Insulating plate processing defect visual inspection and sorting method and system

Abstract

The invention discloses an insulating plate processing defect visual inspection and sorting method and system, particularly relates to the technical field of automatic sorting, and aims to solve the problem that subsequent qualified insulating plates are misjudged due to impact interference generated by sorting action in the prior art. The method comprises the following steps: recording physical characteristic parameters after sorting action, acquiring an initial image of a next insulating plate, predicting a dynamic interference trend and evaluating a pose offset risk, further analyzing interference mode characteristics to decide whether to start compensation, dynamically adjusting a defect detection threshold according to the offset risk when starting, and finally detecting the defect of the next insulating plate. And finally, defect detection is carried out based on the adjusted threshold value, and a sorting execution mechanism is controlled to act. The misjudgment rate is effectively reduced, and the sorting accuracy and the first pass yield of a production line are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic sorting, and more specifically, to a method and system for visual detection and sorting of processing defects of insulating boards. Background Art

[0002] During the production and processing of insulating boards, rapid and accurate detection and sorting of processing defects such as scratches, pits, and stains on their surfaces are crucial for ensuring product quality. Currently, industrial applications generally use automated visual detection and sorting systems to accomplish this task. This system typically includes a visual detection unit and a physical sorting unit; during operation, the insulating boards pass through the detection area in sequence on the conveyor belt. After the industrial camera captures an image and the processing system analyzes it, it determines whether the product is qualified. Subsequently, the system controls the sorting actuator (such as a pneumatic nozzle or a mechanical push rod) to remove the defective products from the production line. This automated method significantly improves the detection efficiency and replaces the traditional manual visual inspection operation mode.

[0003] However, in practical applications, the above technical solutions still have defects that need to be urgently solved: due to the compact layout of the production line, the instantaneous impact generated by the sorting actuator during operation, such as the airflow disturbance formed by the exhaust of the pneumatic nozzle or the minute vibration caused by the movement of mechanical components, will be transmitted to the adjacent subsequent qualified insulating boards. This interference may cause unexpected position offsets or attitude changes of the qualified boards. When they enter the subsequent re-inspection station or directly result in the imaging attitude of this detection not meeting the preset conditions, they are very likely to be misjudged by the system as products with position or shape defects, thus causing the incorrect rejection of qualified products. This misjudgment problem caused by the secondary effect of the sorting action directly affects the accuracy of the final sorting result and reduces the throughput rate of the production line. Summary of the Invention

[0004] In order to overcome the above defects of the prior art, the present invention provides a method and system for visual detection and sorting of processing defects of insulating boards to solve the problems raised in the above background art.

[0005] To achieve the above object, the present invention provides the following technical solutions: A method for visual detection and sorting of processing defects of insulating boards, including: S1. Record the physical characteristic parameters of the corresponding action after the sorting actuator completes the action on the current insulating board, and simultaneously obtain the initial position image of the next insulating board when it enters the detection area; S2. Query the pre-established interference mapping relationship according to the physical characteristic parameters and the board flow sequence position of the next insulating board, and predict the dynamic interference trend of the next insulating board; S3. Based on the dynamic interference trend and the initial position image, comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference; S4. Obtain the current interference mode characteristics by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; determine whether to initiate pose compensation for the next insulating plate by performing similarity matching analysis between the mode characteristics and the interference modes in historical misjudgment cases. S5. When pose compensation is activated, the defect detection and evaluation threshold for the next insulating board is dynamically adjusted based on the actual pose deviation risk. S6. Based on the adjusted defect detection evaluation threshold, perform defect detection on the next insulation board, and control the sorting execution mechanism to perform sorting actions according to the detection results.

[0006] Furthermore, the physical characteristic parameters of the corresponding actions are recorded, including the collection of pressure change data and drive current data generated by the sorting actuator during operation; Acquiring the initial position image when the next insulating plate enters the detection area involves acquiring an image containing the corresponding next insulating plate using an industrial camera under the action of a trigger signal, and recording the timestamp corresponding to the acquisition of the initial position image. The trigger signal is associated with the event of the sorting actuator completing its action.

[0007] Furthermore, querying the pre-established interference mapping relationship includes: combining the sorting action type with the plate flow sequence position, wherein the sorting action type is classified according to the time-domain statistical characteristics of the physical characteristic parameters; The pre-established interference mapping relationship is constructed by analyzing the differential features of the initial position image of the next insulating board relative to the interference-free reference image when it arrives at the detection area based on its board flow sequence position after a specific sorting action is completed in historical data. Predicting dynamic interference trends includes outputting and combining trend vectors that characterize the intensity and direction of interference.

[0008] Furthermore, a comprehensive assessment of the actual pose deviation risk includes converting the predicted deviation direction and deviation amount contained in the dynamic disturbance trend into a virtual deviation field for the initial position image. Extract the actual edge contour of the insulating plate from the initial position image; By comparing the geometric differences between the actual edge contour and the expected edge contour after virtual offset field transformation, a quantitative assessment result of the actual pose offset risk is generated.

[0009] Furthermore, the mode characteristics of the current interference are obtained through multi-scale time-frequency feature analysis of the physical characteristic parameters, including: Wavelet packet decomposition is performed on the recorded physical characteristic parameters to obtain signal components in different frequency bands; Calculate the singularity index and energy entropy of each signal component to form the pattern characteristics of the current interference; By performing similarity matching analysis between pattern features and interference patterns in historical misjudgment cases, the determination of whether to initiate pose compensation for the next insulating plate includes: The matching degree is calculated between the current interference pattern features and the typical misjudged interference pattern features that are constructed in the same way and stored in the historical misjudgment case library; The matching degree is compared with a dynamic alert threshold set based on historical false positive rate statistics; If the matching degree exceeds the dynamic alert threshold, it is determined that the pose compensation of the next insulating plate needs to be initiated.

[0010] Furthermore, calculating the singularity index and energy entropy of each signal component includes: For each signal component, the singularity index is estimated by analyzing the variation of its wavelet transform modulus maxima with scale. At the same time, the proportion of the energy of the signal component in the total energy is calculated and substituted into the Shannon entropy formula to calculate the energy entropy; The singularity index and energy entropy of all signal components are combined in frequency band order to form a multidimensional feature vector, which constitutes the mode characteristics of the current interference.

[0011] Furthermore, the matching degree calculation between the current interference pattern characteristics and the typical misjudgment interference pattern characteristics constructed in the same way stored in the historical misjudgment case database includes: Calculate the Euclidean distance between the current interference pattern feature vector and one or more typical misjudged interference pattern feature vectors pre-stored in the historical misjudgment case library; The reciprocal of the calculated minimum Euclidean distance is used as the matching degree to characterize the similarity between the two.

[0012] Furthermore, the defect detection and evaluation threshold for the next insulating plate is dynamically adjusted based on the actual pose deviation risk, including: The quantitative assessment results of the actual pose deviation risk are mapped to the adjusted weighting coefficients for the defect detection judgment threshold; Based on the adjusted weighting coefficients and the preset benchmark defect detection threshold, the dynamically adjusted defect detection threshold is obtained through linear interpolation. The dynamically adjusted defect detection threshold is used in the defect detection process of the next insulation board.

[0013] Furthermore, performing defect detection on the next insulating board based on the adjusted defect detection evaluation threshold includes: acquiring the actual detection image of the next insulating board in the detection area; and comparing the size and contrast information of the defect features in the actual detection image with the dynamically adjusted defect detection evaluation threshold. The sorting action controlled by the sorting execution mechanism based on the detection results includes: when the size and contrast of the defect feature simultaneously exceed the dynamically adjusted defect detection and evaluation threshold, a sorting instruction is generated and sent to the sorting execution mechanism to remove the corresponding next insulating board; otherwise, the next insulating board is allowed to continue to be conveyed on the assembly line.

[0014] On the other hand, the present invention provides a visual inspection and sorting system for defects in insulating board processing, comprising: The parameter acquisition module is used to record the physical characteristic parameters of the corresponding action after the sorting execution mechanism completes the action on the current insulating plate, and at the same time acquire the initial position image of the next insulating plate when it enters the detection area. The trend prediction module is used to query the pre-established interference mapping relationship based on the physical characteristic parameters and the position of the plate current sequence of the next insulating plate, and predict the dynamic interference trend of the next insulating plate. The risk assessment module is used to comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference based on dynamic interference trends and initial position images. The compensation judgment module is used to obtain the mode characteristics of the current interference by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; and to determine whether to initiate pose compensation for the next insulating plate by performing similarity matching analysis between the mode characteristics and the interference modes in historical misjudgment cases. The risk adjustment module is used to dynamically adjust the defect detection and evaluation threshold for the next insulating board based on the actual pose offset risk when pose compensation is initiated. The action execution module is used to perform defect detection on the next insulation board based on the adjusted defect detection evaluation threshold, and control the sorting execution mechanism to perform sorting actions according to the detection results.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. By sensing the physical disturbances of the sorting action in real time and proactively predicting their impact on subsequent insulation boards, the problem of misjudging qualified products due to secondary interference from mechanical actions is effectively solved. Immediately after the sorting action is completed, its physical characteristic parameters are recorded and the initial position image of the next insulation board is obtained simultaneously. A precise correlation between physical events and visual detection is established. Furthermore, by querying the pre-established interference mapping relationship, the abstract mechanical impact is transformed into a quantifiable dynamic interference trend prediction, realizing the transformation from passive detection to proactive early warning. This prediction method based on the combination of physical mechanisms and historical data provides a reliable basis for accurately assessing the risk of pose deviation.

[0016] 2. By identifying interference pattern characteristics through multi-scale time-frequency analysis and intelligently matching them with historical misjudgment cases, the system dynamically decides whether to activate the compensation mechanism. When compensation is determined to be required, the system adaptively adjusts the defect detection threshold based on the real-time assessment of pose offset risk, enabling the detection standard to be flexibly optimized according to changes in working conditions. This closed-loop optimization mechanism ensures that, in the presence of interference, the detection system can effectively identify real defects while significantly reducing the misjudgment rate caused by pose changes, ultimately improving sorting accuracy while ensuring the first-pass yield of the production line. Attached Figure Description

[0017] Figure 1 This is a flowchart of a visual inspection and sorting method for defects in insulating board processing according to the present invention; Figure 2 This is a schematic diagram of the structure of a visual inspection and sorting system for defects in insulation board processing according to the present invention. Detailed Implementation

[0018] 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.

[0019] Example 1: Figure 1 This invention provides a method for visual inspection and sorting of defects in insulating board processing, comprising: S1. After the sorting execution mechanism completes the action on the current insulating plate, record the physical characteristic parameters of the corresponding action, and at the same time obtain the initial position image when the next insulating plate enters the detection area. S2. Based on the physical characteristic parameters and the position of the plate current sequence of the next insulating plate, query the pre-established interference mapping relationship and predict the dynamic interference trend of the next insulating plate. S3. Based on the dynamic interference trend and the initial position image, comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference; S4. Obtain the current interference mode characteristics by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; determine whether to initiate pose compensation for the next insulating plate by performing similarity matching analysis between the mode characteristics and the interference modes in historical misjudgment cases. S5. When pose compensation is activated, the defect detection and evaluation threshold for the next insulating board is dynamically adjusted based on the actual pose deviation risk. S6. Based on the adjusted defect detection evaluation threshold, perform defect detection on the next insulation board, and control the sorting execution mechanism to perform sorting actions according to the detection results.

[0020] S1. After the sorting actuator completes its action on the current insulating plate, it records the physical characteristic parameters of the corresponding action. At the same time, it acquires the initial position image of the next insulating plate entering the detection area. The implementation is as follows: After the sorting actuator completes its action on the current insulating plate, the data recording process is immediately initiated. This recording of the physical characteristics of the corresponding action is achieved through pressure sensors installed in the pneumatic circuit of the sorting actuator and current sensors in the power supply circuit of the drive motor. The pressure sensors monitor gas pressure changes in real time during the sorting actuator's operation. Their range is selected from 0 to 1 MPa, based on the factory's standard operating pressure range for the pneumatic system, and the sampling frequency is set to 2000 Hz, sufficient to capture millisecond-level pressure transients generated by the pneumatic components. The current sensors collect three-phase drive current data from the drive motor during startup, operation, and shutdown, also with a sampling frequency of 2000 Hz, thus accurately recording current fluctuations caused by changes in motor torque. Both pressure change data and drive current data are synchronously recorded and stored in a buffer in time-series format, with each set of data accompanied by a time stamp accurate to the millisecond level.

[0021] The acquisition of the initial position image of the next insulating plate as it enters the detection area is performed by an industrial camera positioned directly above the detection area. The industrial camera has a resolution of 5 megapixels and is equipped with a ring light source to provide uniform illumination. The exposure of the industrial camera is controlled by a specific trigger signal. The generation of this trigger signal is directly related to the completion of the sorting actuator's action: when the programmable logic controller controlling the sorting actuator issues a level indicating the completion of the action, this level signal is sent to a configurable delay trigger. The delay trigger calculates a delay value based on the constant operating speed of the conveyor belt and the known distance between the leading edge of the next insulating plate and the center point of the detection area. For example, if the conveyor belt speed is 0.5 m / s and the distance is 0.1 m, the delay value is set to 200 milliseconds. After this precise delay, the delay trigger outputs a pulse as a trigger signal to the industrial camera. Upon receiving this trigger signal, the industrial camera performs a global exposure, acquiring a digital image containing the complete next insulating plate, which is the initial position image. The system simultaneously records the timestamp corresponding to the acquisition of this image.

[0022] To ensure precise correlation between pressure change data, drive current data, and the timestamps of the initial position image, all acquisition devices communicate with a central controller, which maintains a high-precision global timer. Upon receiving a signal indicating the start of the sorting actuator's operation, the central controller sends a synchronous start command to the pressure and current sensors and records the current value T0 of the global timer. Once the industrial camera completes image acquisition and returns its timestamp T1, the central controller calibrates the timing information of all sensor data to a relative time axis starting from T0. This mechanism ensures strict temporal alignment of data from different sources; for example, it allows for accurate determination of the pressure sensor reading 50 milliseconds before the initial position image acquisition. All acquired raw data is assigned a unique identifier associated with the production batch and insulation board serial number and stored in a database, providing a strictly synchronized and clearly sourced data foundation for subsequent analysis steps.

[0023] S2. Based on the physical characteristic parameters and the position of the current sequence of the next insulating plate, query the pre-established interference mapping relationship, predict the dynamic interference trend of the next insulating plate, and implement it as follows: After acquiring and synchronizing the physical characteristic parameters and initial position images, the process begins by querying pre-established interference mapping relationships to predict dynamic interference trends. The first step is classifying the sorting action types based on the time-domain statistical characteristics of the physical characteristic parameters. Specifically, the processing unit reads recorded pressure change data and drive current data from the cache. For pressure change data, the root mean square (RMS) and peak value over a complete action cycle are calculated; similarly, the RMS and peak current are calculated for drive current data. For example, in a typical pneumatic sorting action, the peak pressure data might be around 0.7 MPa, while the peak current might be around 5 amperes. Subsequently, these calculated time-domain statistical characteristic values ​​are matched against a predefined action type classification table. This classification table was established during the production line commissioning phase by executing hundreds of standard sorting actions and statistically analyzing their characteristic value ranges. For example, a combination of peak pressure of 0.6 to 0.8 MPa and peak current of 4.5 to 5.5 A is defined as a standard high-impact action type; a combination of peak pressure of 0.3 to 0.5 MPa and peak current of 2 to 3 A is defined as a low-impact action type. This matching categorizes the current sorting action into a specific type. If the calculated characteristic values ​​fall within the overlap region of the two types, the type that matches the most frequently occurring action type in history is preferred, or a decision is made based on the principle that the current characteristic has a slightly higher weight than the pressure characteristic, because current more directly reflects the actual kinetic energy output of the actuator.

[0024] Next, the determined sorting action type is combined with the position of the next insulating board in the board flow sequence. The board flow sequence position refers to the sequential number of the insulating board immediately following the currently sorted one on the conveyor belt; for example, the next one is position 1, and the one after that is position 2. The combination of sorting action type and board flow sequence position is represented as a data pair, such as (Standard Strong Impact, Position 1). This combination serves as a key index for querying interference mapping relationships.

[0025] The core of this step is querying the pre-established interference mapping relationship. This interference mapping relationship is constructed during the offline learning phase before the system is put into formal operation. The construction process is as follows: A large amount of historical production data is collected, which includes initial position images of insulating boards at different board flow sequence positions when they arrive at the detection area after different sorting action types are completed. For each such initial position image, its differential features relative to the interference-free reference image need to be calculated. The interference-free reference image is the average of multiple images of the same type of insulating board in the detection area under the condition that the production line is completely stable and there is no interference from any sorting action. The calculation of differential features first involves grayscale processing of the initial position image and the interference-free reference image, and then using a feature point-based image registration algorithm (e.g., using SIFT or ORB feature detectors to extract feature points and using the RANSAC algorithm to calculate the homography matrix for alignment) to accurately register the two images to eliminate the fundamental differences caused by the slight positional fluctuations of the insulating board itself on the conveyor belt. After registration, the absolute difference of grayscale values ​​of corresponding pixels is calculated. Finally, all difference pixel values ​​are summed and averaged to obtain a scalar difference value, which characterizes the overall degree of image offset. For example, a difference value exceeding 10 may indicate a significant offset. Furthermore, to obtain offset direction information, the offset vector of the centroid coordinates of the difference image relative to the image center point needs to be calculated. By analyzing a large amount of historical data, a representative range of difference feature values ​​and its typical offset direction (e.g., offset along the conveyor belt direction is dominant) are statistically derived for each combination (sorting action type, plate flow sequence position). All these relationships are stored in a two-dimensional lookup table, namely the interference mapping table. The structure of this table can be viewed as a matrix, with row indices representing sorting action types and column indices representing plate flow sequence positions. Each cell stores the statistically obtained average difference value and average offset vector (Δx, Δy) for that combination.

[0026] The process of predicting dynamic interference trends involves querying this interference mapping table. The system uses the currently obtained combination of (sorting action type, board flow sequence position) as an index to query the interference mapping table. The query result is a trend vector representing the intensity and direction of the interference. This trend vector is a two-dimensional vector; its first component represents the estimated offset along the conveyor belt direction, in pixels; the second component represents the estimated offset perpendicular to the conveyor belt direction. For example, the query result might return a trend vector (+15, -2), indicating that the insulating board is predicted to move 15 pixels along the conveyor belt direction and offset 2 pixels to the left. This trend vector is the predicted dynamic interference trend, which will serve as direct input for subsequent assessment of the actual pose offset risk. If the queried index combination does not have a completely matching entry in the mapping table, the nearest neighbor interpolation method is used to find the existing combination that is closest in the sorting action type feature space and board flow sequence position, and its corresponding trend vector is used as the predicted value. The entire prediction process relies on the previously established empirical data model, achieving quantitative anticipation of the secondary interference effects of sorting actions. This method, which uses historical data to statistically model and query mapping relationships, avoids the computational burden of real-time simulation of complex fluid mechanics or structural dynamics, thus ensuring the real-time nature of online prediction.

[0027] S3. Based on the dynamic interference trend and the initial position image, comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference, and implement it as follows: After obtaining the dynamic disturbance trend (i.e., trend vector) and its initial position image of the next insulating plate, a comprehensive assessment of the actual pose shift risk is performed. This assessment first requires converting the predicted offset direction and amount contained in the dynamic disturbance trend into a virtual offset field for the initial position image. Essentially, this conversion transforms a pixel-based two-dimensional translation prediction into a spatial transformation model that can act on the entire image. The virtual offset field is represented in computer memory as a two-dimensional array with the same width and height as the initial position image, where each element stores a two-dimensional displacement vector. This displacement vector is directly filled with the predicted trend vector (Δx, Δy). For example, if the trend vector is (15, -2), then the displacement vector at every position in the virtual offset field is (15, -2), representing a global translational transformation expectation that the entire insulating plate image will move 15 pixels along the conveyor belt direction (positive x-axis) and 2 pixels perpendicular to the conveyor belt direction (negative y-axis). The assumption of a uniform offset field is based on a simplification of the motion pattern of the insulating plate after it is disturbed on the conveyor belt, that is, considering it as a rigid body translation in the short term.

[0028] After constructing the virtual offset field, the next step is to accurately extract the actual edge contours of the insulation board from the initial position image. This process uses the classic Canny edge detection algorithm, and its specific steps and parameter settings are as follows. First, the acquired color initial position image is converted from the RGB color space to a grayscale image. Next, a Gaussian filter is used to convolve the grayscale image to achieve smoothing and noise reduction. The size of the Gaussian kernel is usually 5×5 pixels, and its standard deviation σ can be adjusted according to the image noise level, for example, set to 1.5. Then, the gradient magnitude and direction of the image in the x and y directions are calculated, usually using the Sobel operator. Based on this, non-maximum suppression is performed, retaining only the local gradient maximum points in the gradient direction of each pixel, thereby refining the edges. Finally, a dual thresholding method is applied to identify and connect edges: a high threshold and a low threshold are set, for example, the high threshold is set to 30% of the maximum gradient magnitude in the image, and the low threshold is set to 40% of the high threshold (i.e., 12% of the maximum value). Points with gradient magnitudes above a high threshold are identified as strong edges, those below a low threshold are discarded, and points in between are retained as weak edges only if they are connected to strong edges. In the final binarized edge image, all connected white pixels constitute the edge contour. From all contours, the contour with the largest area and shape closest to the outline of the insulation board is selected as the actual edge contour of the insulation board by calculating the area enclosed by the contour and combining the geometric features of the contour (such as rectangularity).

[0029] Subsequently, a quantitative assessment of the actual pose deviation risk is generated by comparing the geometric difference between the actual edge contour and the expected edge contour after virtual offset field transformation. The process of generating the expected edge contour involves applying a geometric transformation to the actual edge contour. The coordinates (x, y) of each pixel on the actual edge contour are added to the displacement vector (Δx, Δy) of the virtual offset field at that point, resulting in new coordinates (x + Δx, y + Δy). Connecting all the transformed points forms the expected edge contour. The core of the quantitative assessment is calculating the geometric difference between these two contours, which is measured using the Hausdorff distance. The Hausdorff distance is a measure of the maximum mismatch between two sets of points.

[0030] Given the actual edge contour point set A and the expected edge contour point set B, first calculate the directed Hausdorff distance h(A, B) from A to B. This is defined as the maximum distance from each point in point set A to the nearest point in point set B, expressed mathematically as: h(A, B) = max{ min{ d(a, b) for b in B} for a in A}, where d(a, b) represents the Euclidean distance between points a and b. Similarly, calculate the directed Hausdorff distance h(B, A) from B to A. The final Hausdorff distance H(A, B) is the maximum of these two directed distances, i.e., H(A, B) = max( h(A, B), h(B, A) ).

[0031] The calculated Hausdorff distance value (in pixels) is directly used as the quantitative assessment result of the actual pose deviation risk. The larger this value, the greater the deviation between the actual position of the insulating plate and the position predicted based on the historical interference model, i.e., the higher the risk of pose deviation after the current interference. This single scalar risk assessment result provides a clear and comparable numerical basis for subsequent decision-making processes. The entire assessment process achieves an objective and quantitative assessment of potential pose deviation risks by rigorously comparing predictions based on physical experience with real-time visual observation data. If a contour that meets the conditions is not found successfully during the contour extraction stage, the pose deviation risk of the insulating plate is marked with an extremely high preset value (e.g., 1000 pixels), and an anomaly handling process is triggered, such as directly identifying it as an object requiring key monitoring.

[0032] S4. Obtain the mode characteristics of the current interference by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; determine whether to initiate pose compensation for the next insulating plate by performing similarity matching analysis between the mode characteristics and the interference modes in historical misjudgment cases. The implementation is as follows: After completing the actual pose deviation risk assessment, the system performs parallel deep analysis of interference patterns to determine whether pose compensation should be initiated. This process begins with multi-scale time-frequency feature analysis of physical characteristic parameters to extract the pattern features of the current interference. First, wavelet packet decomposition is performed on the recorded physical characteristic parameters. Wavelet packet decomposition uses Daubechies 4th order wavelets as basis functions, performing three-level deep decomposition on both pressure change data and drive current data. Three-level decomposition is chosen because it can divide common interference signals in industrial environments (frequency range typically 0-1000 Hz) into eight frequency bands with clear physical meaning. The decomposition process is achieved by recursively applying high-pass and low-pass filters and downsampling, generating a set of signal components covering different frequency ranges, each component representing the oscillation characteristics of the original signal within a specific frequency band. For example, the third-level decomposition will generate eight frequency band components, numbered (3,0) to (3,7) sequentially from low to high frequency. The (3,0) component contains frequency components from 0 to 62.5 Hz, and the (3,7) component contains frequency components from 437.5 to 500 Hz. These signal components are obtained by reconstructing the wavelet packet decomposition coefficients, preserving the time-domain information within each frequency band, thus laying the foundation for subsequent feature extraction.

[0033] Next, the singularity index and energy entropy of each signal component are calculated. For each reconstructed signal component, the singularity index is estimated by analyzing the variation of its continuous wavelet transform modulus maxima with scale. Specifically, the signal component undergoes a continuous wavelet transform at multiple scales, with a scale of 2^n. 1 Up to 2 5 The Mexican Hat wavelet is used as the mother wavelet. At each scale, local maxima of the wavelet transform coefficients are identified, forming a modulus maxima line. The amplitude variation of the same modulus maxima line at different scales is tracked, and the relationship between the logarithm of the amplitude and the logarithm of the scale is fitted by linear regression. The negative value of the resulting slope is the Lipschitz exponent estimate for that point. The average of the Lipschitz exponents of all modulus maxima points in the entire signal component is taken as the singularity exponent of that component. This exponent quantifies the abrupt change characteristics of the signal in that frequency band; a larger negative value indicates a more pronounced impulse characteristic.

[0034] Simultaneously, the energy entropy of each signal component is calculated. First, the energy of a single signal component is calculated, which is the sum of the squares of all sampled values ​​of that component. Then, the percentage of that component's energy relative to the total energy of all signal components is calculated, i.e., energy percentage pi = Ei / ΣEj, where Ei is the energy of the i-th component, and ΣEj is the sum of the energies of all components. Finally, the energy percentage of each component is substituted into the Shannon entropy formula to calculate the energy entropy H = -Σ (pi×log2(pi)), where i ranges from 1 to N, and N is the total number of signal components (e.g., 8). The level of energy entropy reflects the degree of concentration or dispersion of interference energy across different frequency bands. A high entropy value indicates a uniform energy distribution and complex interference patterns; a low entropy value indicates that the energy is concentrated in a few frequency bands and the interference patterns are simple.

[0035] After completing the feature calculations for all signal components, the singularity indices and energy entropies of all components are concatenated in frequency band order (from low frequency (3,0) to high frequency (3,7)) to form a 16-dimensional feature vector. This multi-dimensional feature vector represents the mode characteristics of the current interference. Before combining, each feature component should be normalized, for example, by using z-score standardization to make its mean 0 and standard deviation 1, in order to eliminate the influence of differences in feature dimensions and numerical ranges.

[0036] The decision-making phase then begins, where the current interference pattern features are compared with those in historical misjudgment cases to determine whether pose compensation should be initiated. First, the matching degree is calculated between the current interference pattern feature vector and the typical misjudged interference pattern feature vectors stored in the historical misjudgment case library. This historical case library, built over a long production process, stores the interference pattern feature vectors corresponding to each misjudgment and its associated misjudgment type. The matching degree is calculated using Euclidean distance. Specifically, the Euclidean distance between the current feature vector and each historical feature vector in the case library is calculated. Since a smaller distance indicates higher similarity, and subsequent judgments require higher similarity values ​​indicating higher risk, the reciprocal of the calculated minimum Euclidean distance dmin is taken as the final matching degree S = 1 / dmin.

[0037] The matching score is then compared to a dynamic alert threshold. This dynamic alert threshold is not a fixed value but is dynamically set based on historical false positive rates. The system periodically (e.g., every 1000 insulating boards processed) calculates the recent false positive rate (number of false positives / total number of detections) and adjusts the threshold accordingly. A base threshold of 0.3 is set; when the recent false positive rate exceeds 5%, the threshold is lowered to 0.25 to improve system sensitivity; when the false positive rate is below 2%, the threshold is raised to 0.35 to reduce false alarms. This dynamic adjustment mechanism allows the system to adapt to changes in the production environment.

[0038] Finally, the judgment logic is as follows: if the calculated matching degree exceeds the currently set dynamic alert threshold, the current interference mode is determined to be highly similar to the historical misjudgment mode, posing a high risk of misjudgment. Therefore, the pose compensation detection mode needs to be activated for the next insulating plate. If the matching degree does not exceed the threshold, the detection proceeds normally. The entire analysis process intelligently matches real-time signal characteristics with the historical experience database, achieving data-driven, forward-looking compensation decisions, effectively improving the robustness and adaptability of the detection system. In the initial stage when the historical case library is empty or data is insufficient, the system can use a higher default threshold (e.g., 0.5) and gradually switch to dynamic adjustment mode as cases accumulate. Simultaneously, a matching degree upper limit is set, and S is truncated when dmin is too small to avoid numerical overflow.

[0039] S5. When pose compensation is initiated, the defect detection and evaluation threshold for the next insulating plate is dynamically adjusted based on the actual pose offset risk, as follows: When the system determines that pose compensation needs to be initiated based on the aforementioned analysis, it begins the step of dynamically adjusting the defect detection evaluation threshold according to the actual pose deviation risk. This process aims to adaptively adjust the stringency of the detection criteria based on the predicted pose instability. The dynamic adjustment operation is performed by the processing unit in the system, which reads the quantitative assessment result of the actual pose deviation risk calculated in step S3, i.e., the Hausdorff distance value, from the storage unit. This value, expressed in pixels, characterizes the degree of deviation between the actual and expected positions of the insulating plate.

[0040] The first step in dynamic adjustment is to map the quantitative assessment result of the actual pose deviation risk to an adjustment weight coefficient for the defect detection judgment threshold. The adjustment weight coefficient is a dimensionless value between 0 and 1, used to precisely control the magnitude of threshold adjustment. The mapping relationship is implemented using an experimentally validated piecewise linear function. First, two key risk threshold benchmarks need to be set: a low-risk threshold and a high-risk threshold. The low-risk threshold is typically set to a small pixel value, such as 5 pixels, indicating that within this deviation range, the pose deviation is considered to have a negligible impact on detection; the high-risk threshold is set to a larger pixel value, such as 25 pixels, indicating that beyond this deviation, the pose deviation may have a significant impact on the detection result. The specific values ​​of these two thresholds are statistically derived by analyzing the correlation between pose deviation and the final false positive rate in a large amount of historical production data. When the actual pose deviation risk value is less than or equal to the low-risk threshold, the adjustment weight coefficient is mapped to 0, indicating that no adjustment to the benchmark threshold is needed at the current risk level. When the actual pose deviation risk value is greater than or equal to the high-risk threshold, the adjustment weight coefficient is mapped to 1, indicating that maximum threshold compensation is required. When the actual pose deviation risk value falls between the low-risk threshold and the high-risk threshold, the adjustment weight coefficient is calculated through linear interpolation. The formula is: Adjustment weight coefficient = (Actual pose deviation risk value - Low-risk threshold) / (High-risk threshold - Low-risk threshold). This mapping relationship ensures a reasonable and continuous proportional relationship between the adjustment weight coefficient and the actual pose deviation risk value.

[0041] Next, based on the adjusted weighting coefficients and the preset baseline defect detection threshold, the dynamically adjusted defect detection threshold is calculated using linear interpolation. The baseline defect detection threshold is a set of optimal values ​​determined during the production line commissioning phase under ideal and stable conditions through statistical analysis of thousands of qualified and defective samples. This set of thresholds typically includes two key parameters: the defect area threshold and the defect contrast threshold. The defect area threshold is measured in square pixels; for example, the baseline value might be set to 50 square pixels. The defect contrast threshold is measured in gray levels; for example, the baseline value might be set to 30 gray levels. Simultaneously, the system also presets a set of verified compensation threshold upper limits; for example, the defect area compensation threshold might be set to 70 square pixels, and the defect contrast compensation threshold might be set to 25 gray levels. The dynamically adjusted defect detection threshold is calculated using the following linear interpolation formulas: Adjusted defect area threshold = Baseline defect area threshold + Adjusted weighting coefficient × (Defect area compensation threshold - Baseline defect area threshold); Adjusted defect contrast threshold = Baseline defect contrast threshold - Adjusted weighting coefficient × (Baseline defect contrast threshold - Defect contrast compensation threshold). Linear interpolation ensures that as the adjustment weighting coefficients increase from 0 to 1, the area threshold smoothly transitions from a strict baseline value to a more lenient compensation value, while the contrast threshold smoothly transitions from a strict baseline value to a more sensitive compensation value. This reverse adjustment strategy is based on a deep understanding of the impact of pose offset: offset may cause changes in the projected area of ​​defects, thus requiring a relaxation of area restrictions; at the same time, it may cause edge blurring and a decrease in contrast, thus requiring a reduction in contrast requirements.

[0042] Finally, the dynamically adjusted defect detection threshold is immediately configured in the image processing unit for defect detection of the next insulating plate. To ensure system stability under extreme conditions, safety boundary limits are set for the adjusted thresholds; for example, the adjusted defect area threshold is not allowed to exceed 100 square pixels, and the adjusted defect contrast threshold is not allowed to be lower than 20 gray levels. These boundary values ​​are determined based on the physical characteristics of the image sensor and the effective range of the detection algorithm. The entire dynamic adjustment mechanism transforms quantified pose risks into specific detection parameter adjustments, enabling the system to possess context awareness and effectively cope with dynamic interference factors in the production process while ensuring detection reliability. After completing the threshold adjustment, the processing unit outputs a ready signal to the control logic, indicating that appropriate evaluation criteria have been configured for subsequent detection tasks.

[0043] S6. Based on the adjusted defect detection evaluation threshold, perform defect detection on the next insulation board, and control the sorting execution mechanism to perform sorting actions according to the detection results. This is implemented as follows: After adjusting the dynamic defect detection threshold for the next insulating board, the system immediately executes the final step of defect detection and sorting control based on this threshold. This process begins with acquiring the actual detection image of the next insulating board in the detection area. The acquisition of the actual detection image is performed by an industrial camera deployed at a fixed position in the detection area. This industrial camera is the same device or model as the camera used to acquire the initial position image to ensure consistent imaging characteristics. When the next insulating board moves on the conveyor belt and completely covers the photoelectric sensor beam in the detection area, the photoelectric sensor generates a trigger signal. This signal, delayed for a specific time (e.g., 50 milliseconds calculated based on the conveyor belt speed), triggers the industrial camera to expose and acquire a high-resolution digital image covering the entire surface of the insulating board—the actual detection image. All parameters of the industrial camera, including 5-megapixel resolution, 200-microsecond exposure time, and a 2x gain, are strictly consistent with the settings used when acquiring the initial position image to eliminate detection errors introduced by changes in imaging conditions.

[0044] The image processing unit performs defect feature analysis on the acquired actual inspection images. The analysis process begins with image preprocessing: converting the color image to grayscale; performing convolution on the grayscale image using a 5×5 pixel Gaussian filter with a standard deviation of 1.0 to suppress noise; and applying an adaptive histogram equalization algorithm with contrast constraints to enhance local image contrast, where the cropping constraint is set to 0.02 and the tile grid size is set to 10×10 pixels. After preprocessing, the defect region identification stage begins: an adaptive threshold segmentation algorithm (using Gaussian weighted average with a neighborhood size of 31×31 pixels) is used to binarize the image; morphological opening operations are performed on the binary image (using 3×3 pixel rectangular structuring elements) to remove salt-and-pepper noise and small irrelevant points; and a connected component scanning algorithm is used to mark all potential defect regions, recording the coordinates of the bounding rectangle of each region.

[0045] For each identified candidate defect region, two key feature parameters are calculated. The size of the defect feature is quantified by counting the number of all foreground pixels within the connected region, expressed in square pixels. The contrast of the defect feature is calculated as follows: First, the average grayscale value of all pixels within the candidate defect region is calculated; then, based on the bounding rectangle of this region, a ring-shaped region is extended outward by 5 pixels to form the background region, and the average grayscale value of all pixels within this ring-shaped region is calculated; finally, the defect contrast is defined as the absolute value of the difference between the average grayscale value within the region and the average grayscale value of the background region, expressed in gray levels. These calculations generate a set of precise feature data (size value, contrast value) for each candidate defect region.

[0046] Next, the size and contrast values ​​of each candidate defect region are rigorously compared with the dynamically adjusted defect detection threshold. The dynamically adjusted defect detection threshold comprises two specific values: the dynamically adjusted defect area threshold (denoted as Afinal, unit: square pixels) calculated in step S5 and the dynamically adjusted defect contrast threshold (denoted as Cfinal, unit: gray level). The comparison logic is decisive: for a candidate defect region, it is only ultimately determined to be a real defect if its size value is greater than Afinal and its contrast value is greater than Cfinal. This is a strict AND logic relationship. If multiple candidate defect regions exist in an actual detection image, the image processing unit will independently perform the above dual-condition judgment on each region. As long as one region is determined to be a real defect, the entire insulation board is determined to be a defective product.

[0047] Based on the above detection results, the system controls the sorting actuator to perform the sorting action. The judgment logic is as follows: the system will generate a sorting instruction only if at least one defect feature exists in the actual detection image, and its size and contrast simultaneously exceed (i.e., are greater than) the dynamically adjusted defect detection judgment thresholds Afinal and Cfinal. The sorting instruction is a digital signal packet containing the insulation board serial number, defect type code, and current timestamp. This instruction is sent in real time to the programmable logic controller (PLC) of the sorting actuator via the industrial Ethernet protocol. The sorting actuator is typically a pneumatic nozzle or electromagnetic push rod installed on one side of the conveyor belt. After receiving the sorting instruction, the PLC continuously reads the encoder pulse signal that rotates synchronously with the conveyor belt. When the pulse count indicates that the insulation board has moved to the physical action point of the sorting actuator, the controller immediately drives the solenoid valve to open the air passage (or connects the electromagnet coil), causing the actuator to produce a rapid action, accurately pushing the unqualified insulation board into the designated waste collection device, thereby completing the removal action.

[0048] If, after comprehensive comparison, none of the candidate defective areas simultaneously meet the threshold conditions for size and contrast (i.e., no area has a size greater than Afinal and a contrast greater than Cfinal), the system determines the insulation board to be a qualified product. In this case, the system will not generate any sorting instructions. The sorting execution mechanism remains stationary, allowing the insulation board to smoothly pass through the detection area and sorting point, continuing to be conveyed on the production line to the next process (such as the packaging station). The entire detection, judgment, and sorting decision-making process must be completed within a preset cycle time, for example, the total time from image acquisition to the issuance of the sorting instruction should not exceed 500 milliseconds, to ensure the continuous operation of the production line. The system database records the complete data chain for each detection, including the original image, feature parameters of all candidate areas, the dynamic threshold used, the final judgment result, and whether a sorting action was triggered. This data is used for production quality statistics, system performance evaluation, and continuous algorithm optimization.

[0049] Example 2: Figure 2 A schematic diagram of a visual inspection and sorting system for defects in insulating board processing is provided according to the present invention. The system includes: The parameter acquisition module is used to record the physical characteristic parameters of the corresponding action after the sorting execution mechanism completes the action on the current insulating plate, and at the same time acquire the initial position image of the next insulating plate when it enters the detection area. The trend prediction module is used to query the pre-established interference mapping relationship based on the physical characteristic parameters and the position of the plate current sequence of the next insulating plate, and predict the dynamic interference trend of the next insulating plate. The risk assessment module is used to comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference based on dynamic interference trends and initial position images. The compensation judgment module is used to obtain the mode characteristics of the current interference by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; and to determine whether to initiate pose compensation for the next insulating plate by performing similarity matching analysis between the mode characteristics and the interference modes in historical misjudgment cases. The risk adjustment module is used to dynamically adjust the defect detection and evaluation threshold for the next insulating board based on the actual pose offset risk when pose compensation is initiated. The action execution module is used to perform defect detection on the next insulation board based on the adjusted defect detection evaluation threshold, and control the sorting execution mechanism to perform sorting actions according to the detection results.

[0050] All calculations involved in the embodiments are dimensionless numerical calculations, and the preset parameters and thresholds in the calculations are set by those skilled in the art according to the actual situation.

[0051] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0052] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0053] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0054] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0055] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0056] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0057] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0058] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0059] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for visual inspection and sorting of defects in insulating board processing, characterized in that, include: S1. After the sorting execution mechanism completes the action on the current insulating plate, record the physical characteristic parameters of the corresponding action, and at the same time obtain the initial position image when the next insulating plate enters the detection area. S2. Based on the physical characteristic parameters and the position of the plate current sequence of the next insulating plate, query the pre-established interference mapping relationship and predict the dynamic interference trend of the next insulating plate. S3. Based on the dynamic interference trend and the initial position image, comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference; S4. The mode characteristics of the current interference are obtained by performing multi-scale time-frequency characteristic analysis on the physical characteristic parameters; By performing similarity matching analysis between the pattern features and the interference patterns in historical misjudgment cases, it is determined whether to initiate pose compensation for the next insulating plate. S5. When pose compensation is activated, the defect detection and evaluation threshold for the next insulating board is dynamically adjusted based on the actual pose deviation risk. S6. Based on the adjusted defect detection evaluation threshold, perform defect detection on the next insulation board, and control the sorting execution mechanism to perform sorting actions according to the detection results.

2. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, Recording the physical characteristic parameters of the corresponding actions includes collecting pressure change data and drive current data generated by the sorting actuator during operation; Acquiring the initial position image when the next insulating plate enters the detection area involves acquiring an image containing the corresponding next insulating plate using an industrial camera under the action of a trigger signal, and recording the timestamp corresponding to the acquisition of the initial position image. The trigger signal is associated with the event of the sorting actuator completing its action.

3. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, The query of pre-established interference mapping relationships includes: combining sorting action types with board flow sequence positions, wherein the sorting action types are classified based on the time-domain statistical characteristics of physical characteristic parameters; The pre-established interference mapping relationship is constructed by analyzing the differential features of the initial position image of the next insulating board relative to the interference-free reference image when it arrives at the detection area based on its board flow sequence position after a specific sorting action is completed in historical data. Predicting dynamic interference trends includes outputting and combining trend vectors that characterize the intensity and direction of interference.

4. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, The comprehensive assessment of actual pose deviation risk includes: converting the predicted deviation direction and deviation amount contained in the dynamic disturbance trend into a virtual offset field for the initial position image; Extract the actual edge contour of the insulating plate from the initial position image; By comparing the geometric differences between the actual edge contour and the expected edge contour after virtual offset field transformation, a quantitative assessment result of the actual pose offset risk is generated.

5. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, The mode characteristics of the current interference are obtained by performing multi-scale time-frequency feature analysis on the physical characteristic parameters, including: Wavelet packet decomposition is performed on the recorded physical characteristic parameters to obtain signal components in different frequency bands; Calculate the singularity index and energy entropy of each signal component to form the pattern characteristics of the current interference; By performing similarity matching analysis between pattern features and interference patterns in historical misjudgment cases, the determination of whether to initiate pose compensation for the next insulating plate includes: The matching degree is calculated between the current interference pattern features and the typical misjudged interference pattern features that are constructed in the same way and stored in the historical misjudgment case library; The matching degree is compared with a dynamic alert threshold set based on historical false positive rate statistics; If the matching degree exceeds the dynamic alert threshold, it is determined that the pose compensation of the next insulating plate needs to be initiated.

6. The method for visual inspection and sorting of defects in insulating board processing according to claim 5, characterized in that, Calculating the singularity index and energy entropy of each signal component includes: For each signal component, the singularity index is estimated by analyzing the variation of its wavelet transform modulus maxima with scale. At the same time, the proportion of the energy of the signal component in the total energy is calculated and substituted into the Shannon entropy formula to calculate the energy entropy; The singularity index and energy entropy of all signal components are combined in frequency band order to form a multidimensional feature vector, which constitutes the mode characteristics of the current interference.

7. The method for visual inspection and sorting of defects in insulating board processing according to claim 5, characterized in that, The matching degree calculation between the current interference pattern features and the typical misjudged interference pattern features constructed in the same way stored in the historical misjudgment case database includes: Calculate the Euclidean distance between the current interference pattern feature vector and one or more typical misjudged interference pattern feature vectors pre-stored in the historical misjudgment case library; The reciprocal of the calculated minimum Euclidean distance is used as the matching degree to characterize the similarity between the two.

8. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, The defect detection and evaluation thresholds for the next insulating plate are dynamically adjusted based on the actual pose deviation risk, including: The quantitative assessment results of the actual pose deviation risk are mapped to the adjusted weighting coefficients for the defect detection judgment threshold; Based on the adjusted weighting coefficients and the preset benchmark defect detection threshold, the dynamically adjusted defect detection threshold is obtained through linear interpolation. The dynamically adjusted defect detection threshold is used in the defect detection process of the next insulation board.

9. The method for visual inspection and sorting of defects in insulating board processing according to claim 1, characterized in that, Defect detection of the next insulation board based on the adjusted defect detection evaluation threshold includes: acquiring the actual detection image of the next insulation board in the detection area; comparing the size and contrast information of the defect features in the actual detection image with the dynamically adjusted defect detection evaluation threshold; The sorting action controlled by the sorting execution mechanism based on the detection results includes: when the size and contrast of the defect feature simultaneously exceed the dynamically adjusted defect detection and evaluation threshold, a sorting instruction is generated and sent to the sorting execution mechanism to remove the corresponding next insulating board; otherwise, the next insulating board is allowed to continue to be conveyed on the assembly line.

10. A visual inspection and sorting system for defects in insulating board processing, used to implement the visual inspection and sorting method for defects in insulating board processing as described in any one of claims 1-9, characterized in that, include: The parameter acquisition module is used to record the physical characteristic parameters of the corresponding action after the sorting execution mechanism completes the action on the current insulating plate, and at the same time acquire the initial position image of the next insulating plate when it enters the detection area. The trend prediction module is used to query the pre-established interference mapping relationship based on the physical characteristic parameters and the position of the plate current sequence of the next insulating plate, and predict the dynamic interference trend of the next insulating plate. The risk assessment module is used to comprehensively assess the risk of actual pose deviation of the next insulating plate due to interference based on dynamic interference trends and initial position images. The compensation judgment module is used to obtain the mode characteristics of the current interference by performing multi-scale time-frequency feature analysis on the physical characteristic parameters; By performing similarity matching analysis between the pattern features and the interference patterns in historical misjudgment cases, it is determined whether to initiate pose compensation for the next insulating plate. The risk adjustment module is used to dynamically adjust the defect detection and evaluation threshold for the next insulating board based on the actual pose offset risk when pose compensation is initiated. The action execution module is used to perform defect detection on the next insulation board based on the adjusted defect detection evaluation threshold, and control the sorting execution mechanism to perform sorting actions according to the detection results.

Images