​of the scratch feature set are standardized using a min-max method. Assuming the density range for the entire sample is 0-5 scratches / mm², the standardized value is (2.1-0) / (5-0) = 0.42; the depth range is 0-0.05mm, so the standardized value is (0.03-0) / 0.05 = 0.60. Simultaneously, the peak and gradient values ​​of the stress distribution feature set are standardized. Assuming the peak value range is 0-30MPa, the standardized value is 24.3 / 30 = 0.81. If the gradient range is 0-3 MPa / mm, the standardized value is 1.8 / 3 = 0.60. Then, according to preset weighting coefficients (scratch weight α = 0.6, stress weight β = 0.4), a fusion calculation is performed: density-peak value = 0.6 × 0.4² + 0.4 × 0.8¹ = 0.576, depth-gradient value = 0.6 × 0.60 + 0.4 × 0.60 = 0.600. Finally, the abnormal feature group for this small region is generated as [0.576, 0.600]. This integration method can transform features from various dimensions into a unified analytical basis, providing an intuitive reference for quality control and significantly improving detection efficiency and problem localization capabilities.

[0070] In step S17, the feature weights of each dimension are adjusted according to the micro-scratch features and the local stress concentration features to obtain an adjusted feature weight set. The quality assessment score of the micro-area is then calculated by combining the adjusted feature weight set with the actual installation deviation value. This includes: acquiring the local stress concentration features and stress distribution features from the abnormal feature group; comparing the micro-scratch density and scratch distribution range according to the abnormal feature group to obtain the corresponding scratch density deviation dataset; adjusting the feature weight ratios of the scratch density deviation dataset, the local stress concentration features, and the stress distribution features to obtain an adjusted feature weight set; performing a preliminary calculation of the regional quality performance based on the adjusted feature weight set and the actual installation deviation value to obtain a preliminary quality performance dataset; and performing data fusion processing on the preliminary quality performance dataset to obtain a quality assessment score.

[0071] Specifically, local stress concentration features and stress distribution features are obtained from a set of abnormal data features, including initial optical transmittance deviation, actual installation deviation value, surface scratch distribution after processing, and edge tensile strength fluctuation data.

[0072] It should be noted that the density of minute scratches is matched with a preset threshold range using data comparison to obtain a corresponding scratch density deviation dataset. In one test, the scratch distribution of a batch of explosion-proof films showed that some areas had 6 scratches per square centimeter, while other areas had only 2, indicating poor uniformity. A scratch density deviation dataset was generated through comparison, with the deviation value controlled within 20%. This comparison method helps identify unevenly distributed areas.

[0073] Specifically, the feature weight ratio adjustment method includes the following steps: First, the scratch density deviation dataset, local stress concentration features, and stress distribution features are standardized and preprocessed to map each feature value to the [0,1] interval. The min-max standardization method is used, with the minimum and maximum values ​​obtained from historical data of the most recent month in the historical database. Then, weight coefficients are assigned based on the relative importance of each feature in the abnormal feature group, following the principles of weight allocation: First, scratch density deviation weighting The weighting is based on two factors: first, the uniformity of the scratch distribution, with higher weights for poorer uniformity; second, the weighting of local stress concentration characteristics. The third factor is the weighting of stress distribution characteristics, determined based on the degree of stress concentration. The weights are set based on the uniformity of distribution; simultaneously, the total weight constraint must be satisfied, with a total weight sum of 1. The specific weight adjustment formula is as follows:

[0074] in, , , These are the baseline weights for each feature. For example, when uneven scratch density distribution is detected in a certain area, such as 6 scratches / cm² in some areas and 2 scratches / cm² in others, and the coefficient of variation (i.e., standard deviation divided by the mean) reaches 0.7, the scratch density deviation weight is... Adjusted to 0.5; when the local stress concentration characteristics show a peak value exceeding the threshold by 30%, the stress concentration weight is adjusted. Adjusted to 0.3; stress distribution weight The corresponding value was adjusted to 0.2. Finally, the adjusted feature weight set was obtained through weighted fusion, which was used for subsequent quality assessment score calculation.

[0075] Furthermore, based on the feature weight set, data mapping is applied, and the actual installation deviation value and deviation range, along with optical transmittance deviation data, are used to calculate the quality assessment score for the micro-area. Specifically, the optical transmittance deviation data is first normalized to a uniform dimension. For example, the normalized value is calculated using a formula: the absolute value of the optical transmittance deviation value is divided by a preset deviation threshold, where the preset deviation threshold is set to ±2%, and the normalized value ranges from 0 to 1.

[0076] The preset deviation threshold is set based on the historical database in step S15. For example, by analyzing the light transmittance of 1000 sets of qualified explosion-proof film samples from historical data, the distribution range of light transmittance deviation is calculated to be the mean ± 2 times the standard deviation, ensuring that the threshold covers more than 95% of normal samples, thus using ±2% as the critical point. Similarly, this threshold is the same as the preset deviation threshold in step S11.

[0077] Then, the normalized optical transmittance deviation value is incorporated into the feature weight set, with its weight set to 0.2 based on the historical database (coordinated with the weights of surface scratches, intensity fluctuations, etc., totaling 1). This weighted value is then fused with the actual installation deviation value (weight 0.3) and other feature weights. Finally, a quality assessment score is calculated using a linear weighted formula, whereby the quality assessment score equals the sum of the normalized values ​​of each feature multiplied by their corresponding weights. These features include optical transmittance deviation, actual installation deviation, scratch density deviation, and local stress concentration characteristics. When the optical transmittance deviation exceeds the threshold, the normalized value is set to 1.0, and the deviation is treated as a "serious anomaly."

[0078] For example, if the optical transmittance deviation of a certain area is -3%, exceeding the threshold, the normalized value is 1.0; the actual installation deviation, after normalization, is 0.8; and the comprehensive score for other features (such as scratch density deviation and local stress concentration features) is 0.7. The weight allocation is: optical transmittance deviation 0.2, actual installation deviation 0.3, and other features 0.5. Therefore, the quality assessment score = (1.0 × 0.2) + (0.8 × 0.3) + (0.7 × 0.5) = 0.79, with a maximum score of 1.0.

[0079] It should be noted that the weights are set based on historical database statistics. For example, historical database analysis shows that in the past 1000 samples, the probability of optical transmittance deviation causing quality problems is 20%, actual installation deviation is 30%, surface scratches are 30%, and intensity fluctuation is 20%. Therefore, the initial weight set is {optical transmittance deviation: 0.2, actual installation deviation: 0.3, scratch distribution: 0.3, intensity fluctuation: 0.2}. This weight set is a baseline value and will be dynamically adjusted in actual testing based on real-time data, such as the correlation influence value in step S16, but the sum will remain at 1 to ensure the adaptability of the evaluation.

[0080] This comprehensive processing can integrate multi-dimensional data to form intuitive quality assessment results, which helps to quickly locate problem areas and improve detection efficiency and accuracy.

[0081] In step S18, determining the quality test result of the explosion-proof film based on the quality assessment score and the preset score threshold includes: when the quality assessment score is determined to be lower than the preset score threshold, the explosion-proof film is determined to be unqualified; when the quality assessment score is determined to be higher than or equal to the preset score threshold, the explosion-proof film is determined to be qualified.

[0082] It should be noted that the final quality assessment score is the basis for determining whether the explosion-proof film meets the product quality standards. For example, the score threshold can be set at 80 points. If the product's quality assessment score is below 80 points, the explosion-proof film is considered substandard; if the quality assessment score is 80 points or higher, the explosion-proof film is considered qualified. However, if the optical transmittance deviation exceeds the safe range independently, such as an absolute value greater than 5%, even if the quality assessment score meets the standards, it must be marked as a risk product and re-inspected to ensure reliability. In summary, this invention discloses a machine vision-based method for detecting explosion-proof films. The method includes acquiring an initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values, through a pre-established explosion-proof film detection and acquisition system; processing the initial multi-dimensional data set to obtain a smoothed data set; extracting features from the smoothed data set to obtain decomposed feature data groups; constructing a feature matrix based on the decomposed feature data groups; annotating feature values ​​deviating from the normal range to obtain an annotated feature matrix; and allocating feature weights based on the annotated feature matrix. A preliminary analysis is conducted on the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence values ​​of each dimension feature. When the correlation influence value is determined to exceed a preset correlation threshold, the surface scratch distribution data is segmented into regions to obtain micro-scratch features, and local stress concentration features are extracted from the edge tensile strength fluctuation data. Based on the micro-scratch features and the local stress concentration features, the feature weights of each dimension are adjusted to obtain an adjusted feature weight set. The quality assessment score of the micro-region is calculated by combining the adjusted feature weight set with the actual installation deviation value. The quality inspection result of the explosion-proof film is determined based on the quality assessment score and a preset score threshold.

[0083] The method comprehensively evaluates the quality of the explosion-proof film through multi-stage testing, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values. This avoids fragmented test results and integrates various indicators of the explosion-proof film, ultimately achieving a comprehensive and scientific assessment of the film's quality.

[0084] Reference Figure 2 The present invention provides a machine vision-based explosion-proof film detection system, comprising: The data acquisition module is used to acquire an initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values, through a pre-established explosion-proof film detection and acquisition system. The data processing module is used to process the initial multi-dimensional data set to obtain a smoothed data set; The feature extraction module is used to extract features based on the smoothed data set to obtain the decomposed feature data set; The feature annotation module is used to construct a feature matrix based on the decomposed feature data group, and to annotate the feature values ​​that deviate from the normal range to obtain the annotated feature matrix. The preliminary analysis module is used to assign feature weights based on the labeled feature matrix and to perform a preliminary analysis on the correlation between surface scratch distribution and edge tensile strength fluctuation, so as to obtain the correlation influence value of each dimension feature. The data extraction module is used to perform region segmentation on the surface scratch distribution data to obtain micro-scratch features when the correlation influence value is determined to exceed the preset correlation threshold, and to extract local stress concentration features from the edge tensile strength fluctuation data. The score evaluation module is used to adjust the feature weights of each dimension according to the micro-scratch characteristics and the local stress concentration characteristics to obtain an adjusted feature weight set, and to calculate the quality evaluation score of the micro-area by combining the adjusted feature weight set and the actual installation deviation value. The test result module is used to determine the quality test result of the explosion-proof film based on the quality assessment score and the preset score threshold.

[0085] The explosion-proof film detection system based on machine vision provided in this embodiment of the invention is used to execute all the process steps of the explosion-proof film detection method based on machine vision in the above embodiment. The working principle and beneficial effects of the two are one-to-one, so they will not be described again.

[0086] This invention also provides an electronic device. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor, such as a data acquisition module. When the processor executes the computer program, it implements the steps in the various machine vision-based explosion-proof film detection method embodiments described above, for example... Figure 1 The step S11 shown. Alternatively, when the processor executes the computer program, it implements the functions of each module / unit in the above-described device embodiments, such as the data acquisition module.

[0087] For example, the computer program may be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the electronic device.

[0088] The electronic device may be a desktop computer, laptop, handheld computer, or smart tablet, etc. The electronic device may include, but is not limited to, a processor and memory. Those skilled in the art will understand that the above components are merely examples of electronic devices and do not constitute a limitation on the electronic device. It may include more or fewer components than described above, or combine certain components, or different components. For example, the electronic device may also include input / output devices, network access devices, buses, etc.

[0089] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the electronic device, connecting all parts of the electronic device via various interfaces and lines.

[0090] The memory can be used to store the computer programs and / or modules. The processor implements various functions of the electronic device by running or executing the computer programs and / or modules stored in the memory and by calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0091] Wherein, if the modules / units integrated in the electronic device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0092] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0093] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A machine vision-based method for detecting explosion-proof films, characterized in that, include: An initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation value, is obtained through a pre-established explosion-proof film detection and acquisition system. The initial multidimensional data set is processed to obtain a smoothed data set; Feature extraction is performed on the smoothed data set to obtain the decomposed feature data set; A feature matrix is ​​constructed based on the decomposed feature data group, and abnormal feature annotations are performed on feature values ​​that deviate from the normal range to obtain the labeled feature matrix. Feature weights are assigned based on the labeled feature matrix, and a preliminary analysis is conducted on the correlation between surface scratch distribution and edge tensile strength fluctuation to obtain the correlation influence values ​​of features in each dimension. When the correlation influence value is determined to exceed the preset correlation threshold, the surface scratch distribution data is segmented into regions to obtain micro-scratch features, and the local stress concentration features are extracted from the edge tensile strength fluctuation data. The feature weights of each dimension are adjusted according to the micro-scratch characteristics and the local stress concentration characteristics to obtain an adjusted feature weight set. The quality assessment score of the micro-area is then calculated by combining the adjusted feature weight set with the actual installation deviation value. The quality test result of the explosion-proof film is determined based on the quality assessment score and the preset score threshold.

2. The explosion-proof film detection method based on machine vision according to claim 1, characterized in that, The step of processing the initial multi-dimensional data set to obtain a smoothed data set includes: The initial multi-dimensional data set is processed to unify its format, resulting in a standard data set. Based on the standard dataset, missing values ​​are interpolated to obtain the complete dataset after filling in the missing values. The outliers in the complete dataset are cleaned to obtain a cleaned, standardized dataset. Based on the specified data set, a smoothing process is performed using a sliding window average to obtain a smoothed initial data set. The smoothed initial data set is normalized to obtain a unified data group; Numerical data transformation is performed on the unified data set to obtain the final smoothed data set.

3. The explosion-proof film detection method based on machine vision according to claim 1, characterized in that, The step of extracting features from the smoothed data set to obtain the decomposed feature data set includes: Defect feature points are separated from the surface scratch distribution data in the smoothed dataset to obtain the separated scratch feature groups. When the feature points in the scratch feature group exceed the preset feature threshold range, the scratch feature group is labeled to obtain the labeled classification data group. The edge tensile strength fluctuation data in the classified data group are parameterized to obtain the decomposed feature data group.

4. The explosion-proof film detection method based on machine vision according to claim 1, characterized in that, The step of constructing a feature matrix based on the decomposed feature data set, and annotating feature values ​​that deviate from the normal range to obtain an annotated feature matrix includes: The decomposed feature data groups are aligned to obtain an aligned data set. Based on the aligned dataset, feature integration is performed to obtain an integrated feature group; The integrated feature groups are used to construct a matrix to obtain the constructed matrix data; When any feature value in the matrix data exceeds the preset feature range, the data corresponding to the feature value is marked to obtain a labeled feature matrix.

5. The explosion-proof film detection method based on machine vision according to claim 1, characterized in that, The process involves assigning feature weights based on the labeled feature matrix and conducting a preliminary analysis of the correlation between surface scratch distribution and edge tensile strength fluctuations to obtain the correlation influence values ​​of features in each dimension, including: Based on the labeled feature matrix, surface scratch distribution and edge tensile strength fluctuation data are extracted and integrated to obtain a distribution dataset; The importance value of each feature is obtained by comparing the distributed dataset with the feature weights in the historical database. When the importance value exceeds a preset standard value threshold, the data corresponding to the importance value will be marked as key association points to obtain the compared association dataset; Based on the aforementioned associated dataset, the scratch distribution and edge tensile strength fluctuation data are processed to obtain the integrated associated impact value.

6. The machine vision-based explosion-proof film detection method according to claim 1, characterized in that, When the correlation influence value is determined to exceed a preset correlation threshold, the surface scratch distribution data is segmented to obtain micro-scratch features, and local stress concentration features are extracted from the edge tensile strength fluctuation data, including: When the correlation influence value exceeds the preset correlation threshold, the surface scratch distribution data is segmented to obtain the segmented scratch area dataset. The micro-scratches are extracted from the scratch region dataset. Local stress concentration features are obtained by extracting edge intensity fluctuation data; When the correlation value between the micro-scratch feature and the local stress concentration feature is higher than a preset correlation value threshold, the micro-scratch feature and the local stress concentration feature are integrated to obtain an abnormal feature group.

7. The explosion-proof film detection method based on machine vision according to claim 1, characterized in that, The adjusted feature weight set obtained by adjusting the feature weights of each dimension according to the micro-scratch features and the local stress concentration features includes: Obtain the local stress concentration features and stress distribution features in the abnormal feature group; Based on the comparison of the density and distribution range of the micro-scratches with the abnormal feature groups, the corresponding scratch density deviation dataset is obtained; The feature weight ratios of the scratch density deviation dataset, local stress concentration features, and stress distribution features are adjusted to obtain the adjusted feature weight set.

8. The explosion-proof film detection method based on machine vision according to claim 7, characterized in that, The quality assessment score for the micro-area is calculated by combining the adjusted feature weight set and the actual installation deviation value, including: Based on the adjusted feature weight set and the actual installation deviation value, a preliminary calculation of the regional quality performance is performed to obtain a preliminary quality performance dataset. The preliminary quality performance dataset is subjected to data fusion processing to obtain a quality assessment score.

9. The machine vision-based explosion-proof film detection method according to claim 1, characterized in that, The step of determining the quality test result of the explosion-proof film based on the quality assessment score and a preset score threshold includes: When the quality assessment score is determined to be lower than a preset score threshold, the explosion-proof film is deemed to be substandard. When the quality assessment score is determined to be higher than or equal to a preset score threshold, the explosion-proof film is deemed to be of qualified quality.

10. A machine vision-based explosion-proof film inspection system, characterized in that, include: The data acquisition module is used to acquire an initial multi-dimensional data set, including optical transmittance deviation, surface scratch distribution, edge tensile strength fluctuation data, and actual installation deviation values, through a pre-established explosion-proof film detection and acquisition system. The data processing module is used to process the initial multi-dimensional data set to obtain a smoothed data set; The feature extraction module is used to extract features based on the smoothed data set to obtain the decomposed feature data set; The feature annotation module is used to construct a feature matrix based on the decomposed feature data group, and to annotate the feature values ​​that deviate from the normal range to obtain the annotated feature matrix. The preliminary analysis module is used to assign feature weights based on the labeled feature matrix and to perform a preliminary analysis on the correlation between surface scratch distribution and edge tensile strength fluctuation, so as to obtain the correlation influence value of each dimension feature. The data extraction module is used to perform region segmentation on the surface scratch distribution data to obtain micro-scratch features when the correlation influence value is determined to exceed the preset correlation threshold, and to extract local stress concentration features from the edge tensile strength fluctuation data. The score evaluation module is used to adjust the feature weights of each dimension according to the micro-scratch characteristics and the local stress concentration characteristics to obtain an adjusted feature weight set, and to calculate the quality evaluation score of the micro-area by combining the adjusted feature weight set and the actual installation deviation value. The test result module is used to determine the quality test result of the explosion-proof film based on the quality assessment score and the preset score threshold.

Images