A method and system for detecting the quality of a glass fiber plate

Abstract

The application relates to the technical field of image recognition, and discloses a glass fiber plate production quality detection method and system, wherein the method comprises the following steps: acquiring a continuous infrared thermal image sequence through an infrared thermal imaging device; constructing a temperature difference image set; identifying a graph element area with abnormal diffusion characteristics; further performing fitting enhancement and saliency detection on the abnormal graph element area; and performing defect classification and quality judgment based on a region feature vector. Compared with the prior art, in the case that the surface texture of the glass fiber plate is complex and thermal pressing disturbance exists, especially in the case that the thermal diffusion behaviors of dry filaments, degumming and air bubble defects are similar and the false detection rate of conventional algorithms is high, the technical problem that high-precision typing judgment cannot be achieved is solved. Due to the introduction of a multi-channel thermal diffusion consistency analysis mechanism and parallel discrimination logic based on a feature vector, the present application realizes infrared graph element abnormal enhancement detection and defect classification and recognition in a complex background, and the accuracy of glass fiber plate production quality detection is improved.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to a method and system for quality inspection in the production of fiberglass boards. Background Technology

[0002] Currently, fiberglass boards, widely used in automotive interiors, building insulation, wind turbine blades, and other fields, rely heavily on the uniformity of hot pressing and the consistency of sizing agent distribution during the manufacturing process for performance stability and product quality. During the hot pressing and cooling stage of fiberglass boards, defects such as "dry fibers," "delamination," or "cavitation" may occur due to factors such as uneven sizing agent coating, disordered fiber arrangement, and deviations in hot pressing temperature control. These defects not only affect structural strength and thermal insulation performance but may also lead to subsequent interlayer separation, warping, and even safety issues.

[0003] Quality inspection in industrial settings primarily employs methods such as manual visual inspection, static infrared thermal imaging, or ultrasonic penetration. However, manual inspection is inefficient and highly susceptible to subjective influences, easily affected by factors such as lighting, angle, and fatigue. While ultrasonic testing possesses penetrating capabilities, the equipment is expensive and requires high coupling with the workpiece surface, making it difficult to rapidly and non-contactly inspect large batches of products. Traditional infrared thermal imaging technology often relies on threshold judgments based on single-frame temperature images, failing to fully utilize the dynamic characteristics of the heat diffusion process and making it difficult to accurately extract subtle internal thermal response anomalies from complex background temperature evolution.

[0004] Furthermore, due to limitations imposed by the complex surface texture of fiberglass boards, uneven local cooling conditions, and process disturbances, traditional image recognition-based infrared detection methods exhibit poor stability and robustness. For instance, in the presence of a slight temperature gradient on the surface, traditional methods may misidentify normal areas as "abnormal heating," leading to misjudgments. Moreover, when cavitation bubbles are located in the middle or bottom layers, their thermal signal propagation paths are complex, making it difficult to discern significant surface thermal map differences in a short time, and thus easily overlooked by traditional methods.

[0005] Therefore, existing technologies cannot fully meet the real-time, high-precision detection requirements for deep-seated minute defects in the high-efficiency production of fiberglass boards. There is an urgent need to propose a novel quality inspection method that integrates time-series thermal diffusion behavior modeling, image recognition, and embedded discrimination logic. This method should possess stable, rapid, non-contact, and embeddable defect identification capabilities even under complex operating conditions, particularly in the real-time monitoring of three types of defects: "dry fibers," "delamination," and "cavitation." This would achieve a complete link from thermal response mechanism analysis to classification and control, thereby improving the intelligent detection capabilities and product consistency control level of the entire production line. Summary of the Invention

[0006] To address the aforementioned technical shortcomings, the purpose of this invention is to propose a method for inspecting the production quality of glass fiberboard. This method aims to solve the technical problem that existing technologies cannot achieve high-precision classification and determination when the surface texture of glass fiberboard is complex and subject to thermal pressure disturbance, especially when the thermal diffusion behavior of dry fibers, degumming, and cavitation defects is similar and conventional algorithms have a high false detection rate.

[0007] To solve the above-mentioned technical problems, the present invention adopts the following technical solution: The present invention provides a method for quality inspection in the production of glass fiberboard.

[0008] The method for testing the production quality of the fiberglass board includes:

[0009] Step S10: During the cooling stage after the glass fiber board completes the hot pressing process, a continuous infrared thermal image sequence S of the glass fiber board surface is acquired using a preset infrared thermal imaging device.

[0010] Step S20: Based on the infrared thermal image sequence S, execute the nonlinear median filtering algorithm based on spatial preservation and the background temperature modeling mechanism guided by local response in sequence, output the background temperature T(t) of the current frame, and construct the background temperature evolution curve; perform background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background temperature evolution curve, and output the temperature difference image set ΔT(x,y,t).

[0011] Step S30: Based on the temperature difference image set ΔT(x,y,t), a multi-channel thermal diffusion model is constructed using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of abnormal primitive regions Ω.

[0012] Step S40: Perform an image processing mechanism based on a combination of fitting enhancement and saliency detection on the abnormal primitive region set Ω, and output the final defect primitive mask M(x,y);

[0013] Step S50: Based on the output final defect primitive mask M(x,y), execute the preset defect identification and quality judgment control logic based on feature vector driving in the preset FPGA control platform, and finally output the quality inspection result.

[0014] Preferably, in step S10, the infrared thermal image sequence S includes multiple two-dimensional infrared thermal image frames arranged in chronological order. Each pixel position (x, y) in each two-dimensional infrared thermal image frame corresponds to a temperature value T(x, y, t) at a time t; where x represents the horizontal coordinate of the pixel position and y represents the vertical coordinate of the pixel position.

[0015] Preferably, in step S20, the steps of sequentially executing a spatially preserved nonlinear median filtering algorithm and a local response-guided background temperature modeling mechanism based on the infrared thermal image sequence S, outputting the current frame background temperature T(t), and constructing a background temperature evolution curve; and performing background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background temperature evolution curve, and outputting the temperature difference image set ΔT(x,y,t), specifically include:

[0016] Step S201: Define an adaptive neighborhood window W(x,y) for each pixel position (x,y) in the infrared thermal image sequence S. The size of the adaptive neighborhood window W(x,y) is determined according to the temperature change rate of the pixel position (x,y) in the frames before and after it. ,in, The default base window size; The gradient is in the time direction; This is the window adjustment coefficient;

[0017] Median filtering is performed on the temperature value T(x,y,t) at time t for each pixel position (x,y) in the infrared thermal image sequence S within an adaptive neighborhood window W(x,y), and the filtered image frame is output. ;

[0018] Step S202: Filter the image frame Extract the time dimension sequence, use linear least squares fitting to construct the background temperature evolution curve based on the time dimension sequence, and output the background temperature T(t) of the current frame;

[0019] Step S203: Calculate the background residual standard deviation of each pixel based on the background temperature evolution curve, and perform background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background residual standard deviation, and output the temperature difference image set ΔT(x,y,t).

[0020] Preferably, in step S30, the step of constructing a multi-channel thermal diffusion model based on the temperature difference image set ΔT(x,y,t) using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation, and outputting the set of anomalous primitive regions Ω by the multi-channel thermal diffusion model, specifically includes:

[0021] Step S301: Construct a multi-channel heat diffusion model. The multi-channel heat diffusion model divides the infrared thermal image sequence S into multiple feature channel segments based on the temperature difference image set ΔT(x,y,t), forming a multi-channel set. The multi-channel set includes the first cooling segment. Second cooling section and the third cooling section Among them, the first cooling section Surface response used to indicate initial cooling; second cooling section Used to represent shallow heat conduction in intermediate cooling; third cooling section Used to represent the deep hysteresis response of late-stage cooling;

[0022] Step S302: The multi-channel thermal diffusion model is applied to the k-th characteristic channel segment in the multi-channel set. Perform the thermal diffusion consistency score calculation process for multi-channel verification and output anomaly candidate points;

[0023] Step S303: If a pixel location is repeatedly marked as an anomalous candidate point in two or more non-adjacent feature channel segments, the multi-channel thermal diffusion model determines that the pixel location is a thermal diffusion anomalous primitive and finally outputs the set of anomalous primitive regions Ω.

[0024] Preferably, in step S30, the multi-channel heat diffusion model is applied to the k-th characteristic channel segment in the multi-channel set. The process of calculating the thermal diffusion consistency score for multi-channel verification and outputting candidate anomalies includes the following steps:

[0025] In the characteristic channel segment Fit an intrachannel temperature trend curve and calculate the first residual between the temperature value T(x,y,t) in the infrared thermogram sequence S at time t and the corresponding channel temperature value in the intrachannel temperature trend curve.

[0026] Simultaneously, the second residual between the temperature value T(x,y,t) in the infrared thermal image sequence S at time t and the neighboring pixels of the pixel position (x,y) is calculated separately.

[0027] The thermal diffusion consistency score at pixel position (x, y) at time t is evaluated using the dual-channel residual consistency distribution assumption method based on the first and second residuals; pixel positions with diffusion consistency scores less than the preset consistency score threshold are marked as abnormal candidate points.

[0028] Preferably, step S40, which involves performing an image processing mechanism based on a combination of fitting enhancement and saliency detection on the set of abnormal primitive regions Ω to output the final defect primitive mask M(x,y), specifically includes:

[0029] Step S401: Obtain the abnormal temperature difference image set corresponding to the pixel position (x,y) in the abnormal primitive region set Ω, and perform logarithmic domain polynomial fitting on the abnormal temperature difference image set within a preset time window to obtain the thermal diffusion rate image G(x,y).

[0030] Step S402: Based on the obtained thermal diffusion rate image G(x,y), construct a neighborhood window centered on the pixel position (x,y), calculate the local mean μ(x,y) and standard deviation σ(x,y), perform saliency detection processing based on the local mean μ(x,y) and standard deviation σ(x,y), and output the saliency score S(x,y); aggregate the pixel positions where the saliency score S(x,y) is greater than the preset saliency score threshold to generate a saliency primitive mask. ;

[0031] Step S403: Mask the saliency primitives Perform primitive-level cross-referencing with the abnormal primitive region set Ω. If a pixel position (x, y) simultaneously satisfies: If the result is positive, it is marked as the final defect primitive, and the final defect primitive mask M(x,y) is output.

[0032] Preferably, in step S50, based on the output final defect primitive mask M(x,y), the preset defect identification and quality judgment control logic driven by feature vectors is executed in a preset FPGA control platform to finally output the quality inspection result. This step specifically includes:

[0033] Step S501: In the preset FPGA control platform, perform maximum temperature difference extraction and time response analysis on each connected region i in the final defect primitive mask M(x,y) to construct its five-dimensional feature vector. , ,in, The maximum temperature contrast within connected region i; Thermal response polarity, A value of +1 indicates a heating response, including cavitation response and degumming response; A value of -1 indicates a cooling response, including the dry wire response; For the response start time; Let i be the area of ​​the connected region. The regional compactness index is defined as follows: ,in, Let i be the perimeter of the connected region i, and let i be the region compactness index. The closer it is to 1, the closer it is to a circle;

[0034] Step S502: For the five-dimensional feature vector Perform the following multi-condition rule judgment and output the corresponding defect classification control code. ;

[0035]

[0036] in, and These are the preset negative temperature contrast threshold and positive temperature contrast threshold, respectively; and These are the preset early response judgment time threshold and the late response judgment time threshold, respectively; and These are the preset lower limit and upper limit for determining the area of ​​the region, respectively. and These are the preset lower limit and upper limit for shape compactness determination, respectively; This indicates a "logical AND" relationship, used to describe a condition that must be met simultaneously for a statement to be valid.

[0037] Step S503: Apply the defect classification control code obtained from the determination. The input is converted into a binary control word and sent to the preset host computer, which then outputs the final quality detection result.

[0038] The present invention also provides a glass fiberboard production quality inspection system comprising:

[0039] The infrared image acquisition module is used to acquire a continuous infrared thermal image sequence S on the surface of the glass fiber board through a preset infrared thermal imaging device during the cooling stage after the hot pressing process of the glass fiber board is completed.

[0040] The background modeling and subtraction module is used to sequentially execute a nonlinear median filtering algorithm based on spatial preservation and a background temperature modeling mechanism guided by local response based on the infrared thermal image sequence S, output the background temperature T(t) of the current frame, and construct the background temperature evolution curve; according to the background temperature evolution curve, the infrared thermal image sequence S is subjected to background subtraction processing based on normalized residual compensation, and the temperature difference image set ΔT(x,y,t) is output.

[0041] The multi-channel thermal diffusion anomaly extraction module is used to construct a multi-channel thermal diffusion model based on the temperature difference image set ΔT(x,y,t) using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of anomalous primitive regions Ω.

[0042] The defect saliency detection module is used to perform an image processing mechanism based on a combination of fitting enhancement and saliency detection on the set of abnormal primitive regions Ω, and output the final defect primitive mask M(x,y);

[0043] The FPGA identification and output control module is used to execute preset feature vector-driven defect identification and quality judgment control logic on a preset FPGA control platform based on the output final defect primitive mask M(x,y), and finally output the quality inspection result.

[0044] The present invention also provides a glass fiberboard production quality inspection device, comprising: a memory, a processor, and a glass fiberboard production quality inspection program stored in the memory and executable on the processor, wherein the glass fiberboard production quality inspection program implements a glass fiberboard production quality inspection method when executed by the processor.

[0045] The present invention also provides a computer program product, including a fiberglass board production quality inspection program, which, when executed by a processor, implements the fiberglass board production quality inspection method.

[0046] The beneficial effects of this invention are as follows: By introducing a multi-channel thermal diffusion consistency analysis mechanism based on image recognition, this invention can effectively extract abnormal temperature difference areas during the cooling stage after hot pressing of glass fiber boards, significantly improving the early identification ability of defects such as dry fibers, degumming and cavitation, and solving the problem of high false detection rate of traditional image recognition methods under complex surface texture and thermal field disturbance conditions.

[0047] This invention further combines the image recognition processing mechanism of fitting enhancement and saliency detection, and builds a parallel discrimination logic based on feature vector driving on the FPGA control platform, realizing accurate classification of defect types and real-time quality judgment, with good online detection capability and industrial deployment adaptability. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating the first embodiment of a method for quality inspection in the production of glass fiberboard according to the present invention.

[0050] Figure 2 This is a schematic diagram comparing the consistency scores of normal and defective areas in the first embodiment of a glass fiberboard production quality inspection method of the present invention.

[0051] Figure 3 This is a schematic diagram comparing the thermal diffusion response curves of the normal area and the defective area in the first embodiment of a glass fiberboard production quality inspection method of the present invention.

[0052] Figure 4 This is a schematic diagram of the equipment for a glass fiberboard production quality inspection method according to the present invention. Detailed Implementation

[0053] 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.

[0054] Example 1: As Figure 1 The diagram shown is a flowchart of the first embodiment of the glass fiberboard production quality inspection method of the present invention, which presents the first embodiment of the glass fiberboard production quality inspection method of the present invention.

[0055] In the first embodiment, the method for inspecting the production quality of the fiberglass board includes:

[0056] Step S10: During the cooling stage after the glass fiber board completes the hot pressing process, a continuous infrared thermal image sequence S of the glass fiber board surface is acquired using a preset infrared thermal imaging device.

[0057] It should be noted that the "infrared thermal image sequence S" refers to a continuous set of infrared thermal imaging data of the glass fiber board surface during the cooling process after hot pressing, essentially a three-dimensional spatiotemporal temperature matrix. Acquiring the thermal image sequence relies on an infrared thermal imaging device positioned at the cooling channel exit or above the stationary workstation. This device includes an uncooled focal plane array with a long-wave infrared sensor chip (e.g., 8~14μm band), a NETD (temperature difference resolution) below 0.05℃, a high-speed image acquisition module supporting frame rates above 30fps, and a lens system with a surface-enhanced heat reflection shielding structure to suppress background thermal radiation interference. The image sequence S should include high-frequency temperature distribution changes throughout the entire natural or forced cooling process of the glass fiber board, and simultaneously record metadata such as the acquisition timestamp, ambient reference temperature, and board number.

[0058] Understandably, this step aims to establish complete time-series data on the thermal diffusion process, allowing the subtle thermal disturbances induced by potential defects within the fiberglass board to be significantly amplified in subsequent analysis. The infrared thermogram sequence S not only captures the transient distribution of the temperature field on the board surface but also implicitly reveals differences in thermal response characteristics caused by factors such as thickness variations in different areas, uneven adhesive content, or abnormal fiber distribution. By acquiring this sequence stably and continuously, the spatiotemporal resolution of subsequent temperature difference residual analysis and anomaly detection can be significantly improved, thus providing a reliable foundation for thermal diffusion modeling and anomaly localization in image recognition algorithms.

[0059] It should be understood that, compared to traditional manual sampling methods that rely solely on visible surface defects or low-resolution infrared snapshots, this invention effectively avoids problems such as incomplete capture of initial thermal traces of defects and inability to reproduce dynamic diffusion paths through continuous, non-contact, high-frame-rate infrared thermal image sequence acquisition. Traditional thermal imaging typically captures only a single frame image at a fixed time cross-section, easily missing early response signals of internal defects such as cavitation and delamination. This method, however, performs "full-process tracking imaging" throughout the entire cooling cycle, combined with a multi-frame temperature difference evolution calculation mechanism, maintaining the ability to completely capture the thermal response of minute anomaly areas even in high-noise environments, thereby achieving higher defect visualization and discernibility during image recognition.

[0060] For example, taking a batch of fiberglass boards with a thickness of 15mm and dimensions of 1200mm × 2400mm as an example, the thermal imaging device is activated within 3 seconds of exiting the hot-pressing section, with an image acquisition frequency of 40 frames / second and a total acquisition time of 10 seconds. Time-series temperature trajectory analysis of the acquired 400-frame thermal image sequence reveals that the temperature in normal areas stabilizes after 5 seconds, while areas containing dry filament defects exhibit significant localized cooling lag under the same cooling conditions, with a maximum residual of 2.1℃; areas containing cavitation defects experience a brief abnormal temperature rise peak within 3 seconds, with a thermal disturbance duration of 0.8 seconds. These phenomena are difficult to distinguish in a single static image, but through the acquisition and tracking of continuous infrared image sequences in this step, the temporal evolution of defects becomes a quantifiable feature, providing a stable input for the subsequent image recognition algorithm to construct a saliency score.

[0061] Step S20: Based on the infrared thermal image sequence S, execute the nonlinear median filtering algorithm based on spatial preservation and the background temperature modeling mechanism guided by local response in sequence, output the background temperature T(t) of the current frame, and construct the background temperature evolution curve; perform background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background temperature evolution curve, and output the temperature difference image set ΔT(x,y,t).

[0062] It should be noted that the "spatial-preserving nonlinear median filtering algorithm" refers to removing high-frequency interference caused by environmental background radiation, camera noise, or lens thermal drift in infrared thermal images while preserving image edges and local structural details. Its core is the introduction of nonlinear median selection logic based on adaptive weight control of pixel grayscale histograms, which can effectively suppress isolated point noise without blurring edge contours. The "local response-guided background temperature modeling mechanism" refers to constructing a dynamic basis model based on slowly changing regions of thermal diffusion response in the image, extracting a time series representing the overall plate reference temperature change trend, and thus dynamically constructing a background temperature map in each frame to remove interference from macroscopic cooling trends on defect identification. The temperature difference image set refers to the pure local abnormal response temperature residual map set obtained after removing the dynamic background from the infrared thermal image sequence S.

[0063] Understandably, the goal of this step is to model and subtract the "background signal" of large-area temperature changes caused by the overall cooling of the fiberglass board in the infrared thermal image, retaining only the "abnormal signal" corresponding to localized non-uniform cooling behavior, thereby significantly enhancing the thermal trace of defects. The introduction of the background temperature evolution curve allows for relative deviation analysis of the temperature value of each pixel based on a dynamic benchmark, greatly improving the robustness of subsequent thermal diffusion consistency judgments. The normalized residual compensation mechanism further suppresses the global offset problem caused by differences in shooting batches or equipment parameters, ensuring the consistency and comparability of the temperature difference image set across batch detection tasks.

[0064] It should be understood that, compared to traditional image recognition methods that directly perform threshold segmentation or texture analysis on the original thermal image, the residual map extraction strategy based on background evolution modeling introduced in this invention effectively solves the problem of defect thermal disturbances being submerged due to the superposition of the overall cooling trend of the board material. Traditional methods often misjudge areas with high brightness or abrupt texture changes in the image as defects, especially in strong convection cooling environments where the overall temperature field changes drastically and it is almost impossible to extract weak defect signals. However, through the dynamic baseline model and residual normalization processing established in this step, the real abnormal thermal behavior at any shooting time can be significantly enhanced, allowing the disturbance signals formed by tiny cavities, thin-layer delamination, or dry filament areas to be clearly displayed in complex backgrounds, improving the sensitivity and stability of the image recognition algorithm for deep defects.

[0065] Step S30: Based on the temperature difference image set ΔT(x,y,t), a multi-channel thermal diffusion model is constructed using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of abnormal primitive regions Ω.

[0066] It should be noted that "local temporal residual evolution" refers to constructing a temperature residual change curve for each pixel in the temperature difference image set along the time dimension, and measuring whether the thermal diffusion behavior of that point is consistent with the surrounding area through statistical characteristics such as the initial response delay, response rate, and duration of the curve. "Multi-channel cross-validation" refers to simultaneously extracting thermal diffusion features from multiple physical dimensions (such as temperature response polarity, maximum fluctuation amplitude, curve skewness and slope, etc.), constructing an independent anomaly scoring channel in each dimension, and eliminating spurious anomaly response points through cross-consistency judgment. Finally, the results of each channel are merged into an anomaly primitive region set Ω, which is used to characterize potential dry filament, degumming, or cavitation defect areas.

[0067] Understandably, the main technical advantage of this step lies in combining time-series analysis with a multi-channel thermal diffusion model to solve the problem of traditional single-frame image processing methods failing to accurately identify weak thermal anomaly regions. Thermal diffusion anomalies are the external manifestation of deep defects "disturbing the thermal field diffusion path" during the cooling process, often exhibiting localized response lag, asymmetrical temperature rise and fall, or unstable response. By using the evolution behavior of the time-series residual curve as the basis for judgment and introducing a multi-channel feature space mutual verification mechanism, false alarms caused by material texture, surface reflection, or equipment drift can be significantly reduced, improving the accuracy and stability of anomaly identification.

[0068] It should be understood that, compared with existing methods that rely solely on single-channel temperature anomalies or abrupt changes in edge texture to identify defects, the multi-channel thermal diffusion consistency analysis mechanism introduced in this invention represents a shift in identification logic towards "dynamic behavior" rather than "static features." Traditional solutions struggle to effectively address the superposition of temperature field disturbances caused by factors such as the multi-layered structure of the sheet material and uneven distribution of the wetting agent, often resulting in missed or false detections. However, the "local temporal residual evolution curve" introduced in this step can determine the persistence of anomalies from a time dimension, and "multi-channel feature cross-validation" can verify response consistency from multiple angles, thereby achieving stable revelation of deep defects in multi-frame thermal images, exhibiting stronger interpretability and robustness.

[0069] For example, such as Figure 3As shown, the horizontal axis represents the sample number (a total of 10 sets of infrared thermogram sequences were collected during the cooling stage after hot pressing of glass fiber boards); the vertical axis represents the consistency score, ranging from 0 to 1, indicating the fusion consistency index of the regional thermal diffusion response curve under multi-channel feature dimensions (response delay, maximum fluctuation, polarity skewness, etc.), with lower values ​​indicating a higher probability of anomalies; the dark blue bars represent the mean consistency score of defective areas in each sample group; the light gray bars represent the mean consistency score of normal areas in the corresponding samples. It can be observed that the scores of defective areas are generally distributed between 0.25 and 0.45, exhibiting significant inconsistencies; the scores of normal areas are concentrated between 0.8 and 0.9, showing highly consistent thermal diffusion behavior; a clear "separation band" is formed between the scores of the two types of areas in each sample group, supporting subsequent automatic classification and identification. Figure 4 As shown, the red curve of the defect point exhibits obvious characteristics such as delayed start-up (response lag of about 0.8s), asymmetric fluctuation (slow heating rate and fast cooling rate), and peak shift; while the blue curve of the normal point shows symmetrical, continuous and stable thermal diffusion behavior, without abrupt changes or delays, revealing how deep defects disturb the local thermal diffusion path and leave identifiable abnormal traces in the time dimension.

[0070] Step S40: Perform an image processing mechanism based on fitting enhancement and saliency detection on the abnormal primitive region set Ω, and output the final defect primitive mask M(x,y);

[0071] It should be noted that "fitting enhancement" refers to using a minimum residual fitting mechanism between multi-frame thermal diffusion response curves and a preset reference curve within the anomalous primitive region Ω to restore and enhance the contours of anomalous regions with weak signals or fragmented shapes. Specific methods include least-squares fitting of the thermal diffusion path and parabolic approximation fitting of the response time window, used to restore anomalous edge loss caused by noise masking or texture interference. "Saliency detection" refers to performing region aggregation analysis based on feature sparsity and response contrast in the image spatial domain, using global sparse saliency mapping and local contrast enhancement mechanisms to detect salient defect regions, enhancing the visual separability of weak anomalous points. Finally, the fitted and enhanced thermal diffusion image is fused with the saliency detection results, and combined with region growing and morphological processing to generate a final defect primitive mask. This mask is used to mark the spatial distribution of suspected dry filament, debonding, or void defects in the fiberglass board.

[0072] Understandably, the main technical effect of this step lies in enhancing the morphological closure and identifiability of abnormal regions through the combined mechanism of "temporal behavior recovery" and "spatial saliency extraction," particularly demonstrating significant repair and extraction capabilities for defect regions with blurred boundaries, small areas, or broken thermal diffusion paths. Through bidirectional cross-constraints between saliency mapping and fitting enhancement, interference from non-defect factors such as background temperature fluctuations, random noise, and false triggering of edge textures is effectively suppressed.

[0073] It should be understood that, compared to existing technologies that rely solely on pixel-level thermal difference thresholding or edge detection algorithms to generate defect masks, the "fitting enhancement + saliency detection" mechanism proposed in this invention provides a more robust multi-source image fusion path: the fitting enhancement mechanism allows low signal-to-noise ratio anomalous regions to still be restored to continuous closed defect contours; the saliency detection mechanism improves the defect region's resistance to background interference; and the image processing mechanism that fuses these two mechanisms can significantly reduce false detection and false negative rates, making it particularly suitable for detecting defects on fiberglass board surfaces with complex textured backgrounds and drastic local temperature fluctuations. Therefore, this step outperforms traditional single-channel detection or static image thresholding methods in terms of the completeness, accuracy, and stability of defect identification.

[0074] Step S50: Based on the output final defect primitive mask M(x,y), execute the preset defect identification and quality judgment control logic based on feature vector driving in the preset FPGA control platform, and finally output the quality inspection result.

[0075] Understandably, this step, through a feature vector-based structured defect description mechanism, achieves rapid transformation from image masking to structured semantic decision-making, thereby realizing integrated processing capabilities from "visual recognition" to "logical judgment" within the FPGA. Unlike traditional methods that rely on GPUs to execute image analysis logic, this invention encapsulates a series of modules such as temperature difference calculation, mask generation, region statistics, and judgment classification into reconfigurable parallel logic circuits. This enables defect identification and output judgment for images with resolutions greater than 1024×1024 with a 50ms delay, meeting the intra-frame processing requirements of high-speed lamination production lines with high-speed board throughput.

[0076] It should be understood that, compared with existing technologies that require post-processing via host computer image analysis software (such as OpenCV or Python scripts), the "FPGA-built-in defect identification and quality judgment control logic" proposed in this invention significantly shortens the response time of the identification link and reduces maintenance costs. Especially in high-throughput, high-speed cooling platen production line scenarios, traditional PCs or edge computing platforms suffer from problems such as: external image transmission bottlenecks; large latency jitter; and unstable response time of classification algorithms when defect types are diverse. This invention, through a structured description centered on feature vectors, introduces hardware primitives such as lookup table mapping, range comparison, and state transition logic into the judgment rules. While maintaining a defect identification accuracy of over 90%, it can still achieve stable judgment output, effectively supporting image recognition-based quality closed-loop control.

[0077] Example 2: Furthermore, the fiberglass board production quality inspection system provided by this invention employs a fiberglass board production quality inspection method from the above embodiments, which can solve a technical problem in fiberglass board production quality inspection. Compared with the prior art, the beneficial effects of the fiberglass board production quality inspection system provided by this invention are the same as those of the fiberglass board production quality inspection method from the above embodiments, and other technical features of the fiberglass board production quality inspection system are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0078] Example 3: This invention provides a quality inspection device for glass fiberboard production. Please refer to... Figure 4A fiberglass board production quality inspection device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which are executed by the at least one processor to enable the at least one processor to perform a fiberglass board production quality inspection method as described in Embodiment 1 above. The fiberglass board production quality inspection device in this embodiment may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital radio receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), vehicle terminals (e.g., vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. This fiberglass board production quality inspection device is merely an example and should not limit the functionality or scope of use of the embodiments of this invention. The fiberglass board production quality inspection device may include a processing device 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in a read-only memory 1002 or a program loaded from a storage device 1003 into a random access memory 1004. Random access memory 1004 also stores various programs and data required for the operation of a fiberglass board production quality inspection device. Processing device 1001, read-only memory 1002, and random access memory 1004 are interconnected via bus 1005. I / O interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows a fiberglass board production quality inspection device to communicate wirelessly or wiredly with other devices to exchange data. Although a fiberglass board production quality inspection device with various systems is shown in the figure, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems may be implemented alternatively.

[0079] Example 4: This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the glass fiberboard production quality inspection method described above. The computer program product provided by this invention can solve a technical problem related to glass fiberboard production quality inspection. Compared with the prior art, the beneficial effects of the computer program product provided by this invention are the same as those of the glass fiberboard production quality inspection method provided in the above embodiments, and will not be repeated here.

[0080] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from read-only memory 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this invention.

[0081] It should be understood that the various parts disclosed in this invention can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics may be combined in any suitable manner in one or more embodiments or examples.

[0082] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for quality inspection in the production of fiberglass boards, characterized in that, The methods include: Step S10: During the cooling stage after the glass fiber board completes the hot pressing process, a continuous infrared thermal image sequence S of the glass fiber board surface is acquired using a preset infrared thermal imaging device. Step S20: Based on the infrared thermal image sequence S, execute the nonlinear median filtering algorithm based on spatial preservation and the background temperature modeling mechanism guided by local response in sequence, output the background temperature T(t) of the current frame, and construct the background temperature evolution curve; perform background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background temperature evolution curve, and output the temperature difference image set ΔT(x,y,t). Step S30: Based on the temperature difference image set ΔT(x,y,t), a multi-channel thermal diffusion model is constructed using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of abnormal primitive regions Ω. Step S40: Perform an image processing mechanism based on a combination of fitting enhancement and saliency detection on the abnormal primitive region set Ω, and output the final defect primitive mask M(x,y); Step S50: Based on the output final defect primitive mask M(x,y), execute the preset defect identification and quality judgment control logic based on feature vector driving in the preset FPGA control platform, and finally output the quality inspection result.

2. The method for quality inspection in the production of glass fiberboard as described in claim 1, characterized in that, In step S10, the infrared thermal image sequence S includes multiple two-dimensional infrared thermal image frames arranged in chronological order. Each pixel position (x, y) in each two-dimensional infrared thermal image frame corresponds to a temperature value T(x, y, t) at a time t; where x represents the horizontal coordinate of the pixel position and y represents the vertical coordinate of the pixel position.

3. The method for quality inspection in the production of glass fiberboard as described in claim 1, characterized in that, In step S20, based on the infrared thermal image sequence S, the nonlinear median filtering algorithm based on spatial preservation and the background temperature modeling mechanism guided by local response are executed sequentially to output the background temperature T(t) of the current frame and construct the background temperature evolution curve. The steps involved in performing background subtraction processing on the infrared thermal image sequence S based on the background temperature evolution curve and outputting a temperature difference image set ΔT(x,y,t) specifically include: Step S201: Define an adaptive neighborhood window W(x,y) for each pixel position (x,y) in the infrared thermal image sequence S. The size of the adaptive neighborhood window W(x,y) is determined according to the temperature change rate of the pixel position (x,y) in the frames before and after it. ,in, The default base window size; The gradient is in the time direction; This is the window adjustment coefficient; Median filtering is performed on the temperature value T(x,y,t) at time t for each pixel position (x,y) in the infrared thermal image sequence S within an adaptive neighborhood window W(x,y), and the filtered image frame is output. ; Step S202: Filter the image frame Extract the time dimension sequence, use linear least squares fitting to construct the background temperature evolution curve based on the time dimension sequence, and output the background temperature T(t) of the current frame; Step S203: Calculate the background residual standard deviation of each pixel based on the background temperature evolution curve, and perform background subtraction processing based on normalized residual compensation on the infrared thermal image sequence S according to the background residual standard deviation, and output the temperature difference image set ΔT(x,y,t).

4. The method for quality inspection in the production of glass fiberboard as described in claim 1, characterized in that, Step S30 involves constructing a multi-channel thermal diffusion model based on the temperature difference image set ΔT(x,y,t) using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of anomalous primitive regions Ω. Specifically, this includes: Step S301: Construct a multi-channel heat diffusion model. The multi-channel heat diffusion model divides the infrared thermal image sequence S into multiple feature channel segments based on the temperature difference image set ΔT(x,y,t), forming a multi-channel set. The multi-channel set includes the first cooling segment. Second cooling section and the third cooling section Among them, the first cooling section Surface response used to indicate initial cooling; second cooling section Used to represent shallow heat conduction in intermediate cooling; third cooling section Used to represent the deep hysteresis response of late-stage cooling; Step S302: The multi-channel thermal diffusion model is applied to the k-th characteristic channel segment in the multi-channel set. Perform the thermal diffusion consistency score calculation process for multi-channel verification and output anomaly candidate points; Step S303: If a pixel location is repeatedly marked as an anomalous candidate point in two or more non-adjacent feature channel segments, the multi-channel thermal diffusion model determines that the pixel location is a thermal diffusion anomalous primitive and finally outputs the set of anomalous primitive regions Ω.

5. The method for quality inspection in the production of glass fiberboard as described in claim 1, characterized in that, In step S40, the step of performing an image processing mechanism based on a combination of fitting enhancement and saliency detection on the abnormal primitive region set Ω to output the final defect primitive mask M(x,y) specifically includes: Step S401: Obtain the abnormal temperature difference image set corresponding to the pixel position (x,y) in the abnormal primitive region set Ω, and perform logarithmic domain polynomial fitting on the abnormal temperature difference image set within a preset time window to obtain the thermal diffusion rate image G(x,y). Step S402: Based on the obtained thermal diffusion rate image G(x,y), construct a neighborhood window centered on the pixel position (x,y), calculate the local mean μ(x,y) and standard deviation σ(x,y), perform saliency detection processing based on the local mean μ(x,y) and standard deviation σ(x,y), and output the saliency score S(x,y); aggregate the pixel positions where the saliency score S(x,y) is greater than the preset saliency score threshold to generate a saliency primitive mask. ; Step S403: Mask the saliency primitives Perform primitive-level cross-referencing with the abnormal primitive region set Ω. If a pixel position (x, y) simultaneously satisfies: If the result is positive, it is marked as the final defect primitive, and the final defect primitive mask M(x,y) is output.

6. The method for quality inspection in the production of glass fiberboard as described in claim 1, characterized in that, In step S50, based on the output final defect primitive mask M(x,y), a preset defect identification and quality judgment control logic driven by feature vectors is executed in a preset FPGA control platform to finally output the quality inspection result. This step specifically includes: Step S501: In the preset FPGA control platform, perform maximum temperature difference extraction and time response analysis on each connected region i in the final defect primitive mask M(x,y) to construct its five-dimensional feature vector. , ,in, The maximum temperature contrast within connected region i; Thermal response polarity, A value of +1 indicates a heating response, including cavitation response and degumming response; A value of -1 indicates a cooling response, including the dry wire response; For the response start time; Let i be the area of ​​the connected region. The regional compactness index is defined as follows: ,in, Let i be the perimeter of the connected region i, and let i be the region compactness index. The closer it is to 1, the closer it is to a circle; Step S502: For the five-dimensional feature vector Perform the following multi-condition rule judgment and output the corresponding defect classification control code. ; in, and These are the preset negative temperature contrast threshold and positive temperature contrast threshold, respectively; and These are the preset early response judgment time threshold and the late response judgment time threshold, respectively; and These are the preset lower limit and upper limit for determining the area of ​​the region, respectively. and These are the preset lower limit and upper limit for shape compactness determination, respectively; This indicates a "logical AND" relationship, used to describe a condition that must be met simultaneously for a statement to be valid. Step S503: Apply the defect classification control code obtained from the determination. The input is converted into a binary control word and sent to the preset host computer, which then outputs the final quality detection result.

7. A fiberglass board production quality inspection system, applied to the fiberglass board production quality inspection method according to any one of claims 1 to 6, characterized in that, The fiberglass board production quality inspection system includes: The infrared image acquisition module is used to acquire a continuous infrared thermal image sequence S on the surface of the glass fiber board through a preset infrared thermal imaging device during the cooling stage after the hot pressing process of the glass fiber board is completed. The background modeling and subtraction module is used to sequentially execute a nonlinear median filtering algorithm based on spatial preservation and a background temperature modeling mechanism guided by local response based on the infrared thermal image sequence S, output the background temperature T(t) of the current frame, and construct the background temperature evolution curve; according to the background temperature evolution curve, the infrared thermal image sequence S is subjected to background subtraction processing based on normalized residual compensation, and the temperature difference image set ΔT(x,y,t) is output. The multi-channel thermal diffusion anomaly extraction module is used to construct a multi-channel thermal diffusion model based on the temperature difference image set ΔT(x,y,t) using a thermal diffusion consistency analysis mechanism based on local temporal residual evolution and multi-channel cross-validation. The multi-channel thermal diffusion model outputs a set of anomalous primitive regions Ω. The defect saliency detection module is used to perform an image processing mechanism based on a combination of fitting enhancement and saliency detection on the set of abnormal primitive regions Ω, and output the final defect primitive mask M(x,y); The FPGA identification and output control module is used to execute preset feature vector-driven defect identification and quality judgment control logic on a preset FPGA control platform based on the output final defect primitive mask M(x,y), and finally output the quality inspection result.

8. A quality inspection device for fiberglass board production, characterized in that, The fiberglass board production quality inspection equipment includes: a memory, a processor, and a fiberglass board production quality inspection program stored in the memory and executable on the processor. When the fiberglass board production quality inspection program is executed by the processor, it implements a fiberglass board production quality inspection method according to any one of claims 1 to 6.

9. A computer program product, characterized in that, The computer program product includes a fiberglass board production quality inspection program, which, when executed by a processor, implements a fiberglass board production quality inspection method according to any one of claims 1 to 6.

Images