A yoga mat surface defect automatic identification method based on intelligent sensors

Abstract

The present application relates to the technical field of intelligent sensing detection and optical scattering method surface defect intelligent identification, and discloses a yoga mat surface defect automatic identification method based on intelligent sensor.In the surface of the measured object, a two-dimensional coordinate and a discrete sampling array are established, and a unified installation posture is installed.Under the condition of fixed oblique incidence, the scattering intensity of each point is scanned according to a multi-angle sequence, and the integrity and range are verified.According to the fixed angle weight, an angle spectrum weight matrix is formed.The residual error is obtained by point-by-point difference with the defect-free reference matrix, and the mutation intensity is synthesized by row-by-row difference to construct a matrix.According to the threshold value generated by the reference mutation statistics, the candidate abnormality is generated.On the grid, the connected domain is extracted by four-neighbor connection, the area is removed if it is insufficient, and the center and area are calculated.The number, position and area of defects are outputted and visualized marked.

Description

Technical Field

[0001] This invention relates to the field of intelligent sensing and detection and optical scattering method for intelligent identification of surface defects, specifically an automatic identification method for yoga mat surface defects based on intelligent sensors. Background Technology

[0002] With the rise of emerging industries such as flexible materials and consumer electronics, the mass production and high-quality manufacturing of elastomer surfaces like yoga mats and fitness mats has become a core aspect of precision manufacturing. Since the final quality of these products directly impacts user safety and comfort, extremely high requirements are placed on the accuracy and low false negative rate of detecting defects such as minor surface bumps, scratches, bubbles, mold marks, and embedded particles. Currently, mainstream methods for detecting surface defects in yoga mats primarily employ machine vision-based two-dimensional grayscale imaging, color imaging, or laser three-dimensional profilometry. These methods typically use industrial cameras to capture images, combined with specific image processing algorithms (such as threshold segmentation, edge detection, feature extraction, template-based or deep learning methods) to identify abnormal areas on the surface.

[0003] However, existing image- or point cloud-based automatic detection technologies generally face several limitations. First, limited by surface reflection, uneven illumination, and the diversity of texture and color inherent in materials, traditional visual methods struggle to reliably distinguish normal surface textures from genuine microscopic defects, easily leading to false positives, false negatives, or inaccurate region segmentation. Second, while methods such as deep learning improve adaptability to some extent, they heavily rely on a large number of manually labeled samples and are sensitive to scene generalization and changes in data distribution, making maintenance and expansion difficult. Meanwhile, while point cloud measurement equipment has good resolution of elevation variations, it is not sensitive to the soft deformation and minute perturbation defects of flexible materials, and its high cost and system complexity make it difficult to apply on a large scale in automated production lines for flexible products. Furthermore, most existing methods cannot achieve accurate quantitative modeling of complex optical scattering responses, making it difficult to fully utilize the physical response relationship between the incident angle, scattering angle, and local surface geometry. Therefore, for non-significant defects such as fine lines and pits on the surface, existing vision, point cloud and traditional inspection equipment often exhibit low detection rates and accuracy. Especially in high production line speed and strict quality control scenarios, this can easily lead to problems such as defective products flowing out, increased pressure on manual re-inspection, and limited automation level of the production line.

[0004] Therefore, this case aims to propose an automatic identification method for yoga mat surface defects based on intelligent sensors. Taking the influence of surface micro-geometry on the scattering angle distribution as the starting point, it replaces the image intensity signal, which is easily affected by texture and color, with a measurable angular spectrum response. Furthermore, it performs benchmark alignment and residual amplification on isomorphic grids and isomorphic angle sets, thereby achieving stable identification of defects such as micro-bumps, fine lines, and pits without relying on massive samples and learning parameters. Summary of the Invention

[0005] This invention provides an automatic identification method for surface defects of yoga mats based on intelligent sensors, which helps to solve the problems mentioned in the background art.

[0006] This invention provides the following technical solution: an automatic identification method for surface defects of yoga mats based on intelligent sensors, comprising:

[0007] S1. Establish a two-dimensional coordinate frame on the upper surface of the object under test, set up a scattering acquisition array composed of discrete sampling points, and keep the installation posture of all sampling points consistent with the outer normal of the object under test.

[0008] S2. Irradiate the object under fixed oblique conditions, perform scattering intensity scanning on each sampling point according to the preset multi-angle sequence and record the intensity reading, and at the same time complete the data integrity check and reading range limitation.

[0009] S3. For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectrum value, and a response matrix covering the entire domain is formed.

[0010] S4. Obtain the baseline response matrix on the defect-free standard sample, and subtract the current response matrix from the baseline response matrix point by point to form the residual value and the residual matrix.

[0011] S5. Perform discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix.

[0012] S6. Based on the mutation intensity statistics of the reference benchmark, a fixed threshold is set, a set of candidate anomalies is generated on the mutation intensity matrix, and the preliminary defect judgment is completed.

[0013] S7. Extract connected components on the discrete grid according to the four-adjacency relationship, remove candidate regions with insufficient area according to the minimum effective area threshold, and complete the sequential numbering of the retained regions and calculate the geometric center coordinates and region area.

[0014] S8. Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and generate corresponding visual markers on the coordinate result map.

[0015] Optionally, establishing a two-dimensional coordinate frame on the upper surface of the object under test, setting up a scattering acquisition array composed of discrete sampling points, and ensuring that the installation orientation of all sampling points is consistent with the external normal of the object under test, specifically includes:

[0016] A two-dimensional rectangular coordinate frame is established on the upper surface of the yoga mat being tested. The origin is set at the lower left corner, the long side is used as the positive direction of the horizontal axis, and the short side is used as the positive direction of the vertical axis. The coordinate unit is millimeters.

[0017] The area to be measured is divided into equally spaced grids according to the set size resolution, with the horizontal and vertical spacing set to the same value;

[0018] An integrated optical scattering angle detector is fixedly installed at each grid point, and the normal direction of the detector is consistent with the outer normal of the object being measured.

[0019] The length parameter of the object under test is quantized at the micrometer level to generate an effective computational domain that is strictly aligned with the grid, and subsequent acquisition and coordinate output are restricted to this effective computational domain.

[0020] Optionally, the step of irradiating the object under fixed oblique illumination conditions, performing a scattering intensity scan at each sampling point according to a preset multi-angle sequence and recording the intensity readings, while simultaneously completing data integrity checks and limiting the reading range, specifically includes:

[0021] A parallel light source device is installed obliquely above the surface of the object being measured, with a fixed incident angle of 45 degrees. The incident direction is located in the plane formed by the surface normal and the horizontal axis, and the light spot covers all sampling points.

[0022] A discrete scattering angle sequence is set, with the scattering angle range being 20 to 80 degrees and the angle step size being 10 degrees, resulting in seven discrete scattering angles;

[0023] For each sampling point, the receiving direction is adjusted angle by angle according to the discrete scattering angle sequence, and the scattering intensity is read to form a multi-angle intensity sequence;

[0024] Check the integrity of the data collection. If there are any data items that were not successfully collected, terminate the process and provide a re-collection conclusion.

[0025] Obtain the detector's saturation upper limit and noise lower limit, and limit the original intensity reading to the effective range; mark the point of overexposure when the maximum reading reaches the saturation upper limit, and mark the point of underexposure when the maximum reading is not higher than the noise lower limit; determine that the calibration is invalid and terminate the detection when the noise lower limit is not less than the saturation upper limit.

[0026] Optionally, for each sampling point, the multi-angle scattering intensities are weighted and summarized according to a fixed angle weighting rule to obtain a scalar response of the angular spectral intensity, forming a response matrix covering the entire domain, specifically including:

[0027] For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectral power value;

[0028] Arrange the scalar responses of all sampling points in row and column order to form an angular spectral value matrix.

[0029] Optionally, obtaining the baseline response matrix on the defect-free standard sample and subtracting the current response matrix from the baseline response matrix point by point to form the residual value and the residual matrix specifically includes:

[0030] A standard yoga mat without physical defects was selected as the reference sample, and multi-angle scattering intensity was collected according to step S2.

[0031] The original intensity readings of the reference sample are subjected to range limitation and integrity verification. After confirming that the reference data is complete, the reference angular spectral strength matrix is ​​generated according to step S3.

[0032] The angular spectral power matrix of the current object under test is subtracted point by point from the reference angular spectral power matrix to generate residual values ​​and residual matrices.

[0033] Optionally, the step of performing discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix specifically includes:

[0034] Discrete differencing is performed on the residual matrix along the horizontal and vertical directions, with one-sided differencing used for the matrix boundaries and central differencing used for the matrix interior.

[0035] The horizontal and vertical difference results are combined to form a mutation intensity index, creating a mutation intensity matrix covering the entire domain.

[0036] Optionally, the mutation intensity statistics based on the reference benchmark are set with a fixed threshold, a set of candidate outliers is generated on the mutation intensity matrix, and a preliminary defect determination is completed, specifically including:

[0037] The spatial abrupt change intensity is calculated on the angular spectral strength matrix of the reference sample, and the average and maximum values ​​of the results are obtained.

[0038] An interpolation threshold for mutation detection is set by performing interpolation between the average and maximum values ​​using a fixed ratio of 0.5.

[0039] Sampling points above a fixed threshold are selected from the mutation intensity matrix of the object under test to form a set of candidate anomalies;

[0040] When the candidate anomaly set is empty, output a defect-free conclusion and end the process; when the candidate anomaly set is not empty, proceed to connected component processing.

[0041] Optionally, the step of extracting connected components on the discrete grid according to four-adjacency relationships, eliminating candidate regions with insufficient area according to the minimum effective area threshold, sequentially numbering the retained regions, and calculating the geometric center coordinates and region area specifically includes:

[0042] An undirected graph structure is constructed on a discrete grid using four-adjacency relationships, and the connected components of the candidate outlier set are extracted accordingly.

[0043] Calculate the number of sampling points and the corresponding physical area contained in each connected component;

[0044] Set the minimum effective area threshold to the sum of the areas of two grid cells, and remove connected components with an area smaller than this threshold.

[0045] The remaining connected components are renumbered sequentially, and the geometric center coordinates and area of ​​each connected component are calculated.

[0046] Optionally, the output includes the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and generates corresponding visual markers on the coordinated result map, specifically including:

[0047] Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect region;

[0048] On the coordinated result map, generate dot marks for the geometric center of each defect.

[0049] The present invention has the following beneficial effects:

[0050] 1. This invention first explicitly proposes establishing a strict two-dimensional coordinate system on the surface of the yoga mat being tested, precisely corresponding the actual physical dimensions with digital quantization, and employing a micrometer-level quantization resolution to effectively avoid the randomness and non-uniformity of the measurement point arrangement. Through uniform spacing, a regular row and column sampling array is formed, and the installation posture of each sampling point is strictly consistent with the normal to the outside of the mat. This not only ensures the reproducibility and data consistency of subsequent testing but also provides standardized input for algorithm processing.

[0051] 2. A multi-angle oblique illumination condition is introduced, and the scattering intensity is collected at each measurement point through angle scanning, using a fixed step size to cover all sampling points. The acquisition process is coupled with an automated data integrity check mechanism, with strict stop and prompt mechanisms for abnormal situations such as missing data, overexposure, and underexposure. This improves the reliability and robustness of data acquisition, while significantly enhancing the sensitivity and recognition capability for minute surface defects through the fusion of multi-angle information. Compared with traditional surface detection using a single light source angle or a single data channel, this scheme effectively eliminates the weakness of angle dependence, enabling the algorithm to exhibit higher robustness and resolution when facing complex textures and microscopic defects.

[0052] 3. The optical signals from each sampling point at multiple discrete angles are weighted and integrated to generate a unique scalar response, which is then used to construct an angular spectrum response matrix covering the entire detection area. This aggregation strategy effectively utilizes multi-dimensional optical information, enhancing the contrast between the defect signal and the background. This method overcomes the limitations of traditional "single threshold" or "single angle" methods for locating abnormal areas, extracting more essential surface physical characteristics through weighting. Simultaneously, matrix-based storage facilitates subsequent automated processing and batch comparison, effectively supporting rapid detection and back-end traceability in large-scale production scenarios.

[0053] 4. By collecting standard, defect-free sample data, a baseline response matrix is ​​formed, and point-by-point residual analysis is performed between the tested samples and the baseline. This method is essentially "personalized calibration + differential analysis," which can eliminate systematic errors such as differences between equipment and material batches, making the detection more targeted and accurate. The introduction of the residual matrix further highlights defect signals and effectively suppresses interference from non-defect factors such as the external environment and material texture. Compared to existing detection methods that rely solely on absolute signal thresholds, this method improves the ability to capture minute differences and local anomalies, achieving a balance between high sensitivity and high accuracy.

[0054] 5. This invention employs a discrete difference algorithm along the row and column directions to synthesize a mutation intensity index, forming a mutation intensity matrix. This method, through second-order analysis of the spatial changes in the response matrix, can effectively detect minute mutations and structural anomalies on the surface. Its advantage lies not only in locating point-like and line-like defects but also in distinguishing between local mutations and large-scale smooth changes, thus improving detection sensitivity and robustness.

[0055] 6. Based on the mutation intensity distribution of standard samples, an adaptive threshold method is used to screen candidate outliers. By statistically analyzing the average and maximum values, an interpolation method is used to set the judgment threshold, dynamically adapting to different material batches and process fluctuations. This mechanism effectively avoids the subjectivity of manual threshold setting, reduces missed detections and false judgments, and also adapts to fluctuations in material, lighting, and other conditions during production. Unlike conventional "human experience thresholds" or "one-size-fits-all" strategies, the data-driven threshold setting of this invention significantly improves the algorithm's generalization ability and detection consistency, making it feasible for practical industrial applications.

[0056] 7. By segmenting connected components using four-adjacency relationships on the set of outliers and setting a minimum effective area threshold to filter out noisy regions, this scheme can effectively identify both point and area defects, and eliminate false defects caused by noise and isolated points. The center coordinates and area of ​​each connected component are calculated, achieving precise quantification of defects. Compared to traditional schemes that only mark points, rely on manual judgment, or depend on subjective experience, this method improves the automation and objectivity of defect identification, provides data support for subsequent defect tracking, tracing, and repair, and promotes the full automation of the intelligent detection process.

[0057] 8. The final output of the solution includes the total number of defects, center coordinates, and area, and automatically generates visual markers such as dots on the result image. This improves the intuitiveness and usability of the inspection results, facilitating direct interpretation and subsequent processing by production operators, and also supports long-term data archiving and traceability. Unlike traditional "text description" or "manual illustration" methods, this method achieves quantitative, accurate, and batch defect reporting, improving the digitalization and intelligence level of surface quality control. Attached Figure Description

[0058] Figure 1 This is a schematic diagram of the process of the present invention.

[0059] Figure 2 This is a schematic diagram of the coordinate system of the present invention.

[0060] In the diagram: 1-origin, 2-short side, 3-Y-axis, 4-X-axis, 5-long side. Detailed Implementation

[0061] 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.

[0062] Example, refer to Figure 1 An automatic identification method for surface defects of yoga mats based on smart sensors, comprising:

[0063] S1. Establish a two-dimensional coordinate frame on the upper surface of the object under test, set up a scattering acquisition array composed of discrete sampling points, and keep the installation posture of all sampling points consistent with the outer normal of the object under test.

[0064] S2. Irradiate the object under fixed oblique conditions, perform scattering intensity scanning on each sampling point according to the preset multi-angle sequence and record the intensity reading, and at the same time complete the data integrity check and reading range limitation.

[0065] S3. For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectrum value, and a response matrix covering the entire domain is formed.

[0066] S4. Obtain the baseline response matrix on the defect-free standard sample, and subtract the current response matrix from the baseline response matrix point by point to form the residual value and the residual matrix.

[0067] S5. Perform discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix.

[0068] S6. Based on the mutation intensity statistics of the reference benchmark, a fixed threshold is set, a set of candidate anomalies is generated on the mutation intensity matrix, and the preliminary defect judgment is completed.

[0069] S7. Extract connected components on the discrete grid according to the four-adjacency relationship, remove candidate regions with insufficient area according to the minimum effective area threshold, and complete the sequential numbering of the retained regions and calculate the geometric center coordinates and region area.

[0070] S8. Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and generate corresponding visual markers on the coordinate result map.

[0071] By first establishing a standardized two-dimensional coordinate system and discrete sampling point array, every location in the detection area can be accurately mapped, resulting in highly repeatable data and improved detection comprehensiveness. Scattering scanning under multi-angle oblique illumination conditions, combined with point-by-point acquisition and real-time integrity verification, effectively solves the problems of local information loss and errors caused by factors such as angle and illumination. Furthermore, weighted fusion of angular spectrum responses further enhances signal representativeness and anti-interference capabilities. Using the reference response matrix of a defect-free standard sample enables personalized differential analysis, minimizing the influence of equipment, material differences, and external environmental factors. The calculation of the residual matrix and abrupt change intensity helps to quickly highlight anomalies from large, uniform areas. Adaptive threshold screening and the elimination of four-adjacent connected components effectively reduce false positives and random noise. Finally, through the visual marking of the number, location, and area of ​​defects, fully automated, batch, and high-precision surface defect detection is achieved. Compared with existing technologies, this method improves detection accuracy, consistency, and efficiency, reduces subjective human error and detection blind spots, and is suitable for batch, standardized production lines, strongly supporting intelligent control of surface quality.

[0072] Reference Figure 2 The step of establishing a two-dimensional coordinate frame on the upper surface of the object under test, setting up a scattering acquisition array composed of discrete sampling points, and ensuring that the installation orientation of all sampling points is consistent with the external normal of the object under test, specifically includes:

[0073] A two-dimensional rectangular coordinate frame is established on the upper surface of the yoga mat being tested. The origin is set at the lower left corner, the long side is used as the positive direction of the horizontal axis, and the short side is used as the positive direction of the vertical axis. The coordinate unit is millimeters.

[0074] The area to be measured is divided into equally spaced grids according to the set size resolution, with the horizontal and vertical spacing set to the same value;

[0075] An integrated optical scattering angle detector is fixedly installed at each grid point, and the normal direction of the detector is consistent with the outer normal of the object being measured.

[0076] The length parameter of the object under test is quantized at the micrometer level to generate an effective computational domain that is strictly aligned with the grid, and subsequent acquisition and coordinate output are restricted to this effective computational domain.

[0077] Further specific implementation steps include:

[0078] Establish a two-dimensional rectangular coordinate system on the surface of the yoga mat. Let the bottom left corner be the origin, and set... The positive direction of the axis is along the long side of the yoga mat, moving away from the origin. The positive axis is along the shorter side of the yoga mat away from the origin, and the longer side is longer than the shorter side. The coordinate unit is millimeters, and the entire yoga mat area is defined as... ;in, For yoga mats The actual length of the shaft; For yoga mats The actual length of the shaft;

[0079] Set the size quantization resolution to millimeters;

[0080] Along the entire area Axial spacing ,along Axial spacing Divided into individual columns and Each row, set column indexes. Set row index Then the first The coordinates of the sampling points are: , , ;in, , The grid is respectively in , Spacing in the direction; This represents the total number of columns; This represents the total number of rows. , The first line, number The coordinate values ​​corresponding to the column; For the first The two-dimensional coordinates of each sampling point; and satisfying , , ;

[0081] An integrated optical scattering angle detector is fixedly installed at each point, and its normal points are aligned with the outer normal of the yoga mat.

[0082] Quantify the length in micrometers: , ;in, Quantized in micrometers Integer value; Quantized in micrometers Integer value;

[0083] calculate and Greatest common divisor micrometer; among which, To find the greatest common divisor (GCD) function of two integers;

[0084] make, millimeters , ;

[0085] And set the length for strict alignment after quantization: millimeters millimeters; among which, , The effective length to be quantized and strictly aligned with the grid;

[0086] Strictly speaking, , ;

[0087] Set the effective computational domain used for sampling and coordinate output to .

[0088] By establishing a standard Cartesian coordinate system on the yoga mat surface and quantizing coordinates using physical boundaries and millimeters as units, the measured area was standardized and digitized, facilitating subsequent batch acquisition and data comparison. Furthermore, micron-level dimensional resolution was employed to divide the area into equally spaced sections and establish strict row and column indices, ensuring uniform distribution and traceability of the sampling points. All sampling points were equipped with sensors whose orientation was aligned with the surface normal, improving measurement accuracy and spatial uniformity. The grid spacing was standardized using the greatest common divisor, avoiding irregularities and data distortion that might result from manual sampling. This step also ensured strict alignment between the effective acquisition domain boundary and data output, guaranteeing that all analysis results came from valid and standardized data areas. Compared to existing methods of "experience-based sampling" and "manual acquisition," this invention significantly improves spatial digitization, sampling uniformity, quantification accuracy, and automation capabilities. This approach not only enhances the systematic nature and reproducibility of the detection, facilitating comparative analysis with historical data, but also provides high-quality input for data-driven intelligent algorithms.

[0089] The process involves irradiating the object under fixed oblique illumination conditions, performing a scattering intensity scan at each sampling point according to a preset multi-angle sequence, recording the intensity readings, and simultaneously performing data integrity checks and limiting the reading range. Specifically, this includes:

[0090] A parallel light source device is installed obliquely above the surface of the object being measured, with a fixed incident angle of 45 degrees. The incident direction is located in the plane formed by the surface normal and the horizontal axis, and the light spot covers all sampling points.

[0091] A discrete scattering angle sequence is set, with the scattering angle range being 20 to 80 degrees and the angle step size being 10 degrees, resulting in seven discrete scattering angles;

[0092] For each sampling point, the receiving direction is adjusted angle by angle according to the discrete scattering angle sequence, and the scattering intensity is read to form a multi-angle intensity sequence;

[0093] Check the integrity of the data collection. If there are any data items that were not successfully collected, terminate the process and provide a re-collection conclusion.

[0094] Obtain the detector's saturation upper limit and noise lower limit, and limit the original intensity reading to the effective range; mark the point of overexposure when the maximum reading reaches the saturation upper limit, and mark the point of underexposure when the maximum reading is not higher than the noise lower limit; determine that the calibration is invalid and terminate the detection when the noise lower limit is not less than the saturation upper limit.

[0095] Further specific implementation steps include:

[0096] Install a parallel light source device diagonally above the surface of the yoga mat and fix the angle of incidence. The incident direction lies within the incident plane, which is formed by the intersection of the normal to the upper surface of the yoga mat and... The plane stretched by the axis; the light source emits monochromatic collimated light, and the light spot covers all the measuring points;

[0097] Set the scattering angle domain as , with step size Discretized From one angle , ;in, These are the minimum and maximum values ​​of the scattering angle, respectively; The angle is the distance from the walking distance; The total number of discrete angles; For the first A discrete scattering angle; Angle index;

[0098] Each The point detector has a built-in angle scanning mechanism that sequentially points the receiving direction within the incident plane. And read out the scattering intensity: ;in, For the first The measuring point at the ... The original scattered brightness at each scattering angle;

[0099] Construct the missing set: ;in, This is the set of ternary indexes that were not successfully collected.

[0100] like If the data is incomplete, the detection is stopped and a conclusion requiring resampling is output; where, For set The number of elements in the middle;

[0101] The physical saturation limit of the detector is denoted as The lower limit of noise is denoted as ;

[0102] Bound the original readings to ;in, For the original readings or Bounded brightness after clamping; and These are operations to retrieve the minimum and maximum values, respectively.

[0103] like If so, it is marked as an overexposure point;

[0104] like If so, it is marked as an underexposed point;

[0105] If it appears If the calibration fails, the calibration is deemed invalid, the test is stopped, and a calibration requirement is output.

[0106] The steps for multi-angle optical scattering acquisition and data integrity assurance of the tested object under fixed oblique illumination conditions are refined. First, a fixed-angle light source from above is used to ensure that each sampling point is illuminated under standardized lighting conditions, reducing the interference of environmental changes on the detection. Then, through a preset multi-angle sequence, the scattering intensity of each sampling point is scanned and acquired sequentially, achieving sufficient acquisition of the anisotropic characteristics of the surface. An automatic data integrity check mechanism is set up during the acquisition process. When any sampling point or angle fails to acquire data, it automatically prompts and stops the detection, avoiding distortion of the overall results due to missing data. Range limitation processing is also performed on the acquired data, including the identification and marking of overexposed and underexposed points, further improving the reliability and comparability of the data. In addition, if the acquisition equipment calibration malfunctions, it can automatically determine and output calibration conclusions. Compared with traditional detection methods that rely solely on a single angle, manual judgment, or simple light sources, this invention effectively improves the detection capability for complex textures and minute defects through meticulous measures such as multi-angle, multi-point, and automatic integrity verification. It reduces omissions and misjudgments, making the overall process more rigorous and reliable, and is particularly suitable for large-area, high-requirement quality control applications.

[0107] For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain a scalar response of the angular spectral intensity, forming a response matrix covering the entire domain, specifically including:

[0108] For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectral power value;

[0109] Arrange the scalar responses of all sampling points in row and column order to form an angular spectral value matrix.

[0110] Further specific implementation steps include:

[0111] Calculate the angular spectral power value at the measuring point scalar value :

[0112] ;

[0113] Construct the power value matrix .

[0114] Specific rules were established for the weighted integration of multi-angle signals at each sampling point and the generation of the global response matrix. By weighting and summarizing the multi-angle scattering intensity at each point according to fixed weights, a unique scalar response is obtained. Then, the scalar responses from all sampling points are integrated into a single response matrix. This effectively integrates information from multiple angles and directions, transforming previously scattered multi-dimensional data into comparable data in a single dimension, making the signal more concentrated and improving sensitivity to defects. The establishment of the response matrix also facilitates subsequent batch analysis, differential comparison, and defect localization, improving overall processing efficiency and data consistency. Unlike the common "single-angle threshold judgment" or "manual qualitative observation" methods in existing technologies, this invention utilizes algorithms to highly fuse multi-angle data, improving the ability to identify minute anomalies and facilitating automated, batch-based intelligent detection processes. This approach is particularly suitable for the practical needs of production processes for the subtle differentiation of surface defects and batch quantitative detection.

[0115] The process of obtaining a baseline response matrix on a defect-free standard sample, and then subtracting the current response matrix from the baseline response matrix point by point to form a residual value and a residual matrix, specifically includes:

[0116] A standard yoga mat without physical defects was selected as the reference sample, and multi-angle scattering intensity was collected according to step S2.

[0117] The original intensity readings of the reference sample are subjected to range limitation and integrity verification. After confirming that the reference data is complete, the reference angular spectral strength matrix is ​​generated according to step S3.

[0118] The angular spectral power matrix of the current object under test is subtracted point by point from the reference angular spectral power matrix to generate residual values ​​and residual matrices.

[0119] Further specific implementation steps include:

[0120] Select a standard yoga mat without any physical defects and collect raw readings. Constructing a baseline missing set ;in, For the first time on a defect-free standard yoga mat The sampling point at the th sampling point The original scattered brightness measured at each scattering angle;

[0121] like If the baseline data is incomplete, the detection will be stopped and a message indicating that resampling is required will be output.

[0122] When the baseline data is complete, the original readings are bounded as follows:

[0123] ;in, To be Bounded scattering brightness;

[0124] Calculate the baseline value The baseline power value matrix is ​​obtained. ;in, For the defect-free benchmark yoga mat in the first Column, No. Angular spectral density values ​​at the sampling points;

[0125] Calculate the first Point residual value ;

[0126] Constructing the residual matrix .

[0127] This paper proposes a method to obtain a baseline response using a defect-free standard sample and then perform point-by-point differencing with the response matrix of the sample under test to form a residual matrix. This step quantifies the differences in the detected data as residuals by comparing the response matrices of the standard sample and the current object being tested. This method effectively eliminates systematic and accidental interferences such as those between equipment, batches, and ambient lighting, improving the accuracy and specificity of defect identification. The baseline data acquisition also requires integrity verification and range constraints to ensure the quality of the comparison data. The residual matrix can intuitively highlight defect areas, achieving precise defect location and quantification. Unlike conventional methods that rely on absolute numerical thresholds or manual experience standards, this invention, through personalized and dynamic standard sample differencing, adapts to the actual situation of varying product, equipment, and process batches, improving detection sensitivity and accuracy, and providing fundamental support for intelligent quality control.

[0128] The step of performing discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix specifically includes:

[0129] Discrete differencing is performed on the residual matrix along the horizontal and vertical directions, with one-sided differencing used for the matrix boundaries and central differencing used for the matrix interior.

[0130] The horizontal and vertical difference results are combined to form a mutation intensity index, creating a mutation intensity matrix covering the entire domain.

[0131] Further specific implementation steps include:

[0132] Scalar for calculating the mutation intensity of residual power ;in, ;

[0133] ; For along Discrete difference of direction; For along Discrete difference of direction;

[0134] Constructing a mutation matrix .

[0135] This involves discrete differencing of the residual matrix and generation of the mutation intensity matrix. This step synthesizes a mutation intensity index by differencing the residual values ​​along both row and column directions, forming a mutation intensity matrix. This effectively distinguishes smooth regions from abrupt changes, isolating areas of drastic spatial variation (i.e., potential defect areas) from the overall picture. Compared to analyzing only the residuals themselves, this spatial differencing method is more sensitive to surface mutations, adapting to defects of various shapes, orientations, and sizes, and is particularly effective for edge defects, cracks, and dents. Furthermore, combining this with the synthesized mutation intensity index helps to integrate changes in multiple directions, further enhancing the ability to focus on anomalous regions. Unlike traditional detection methods that rely solely on global mean, variance, or a single criterion, this approach lays a solid foundation for subsequent automatic screening, segmentation, and quantification, improving the accuracy and reliability of defect detection.

[0136] The mutation intensity statistics based on the reference benchmark are set with a fixed threshold, a set of candidate outliers is generated on the mutation intensity matrix, and a preliminary defect determination is completed, specifically including:

[0137] The spatial abrupt change intensity is calculated on the angular spectral strength matrix of the reference sample, and the average and maximum values ​​of the results are obtained.

[0138] An interpolation threshold for mutation detection is set by performing interpolation between the average and maximum values ​​using a fixed ratio of 0.5.

[0139] Sampling points above a fixed threshold are selected from the mutation intensity matrix of the object under test to form a set of candidate anomalies;

[0140] When the candidate anomaly set is empty, output a defect-free conclusion and end the process; when the candidate anomaly set is not empty, proceed to connected component processing.

[0141] Further specific implementation steps include:

[0142] Calculate the spatial mutation intensity of the baseline power value ;in, ;

[0143] ; , The benchmark values ​​are respectively at and Discrete difference of direction;

[0144] calculate average ;

[0145] Get maximum value ;

[0146] Set mutation detection threshold ;

[0147] Construct a set of candidate anomalies ;

[0148] like The output is flawless and provides... , empty defect set and empty area; where, The number of defects;

[0149] like If so, then continue with defect identification.

[0150] A method for setting adaptive thresholds based on baseline mutation intensity statistics and filtering outomas using a mutation intensity matrix was defined. Specifically, the average and maximum mutation intensities of standard samples were calculated, and an interpolation method was used to set the judgment threshold. This ensures that the threshold can adapt to changes in product batches while maintaining high sensitivity to anomalies. After automatically filtering the outomas, if no anomalies are found, a "no defect" conclusion is output directly, improving detection efficiency. If anomalies are found, further processing is initiated. This approach differs from previous manual threshold settings or one-size-fits-all criteria, effectively avoiding "missed detections" and "false alarms," ​​balancing flexibility and accuracy. Data-driven threshold adjustment makes the algorithm more adaptive and industrially applicable, especially suitable for applications with highly variable production processes and diverse defect types.

[0151] The process of extracting connected components on a discrete grid based on four-adjacency relationships, eliminating candidate regions with insufficient area according to a minimum effective area threshold, sequentially numbering the retained regions, and calculating the geometric center coordinates and area of ​​each region specifically includes:

[0152] An undirected graph structure is constructed on a discrete grid using four-adjacency relationships, and the connected components of the candidate outlier set are extracted accordingly.

[0153] Calculate the number of sampling points and the corresponding physical area contained in each connected component;

[0154] Set the minimum effective area threshold to the sum of the areas of two grid cells, and remove connected components with an area smaller than this threshold.

[0155] The remaining connected components are renumbered sequentially, and the geometric center coordinates and area of ​​each connected component are calculated.

[0156] Further specific implementation steps include:

[0157] Setting four-adjacency relationships on a discrete grid :

[0158] like ,but ;by For vertex set, The connected components of an undirected graph with edge set are defined as the set of defective connected components. ;in, This is the four-adjacency relation symbol; For the first A set of points in a connected domain; For connected component indexes;

[0159] Calculate the first Area of ​​each connected region ;in, For set The number of elements;

[0160] Set minimum effective area threshold ;

[0161] If for a certain have Then remove the first one. Connected components;

[0162] If all connected components are removed, the output is defect-free and gives... Otherwise, renumber all retained connected components sequentially. ;

[0163] Get the Geometric center coordinates of the defective connected domain :

[0164] .

[0165] The steps for extracting connected components from anomaly sets using four-adjacency relationships and removing false defects based on area thresholds have been refined. By segmenting connected components using four-adjacency relationships, spatially related anomalies can be naturally grouped into a single defect region, avoiding false alarms from single-point noise and discrete anomalies. Setting a minimum effective area threshold automatically filters out small-area ineffective defect regions, further reducing false positives. Each retained defect region is individually numbered, and its geometric center and area are calculated, facilitating subsequent quantitative description and automatic recording of defects. This step is highly practical in actual production, accurately describing and classifying defects while reducing the burden of manual post-processing.

[0166] The output includes the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and corresponding visual markers are generated on the coordinated result map. Specifically, this includes:

[0167] Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect region;

[0168] On the coordinated result map, generate dot marks for the geometric center of each defect.

[0169] Further specific implementation steps include:

[0170] Output number of defects Coordinates of the center of each defect Area of ​​each defect region ;

[0171] exist Marked with a dot .

[0172] The entire testing process has been automated and visualized, outputting the final results. Specifically, it outputs the number of defects, the coordinates of each defect's center, and its area, all visually marked on a coordinate graph. This greatly facilitates direct interpretation and production decision-making for operators, while also providing quantitative data for production traceability, process improvement, and product repair. Automated visualization replaces traditional manual judgment, recording, and paper-based statistics, making test results more intuitive, objective, and easier to archive. This approach effectively promotes the digitalization, streamlining, and intelligentization of quality control, and also facilitates seamless integration and data sharing with upstream and downstream systems.

[0173] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0174] The above description is only a preferred embodiment of the present invention. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the technical principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for automatic identification of surface defects on yoga mats based on intelligent sensors, characterized in that, include: S1. Establish a two-dimensional coordinate frame on the upper surface of the object under test, set up a scattering acquisition array composed of discrete sampling points, and keep the installation posture of all sampling points consistent with the outer normal of the object under test. S2. Irradiate the object under fixed oblique conditions, perform scattering intensity scanning on each sampling point according to the preset multi-angle sequence and record the intensity reading, and at the same time complete the data integrity check and reading range limitation. S3. For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectrum value, and a response matrix covering the entire domain is formed. S4. Obtain the baseline response matrix on the defect-free standard sample, and subtract the current response matrix from the baseline response matrix point by point to form the residual value and the residual matrix. S5. Perform discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix. S6. Based on the mutation intensity statistics of the reference benchmark, a fixed threshold is set, a set of candidate anomalies is generated on the mutation intensity matrix, and the preliminary defect judgment is completed. S7. Extract connected components on the discrete grid according to the four-adjacency relationship, remove candidate regions with insufficient area according to the minimum effective area threshold, and complete the sequential numbering of the retained regions and calculate the geometric center coordinates and region area. S8. Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and generate corresponding visual markers on the coordinate result map.

2. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 1, characterized in that, The process of establishing a two-dimensional coordinate frame on the upper surface of the object under test, setting up a scattering acquisition array composed of discrete sampling points, and ensuring that the installation orientation of all sampling points is consistent with the external normal of the object under test specifically includes: A two-dimensional rectangular coordinate frame is established on the upper surface of the yoga mat being tested. The origin is set at the lower left corner, the long side is used as the positive direction of the horizontal axis, and the short side is used as the positive direction of the vertical axis. The coordinate unit is millimeters. The area to be measured is divided into equally spaced grids according to the set size resolution, with the horizontal and vertical spacing set to the same value; An integrated optical scattering angle detector is fixedly installed at each grid point, and the normal direction of the detector is consistent with the outer normal of the object being measured. The length parameter of the object under test is quantized at the micrometer level to generate an effective computational domain that is strictly aligned with the grid, and subsequent acquisition and coordinate output are restricted to this effective computational domain.

3. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 2, characterized in that, The process involves irradiating the object under fixed oblique illumination conditions, performing a scattering intensity scan at each sampling point according to a preset multi-angle sequence, recording the intensity readings, and simultaneously performing data integrity checks and limiting the reading range. Specifically, this includes: A parallel light source device is installed obliquely above the surface of the object being measured, with a fixed incident angle of 45 degrees. The incident direction is located in the plane formed by the surface normal and the horizontal axis, and the light spot covers all sampling points. A discrete scattering angle sequence is set, with the scattering angle range being 20 to 80 degrees and the angle step size being 10 degrees, resulting in seven discrete scattering angles; For each sampling point, the receiving direction is adjusted angle by angle according to the discrete scattering angle sequence, and the scattering intensity is read to form a multi-angle intensity sequence; Check the integrity of the data collection. If there are any data items that were not successfully collected, terminate the process and provide a re-collection conclusion. Obtain the detector's saturation upper limit and noise lower limit, and limit the original intensity reading to the effective range; mark the point of overexposure when the maximum reading reaches the saturation upper limit, and mark the point of underexposure when the maximum reading is not higher than the noise lower limit; determine that the calibration is invalid and terminate the detection when the noise lower limit is not less than the saturation upper limit.

4. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 3, characterized in that, For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain a scalar response of the angular spectral intensity, forming a response matrix covering the entire domain, specifically including: For each sampling point, the multi-angle scattering intensity is weighted and summarized according to a fixed angle weighting rule to obtain the scalar response of the angular spectral power value; Arrange the scalar responses of all sampling points in row and column order to form an angular spectral value matrix.

5. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 4, characterized in that, The process of obtaining a baseline response matrix on a defect-free standard sample, and then subtracting the current response matrix from the baseline response matrix point by point to form a residual value and a residual matrix, specifically includes: A standard yoga mat without physical defects was selected as the reference sample, and multi-angle scattering intensity was collected according to step S2. The original intensity readings of the reference sample are subjected to range limitation and integrity verification. After confirming that the reference data is complete, the reference angular spectral strength matrix is ​​generated according to step S3. The angular spectral power matrix of the current object under test is subtracted point by point from the reference angular spectral power matrix to generate residual values ​​and residual matrices.

6. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 5, characterized in that, The step of performing discrete difference operations along the row and column directions on the residual matrix to synthesize the mutation intensity index and obtain the mutation intensity matrix specifically includes: Discrete differencing is performed on the residual matrix along the horizontal and vertical directions, with one-sided differencing used for the matrix boundaries and central differencing used for the matrix interior. The horizontal and vertical difference results are combined to form a mutation intensity index, creating a mutation intensity matrix covering the entire domain.

7. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 6, characterized in that, The mutation intensity statistics based on the reference benchmark are set with a fixed threshold, a set of candidate outliers is generated on the mutation intensity matrix, and a preliminary defect determination is completed, specifically including: The spatial abrupt change intensity is calculated on the angular spectral strength matrix of the reference sample, and the average and maximum values ​​of the results are obtained. An interpolation threshold for mutation detection is set by performing interpolation between the average and maximum values ​​using a fixed ratio of 0.

5. Sampling points above a fixed threshold are selected from the mutation intensity matrix of the object under test to form a set of candidate anomalies; When the candidate anomaly set is empty, output a defect-free conclusion and end the process; when the candidate anomaly set is not empty, proceed to connected component processing.

8. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 7, characterized in that, The process of extracting connected components on a discrete grid based on four-adjacency relationships, eliminating candidate regions with insufficient area according to a minimum effective area threshold, sequentially numbering the retained regions, and calculating the geometric center coordinates and area of ​​each region specifically includes: An undirected graph structure is constructed on a discrete grid using four-adjacency relationships, and the connected components of the candidate outlier set are extracted accordingly. Calculate the number of sampling points and the corresponding physical area contained in each connected component; Set the minimum effective area threshold to the sum of the areas of two grid cells, and remove connected components with an area smaller than this threshold. The remaining connected components are renumbered sequentially, and the geometric center coordinates and area of ​​each connected component are calculated.

9. The method for automatic identification of yoga mat surface defects based on intelligent sensors according to claim 8, characterized in that, The output includes the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect, and corresponding visual markers are generated on the coordinated result map. Specifically, this includes: Output the number of defects, the geometric center coordinates of each defect, and the area of ​​each defect region; On the coordinated result map, generate dot marks for the geometric center of each defect.

Images