A visual inspection method and system for the quality of wood-plastic composite flooring

Abstract

This application relates to the field of quality inspection technology for wood-plastic composite (WPC) flooring, specifically disclosing a visual inspection method and system for WPC flooring. The method includes: acquiring a surface image of the WPC flooring and identifying potential defect areas based on a visual recognition model; performing non-contact physical detection on the potential defect areas to obtain their response characteristics; verifying whether the potential defect areas are genuine defects and generating verification results; adjusting and calibrating the recognition logic of the visual recognition model based on the verification results; extracting feature information from the verified genuine defect areas, associating it with a durability rule set, predicting the service life range and / or durability risk level of the WPC flooring, and outputting the prediction results. This application improves the accuracy of defect identification and enables quantitative prediction of the product's service life in complex outdoor environments.

Description

Technical Field

[0001] This application relates to the field of quality inspection technology for wood-plastic composite flooring, and in particular to a visual inspection method and system for the quality of wood-plastic composite flooring. Background Technology

[0002] On automated production lines for wood-plastic composite (WPC) flooring, traditional visual inspection systems, which use industrial cameras and image analysis algorithms (such as edge detection) to identify surface cracks, have become the mainstream quality control method. However, to meet the market's higher demands for outdoor durability, manufacturers have introduced new fiber reinforcing agents and optimized processes, which has unexpectedly led to a fundamental change in the initial crack morphology. These new cracks typically manifest as extremely fine, irregularly shaped micro-fractures that are highly similar to normal wood grain. Their blurred edge features render the identification rules designed for clearly geometric cracks, upon which traditional visual inspection relies, ineffective, significantly reducing system sensitivity and resulting in serious missed detections.

[0003] To address the issue of missed detections, technicians needed to recalibrate the system. However, the insidious nature of the new cracks made it extremely easy to make mistakes when manually screening "qualified" and "defective" samples. If floorboards containing minute cracks were mistakenly labeled as "qualified samples" and input into the system for learning, the algorithm would incorrectly classify defect features as "normal," effectively relaxing internal inspection standards. This relaxed defect judgment logic due to calibration deviation conflicted with the system's final comprehensive qualification judgment logic, causing a batch of floorboards containing potential defects that should have been intercepted to be mistakenly judged as qualified products and released from the factory.

[0004] A more profound problem is that these minute cracks leaving the factory will accelerate their propagation under harsh outdoor conditions such as sun exposure, rain, and temperature and humidity cycles, causing serious failures such as cracking and peeling to occur far before the product reaches its promised lifespan. Furthermore, once these problems occur, existing visual inspection systems, due to their incorrectly calibrated standards, are unable to effectively trace and locate early defects in the production process, and are completely unable to establish a correlation between the microscopic defect features captured during production and the performance degradation of the product under long-term use. This exposes a fundamental limitation of the current quality inspection system: it can only judge the "conformity" at the moment of production, lacking the ability to predict and assess the long-term lifespan of the product, and unable to prevent batch quality risks caused by potential microscopic defects from the source. Summary of the Invention

[0005] This application proposes a visual inspection method and system for the quality of wood-plastic composite flooring, aiming to solve the technical problems of existing visual inspection systems for wood-plastic composite flooring quality, such as missed detection, misjudgment, and lack of ability to predict the long-term performance of products when faced with complex micro-cracks caused by new materials and new processes.

[0006] In a first aspect, this application provides a visual inspection method for the quality of wood-plastic composite flooring, comprising the following steps:

[0007] Acquire surface images of the wood-plastic composite flooring and identify potential defect areas in the surface images based on a visual recognition model;

[0008] Non-contact physical detection is performed on the potential defect area to obtain the response characteristics of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response characteristics include acoustic characteristics and thermal characteristics;

[0009] Based on the acoustic and thermal characteristics, the system verifies whether the potential defect region is a real defect and generates verification results.

[0010] Based on the verification results, the recognition logic parameters of the visual recognition model are adjusted and calibrated.

[0011] The feature information of potential defect areas verified as real defects is extracted and correlated with a pre-established set of durability rules that characterize the degradation behavior of materials under outdoor environmental stress. The service life range and / or durability risk level of the wood-plastic flooring are predicted, and the prediction results are output.

[0012] As a further embodiment of this application, the step of acquiring a surface image of the wood-plastic composite flooring and identifying potential defect areas of the surface image based on a visual recognition model includes:

[0013] Using an industrial camera, the surface of the wood-plastic composite flooring is photographed to obtain a surface image of the wood-plastic composite flooring;

[0014] The surface image undergoes preprocessing including grayscale conversion, noise removal, and contrast enhancement.

[0015] Based on a visual recognition model, an anomaly detection method based on local texture differences is used to analyze the preprocessed surface image and identify potential defect areas that have statistical differences from the preset normal wood-plastic composite flooring texture features.

[0016] As a further embodiment of this application, the step of analyzing the preprocessed surface image and identifying potential defect areas that statistically differ from the preset normal wood-plastic composite flooring texture features using an anomaly detection method based on local texture differences includes:

[0017] The preprocessed surface image is divided into multiple local image blocks;

[0018] Calculate the local second-order moment features for each of the local image patches, wherein the local second-order moment features include at least one of energy, entropy, and moment of inertia;

[0019] Calculate the deviation of the local second-order moment feature from the preset statistical distribution of normal wood-plastic composite wood flooring texture features;

[0020] Local image blocks whose deviation values ​​exceed a preset deviation threshold are marked as abnormal image blocks;

[0021] The abnormal image blocks that are spatially adjacent are combined to form at least one connected region, and each connected region is identified as a potential defect region.

[0022] The adjustable parameters included in the recognition logic of the visual recognition model include the statistical distribution of the texture features of the normal wood-plastic composite flooring and / or the deviation threshold.

[0023] As a further embodiment of this application, the acoustic feature includes the amplitude attenuation rate of the echo signal, and the thermal feature includes the attenuation rate of the surface temperature.

[0024] The step of verifying whether the potential defect region is a real defect based on the acoustic and thermal characteristics, and generating a verification result, includes:

[0025] When the amplitude attenuation rate exceeds a preset attenuation threshold and the attenuation rate is lower than a preset rate threshold, a verification result is generated that the potential defect area is a real defect.

[0026] As a further embodiment of this application, the acoustic features include at least one of the main frequency offset of the echo signal, the waveform distortion index, and the phase change amount, and the thermal features include at least one of the multi-point temperature gradient change rate, the thermal diffusion anisotropy index, and the surface temperature fluctuation frequency.

[0027] The step of verifying whether the potential defect region is a real defect based on the acoustic and thermal characteristics, and generating a verification result, includes:

[0028] Based on the acoustic and thermal features, a multidimensional feature vector is constructed.

[0029] The multidimensional feature vector is input into a trained pattern recognition model, and the probability value of the potential defect region being a real defect is output; wherein, the pattern recognition model is a Bayesian classifier or a fuzzy logic reasoning model.

[0030] When the probability value exceeds a preset confidence threshold, a verification result is generated indicating that the potential defect area is a real defect.

[0031] As a further embodiment of this application, the step of adjusting and calibrating the recognition logic parameters of the visual recognition model based on the verification results includes:

[0032] The generated verification results are bound to the corresponding potential defect area images to form labeled training samples;

[0033] When the preset calibration triggering conditions are met, the visual recognition model is retrained or incrementally learned using the training samples in order to adjust and calibrate the parameters of the visual recognition model.

[0034] The calibration triggering condition includes at least the following: the cumulative number of training samples collected reaches a preset threshold.

[0035] As a further embodiment of this application, the step of extracting feature information of potential defect areas verified as real defects, associating it with a pre-established set of durability rules characterizing the deterioration behavior of materials under outdoor environmental stress, predicting the service life range and / or durability risk level of the wood-plastic flooring, and outputting the prediction results includes:

[0036] Visual morphological features and physical field signal features of potential defect regions verified as real defects are extracted as feature information; wherein, the visual morphological features include at least one of length, width, and curvature, and the physical field signal features include at least one of the acoustic features of echo signal amplitude attenuation rate and the thermal features of surface temperature attenuation rate.

[0037] The extracted feature information and preset outdoor environmental conditions are input together into a pre-established durability rule set that characterizes the degradation behavior of materials under outdoor environmental stress; wherein, the durability rule set stores correlation relationships, and the correlation relationships define the mapping between different combinations of visual morphological features, physical field signal features and outdoor environmental conditions and the corresponding degradation rate.

[0038] The durability rule set is used to match the corresponding correlation based on the input feature information and outdoor environmental conditions to obtain the degradation rate corresponding to the potential defect area;

[0039] Based on the deterioration rate, the visual morphological characteristics of the potential defective areas, and the preset product failure threshold, the estimated time interval required for the wood-plastic flooring to reach a failure state is calculated.

[0040] The estimated time interval is used as the estimated service life interval of the wood-plastic flooring, and mapped to the corresponding durability risk level;

[0041] The output includes the predicted service life range and / or durability risk level.

[0042] As a further embodiment of this application, the physical detection further includes spectral detection, and the response characteristics further include chemical composition characteristics obtained based on spectral detection;

[0043] Before the step of calculating the estimated time interval required for the wood-plastic flooring to reach a failure state based on the deterioration rate, the visual morphological characteristics of the potential defect area, and a preset product failure threshold, the following steps are also included:

[0044] Based on the characteristics of the chemical components, a preset chemical correction mapping table is consulted to obtain the corresponding chemical correction coefficients;

[0045] Multiply the degradation rate obtained by matching by the chemical correction factor to obtain the corrected degradation rate;

[0046] The degradation rate used to subsequently calculate the estimated time interval required for the wood-plastic composite flooring to reach a failure state is a corrected degradation rate.

[0047] As a further embodiment of this application, the physical detection is implemented based on a local detection device, which includes an integrated component and a motion platform for driving the integrated component. The integrated component includes an acoustic probe, a thermal probe, and a spectral probe.

[0048] The process for obtaining the response features includes:

[0049] Control the motion platform to position the integrated component directly above the potential defect area;

[0050] The acoustic probe, thermal probe, and spectral probe are driven synchronously to emit sound waves, perform local heating, and emit excitation light into the potential defect area, respectively, and receive echo signals, acquire temperature change sequences, and collect scattering spectral signals accordingly.

[0051] The acoustic features are extracted from the echo signal, the thermal features are extracted from the temperature change sequence, and the chemical component features are extracted from the scattering spectrum signal.

[0052] Secondly, this application also provides a visual inspection system for the quality of wood-plastic composite flooring, comprising:

[0053] The image acquisition module is used to acquire surface images of the wood-plastic composite flooring and, based on a visual recognition model, identify potential defect areas in the surface images.

[0054] The feature acquisition module is used to perform non-contact physical detection on the potential defect area to obtain the response features of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response features include acoustic features and thermal features;

[0055] The defect identification module is used to verify whether the potential defect area is a real defect based on the acoustic and thermal characteristics, and to generate a verification result.

[0056] The visual calibration module is used to adjust and calibrate the recognition logic parameters of the visual recognition model based on the verification results.

[0057] The results output module is used to extract feature information of potential defect areas that have been verified as real defects, associate them with a pre-established set of durability rules that characterize the deterioration behavior of materials under outdoor environmental stress, predict the service life range and / or durability risk level of the wood-plastic flooring, and output the prediction results.

[0058] The technical solution according to the embodiments of this application has at least the following beneficial effects:

[0059] This application integrates visual scanning with acoustic and thermal physical field verification. This method can "see through" the internal structure of materials, overcoming the limitations of traditional pure optical inspection in identifying defects with concealed shapes or similar textures, thus improving the accuracy and reliability of defect identification. Simultaneously, it utilizes the objective "truth values" provided by physical field detection to drive the calibration of the visual recognition model, avoiding the risks of systemic learning errors and standard relaxation caused by subjective human misjudgment of samples. Most importantly, this method, by associating verified early defect features with a set of material durability behavior rules, achieves quantitative prediction of product lifespan in complex outdoor environments, completing a paradigm shift in quality control from "immediate conformity judgment" to "long-term performance prediction."

[0060] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0061] The accompanying drawings are used to provide a further understanding of the technical solutions of this application and constitute a part of the specification. They are used together with the embodiments of this application to explain the technical solutions of this application and do not constitute a limitation on the technical solutions of this application.

[0062] Figure 1 This is a flowchart illustrating a visual inspection method for the quality of wood-plastic composite flooring provided in an embodiment of this application.

[0063] Figure 2 This is a schematic diagram of the architecture of a visual inspection system for the quality of wood-plastic composite flooring provided in an embodiment of this application. Detailed Implementation

[0064] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this application, and not all embodiments. The components of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0065] Traditional visual inspection systems for WPC (wood-plastic composite) flooring quality suffer from a significant decrease in sensitivity of image edge detection rules when faced with cracks that are more complex in shape and less obvious in the early stages, caused by new materials and processes. This leads to the system frequently misclassifying these real but unusually shaped micro-cracks as normal surface textures or ignoring them altogether when processing new batches of flooring, resulting in missed detections. Furthermore, because these new cracks are extremely fine in the early stages of production, their visual characteristics are highly similar to normal WPC textures. Even experienced technicians are easily influenced when manually selecting "qualified" samples for training and calibration, leading to incorrect system calibration and further exacerbating the missed detection problem. More importantly, existing systems lack the ability to predict the long-term performance of products, failing to effectively correlate and assess production quality with product lifespan.

[0066] In this regard, such as Figure 1 As shown, this application proposes a visual inspection method for the quality of wood-plastic composite flooring, including the following steps:

[0067] S110, acquire a surface image of the wood-plastic composite flooring, and identify potential defect areas in the surface image based on a visual recognition model;

[0068] S120, perform non-contact physical detection on the potential defect area to obtain the response characteristics of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response characteristics include acoustic characteristics and thermal characteristics;

[0069] S130, based on the acoustic and thermal characteristics, verify whether the potential defect area is a real defect, and generate a verification result;

[0070] S140, Based on the verification results, adjust and calibrate the parameters of the recognition logic of the visual recognition model;

[0071] S150: Extract feature information of potential defect areas that have been verified as real defects, associate them with a pre-established set of durability rules that characterize the deterioration behavior of materials under outdoor environmental stress, predict the service life range and / or durability risk level of the wood-plastic flooring, and output the prediction results.

[0072] The visual inspection method for the quality of wood-plastic composite flooring proposed in this application aims to improve the quality control level of wood-plastic composite flooring during the production process, especially for potential defects that are not obvious at first but have a significant impact on product durability.

[0073] "Plastic composite flooring" refers to flooring materials made of wood fiber and plastic composites, characterized by combining the natural appearance of wood with the durability of plastic. "Visual recognition model" refers to a model built based on machine learning or deep learning algorithms, capable of identifying and classifying visual patterns by analyzing image data; in this application, it is used to identify potential defects on the surface of the plastic composite flooring. "Potential defect area" refers to the area where the visual recognition model initially determines may contain defects. "Non-contact physical detection" refers to a detection method that does not directly contact the surface of the plastic composite flooring, but rather obtains its internal or surface characteristics through the interaction of physical fields (such as sound waves and heat) with the material. "Response characteristics" refer to specific physical quantities exhibited by the material in response to the detection signal during physical detection, such as acoustic and thermal characteristics. "Acoustic detection" typically refers to methods that utilize the propagation, reflection, and attenuation characteristics of sound waves in materials to detect internal structures or defects. "Thermal detection" typically refers to methods that detect material defects by applying or removing heat and monitoring changes in the material's surface temperature. "Durability rule set" refers to a pre-established knowledge base or model that contains the laws governing the degradation behavior of materials under different outdoor environmental stresses, used to predict the service life and risk level of products.

[0074] First, potential defect areas need to be identified. Specifically, high-resolution surface images of the WPC flooring can be obtained by photographing the surface with an industrial camera. These images are then fed into a pre-trained visual recognition model. The visual recognition model is designed to analyze features such as texture, color, and geometry in the images to identify areas that differ from normal WPC flooring surfaces and mark these areas as potential defect areas. As an alternative approach, surface images of the WPC flooring can also be obtained using a scanner, or images containing more spectral information can be acquired using multispectral imaging equipment to provide richer visual features for model analysis.

[0075] For acoustic detection, an ultrasonic probe can be used to emit high-frequency sound waves into a potential defect area and receive the echo signals reflected back from within the material. By analyzing parameters such as the amplitude, frequency, and phase of the echo signals, acoustic characteristics can be obtained, such as the attenuation of sound waves or changes in propagation speed within the defect area. For thermal detection, an infrared thermal imager can be used to locally heat or cool the potential defect area and monitor its surface temperature changes over time in real time. By analyzing the temperature change sequence, thermal characteristics can be obtained, such as the rate of heat diffusion or anomalies in the temperature gradient within the defect area. These acoustic and thermal characteristics can reflect the structural integrity, density changes, or thermal conductivity of the material's interior, thus providing physical evidence for the authenticity of the defect. For example, a region with internal voids may experience increased sound wave attenuation and slower heat diffusion.

[0076] Next, based on the acquired acoustic and thermal features, the system verifies whether the potential defect area is a real defect and generates a verification result. This step is crucial for secondary confirmation of the visual recognition result using physical detection data. For example, a series of thresholds can be set; when both acoustic and thermal features simultaneously meet preset anomaly conditions, the potential defect area can be determined to be a real defect. The system then generates a verification result, clearly indicating whether the potential defect area has been confirmed as a real defect.

[0077] Based on the verification results, the recognition logic of the visual recognition model is parameter-adjusted and calibrated. This step aims to optimize the performance of the visual recognition model by utilizing the verification results of physical detection, enabling it to more accurately identify novel defects. Specifically, images of potential defect areas verified as genuine defects, as well as images of areas verified as non-defective, can be used as training samples with correct labels. During retraining, the model's recognition logic (e.g., feature extraction algorithms, classifier parameters, or thresholds used to identify defects) is adjusted and optimized based on these new, physically verified samples. For example, if the model previously misclassified a novel microcrack as normal texture, upon receiving samples verified as genuine defects by physical detection, the model will adjust its internal parameters to increase its sensitivity to such cracks, thereby more accurately identifying them in subsequent detections.

[0078] Finally, the service life range and / or durability risk level of the WPC flooring are predicted, and the prediction results are output. This step is one of the core innovations of this application, combining defect detection during the production stage with long-term performance prediction of the product in outdoor environments. Specifically, various feature information, such as visual morphological features and physical field signal features, can be extracted from potential defect areas verified as genuine defects. This feature information is then input into a pre-established durability rule set. This rule set stores a large amount of correlation knowledge about how different types and degrees of defects affect the degradation rate and service life of WPC flooring materials under different outdoor environmental stresses (such as ultraviolet radiation, humidity cycling, temperature changes, etc.). By matching the extracted defect feature information with the rule set, the degradation rate of the WPC flooring in a specific outdoor environment can be predicted, thereby calculating its expected service life range and mapping it to the corresponding durability risk level (e.g., high risk, medium risk, low risk). Finally, the system outputs a report containing these prediction results, providing manufacturers with valuable information about the long-term performance of the product.

[0079] The working principle of this application lies in constructing a multi-level detection system of "visual initial screening - physical field verification - lifespan prediction". First, a high-resolution camera is used for visual scanning to identify potential defect areas. Then, non-contact acoustic and thermal detection methods are introduced to obtain independent physical response characteristics at the internal structure level of the material, objectively verifying the "truth value" of the visual recognition results. This accurately distinguishes between real microcracks and similar textures, effectively solving the problem of missed detection and misjudgment caused by the concealment of new crack morphologies. The objective results obtained from the verification are also used to drive the adaptive calibration of the visual recognition model, ensuring its continuous optimization. Finally, the system correlates the verified defect features with a "durability rule set" established based on accelerated aging experiments. Through a quantitative model, the expansion behavior of the defect in a specific outdoor environment is predicted, thereby outputting the product's expected service life range and risk level, achieving a leap from immediate quality judgment to long-term performance evaluation.

[0080] In summary, this application, by integrating visual scanning with acoustic and thermal physical field verification, enables a method that "sees through" the internal structure of materials, overcoming the limitations of traditional pure optical inspection in identifying defects with concealed shapes or similar textures, thus improving the accuracy and reliability of defect identification. Simultaneously, by utilizing the objective "truth values" provided by physical field detection to drive the calibration of the visual recognition model, it avoids the risks of systemic learning errors and standard relaxation caused by subjective human misjudgment of samples. Most importantly, this method, by associating verified early defect features with a set of material durability behavior rules, achieves quantitative prediction of product lifespan in complex outdoor environments, completing a paradigm shift in quality control from "immediate conformity judgment" to "long-term performance prediction."

[0081] In some embodiments of this application, the step of acquiring a surface image of the wood-plastic composite flooring and identifying potential defect areas in the surface image based on a visual recognition model preferably includes:

[0082] Using an industrial camera, the surface of the wood-plastic composite (WPC) flooring is photographed to obtain surface images. The industrial camera typically refers to a high-performance digital camera designed specifically for industrial environments. Its features include high resolution, high frame rate, high stability, and precise synchronous triggering, ensuring rapid and accurate capture of detailed surface images of the WPC flooring on the production line, providing high-quality raw data for subsequent defect detection.

[0083] The surface image undergoes preprocessing including grayscale conversion, noise removal, and contrast enhancement. Specifically, grayscale conversion is the process of converting a color image to a grayscale image, aiming to simplify image data, reduce computational load, and preserve brightness information, which is usually sufficient for texture analysis. Noise removal aims to eliminate random interference introduced during image acquisition, such as smoothing the image using algorithms like median filtering and Gaussian filtering, to improve image clarity and the accuracy of subsequent feature extraction. Contrast enhancement adjusts the image's brightness range, making the grayscale differences between different regions more pronounced, thereby highlighting the visual features of potential defects and facilitating analysis by visual recognition models.

[0084] Based on a visual recognition model, an anomaly detection method based on local texture differences is employed to analyze preprocessed surface images and identify potential defect areas that statistically differ from the pre-defined normal texture features of WPC (Wall-in-Plastic) flooring. The visual recognition model is trained to recognize normal texture patterns on the WPC flooring surface. The core of this local texture difference-based anomaly detection method lies in analyzing the texture features of local areas of the image and comparing them with pre-defined normal WPC flooring texture features. When a significant statistical difference exists between the local texture features and the normal texture features, the area is identified as a potential defect region. This method can effectively detect various forms of anomalies such as surface cracks, scratches, dents, and foreign objects, because these defects typically cause significant changes in local texture.

[0085] Through the above technical solution, this application can achieve preliminary, efficient, and accurate identification of potential defect areas on the surface of wood-plastic composite flooring. Specifically, the use of an industrial camera ensures the accuracy and stability of image acquisition; the image preprocessing step effectively improves image quality, laying a good foundation for subsequent analysis; and the anomaly detection method based on local texture differences enables the visual recognition model to sensitively capture subtle surface anomalies, even defects that are difficult to detect with the naked eye can be preliminarily identified. This detailed and systematic image acquisition and preliminary identification process improves the accuracy and robustness of potential defect area identification, providing reliable input for subsequent physical detection and verification of real defects, thereby improving the efficiency and reliability of the overall detection method.

[0086] Based on the above implementation method, the preferred step of using an anomaly detection method based on local texture differences to analyze the preprocessed surface image and identify potential defect areas that statistically differ from the preset normal wood-plastic composite flooring texture features includes:

[0087] The preprocessed surface image is divided into multiple local image blocks;

[0088] Calculate the local second-order moment features for each of the local image patches, wherein the local second-order moment features include at least one of energy, entropy, and moment of inertia;

[0089] Calculate the deviation of the local second-order moment feature from the preset statistical distribution of normal wood-plastic composite wood flooring texture features;

[0090] Local image blocks whose deviation values ​​exceed a preset deviation threshold are marked as abnormal image blocks;

[0091] The abnormal image blocks that are spatially adjacent are combined to form at least one connected region, and each connected region is identified as a potential defect region.

[0092] The adjustable parameters included in the recognition logic of the visual recognition model include the statistical distribution of the texture features of the normal wood-plastic composite flooring and / or the deviation threshold.

[0093] To analyze the local texture characteristics of the image, the preprocessed surface image is segmented into multiple local image patches. This segmentation operation aims to decompose the overall image into smaller, independently analyzable regions, in order to capture subtle texture variations in local areas.

[0094] Calculate the local second-order moment features for each local image patch. Energy features typically measure the uniformity or flatness of image texture; high energy values ​​may indicate less texture variation. Entropy features reflect the randomness or complexity of image texture; high entropy values ​​may indicate a more disordered texture. Inertia moments characterize the contrast or intensity of local variations in image texture; high inertia moments may mean significant local differences in texture. By calculating these features, the unique texture properties of each local image patch can be quantified.

[0095] Subsequently, the deviation value between the local second-order moment features and the statistical distribution of the preset normal WPC flooring texture features is calculated. The statistical distribution of the normal WPC flooring texture features is typically established as a benchmark through analysis and modeling of a large number of defect-free WPC flooring images. The deviation value quantifies the degree of difference between the texture of the current local image patch and the normal texture, and can be calculated, for example, using statistical distance or similarity metrics.

[0096] When the deviation value exceeds a preset deviation threshold, the corresponding local image patch is marked as an abnormal image patch. The deviation threshold is an adjustable parameter, and its setting determines the sensitivity of identifying abnormal textures. By setting an appropriate threshold, regions with significant statistical differences from normal textures can be distinguished.

[0097] To integrate discrete anomalous image patches into meaningful defect regions, spatially adjacent anomalous image patches are grouped together to form at least one connected region. Each connected region is then identified as a potential defect region. This combination operation helps filter out isolated noise points and ensures that the identified defect regions have a certain degree of spatial continuity and physical meaning.

[0098] It is worth noting that the adjustable parameters included in the recognition logic of the visual recognition model include the statistical distribution of the texture features of the normal wood-plastic composite flooring and / or the deviation threshold. The flexibility of these parameters allows the visual recognition model to be finely adjusted and optimized according to actual application scenarios, the characteristics of different batches of materials, or specific defect types, thereby improving the accuracy and adaptability of detection.

[0099] This application effectively detects defects exhibiting texture anomalies, such as minute cracks, scratches, dents, or material inhomogeneity, through refined texture feature extraction and statistical comparison of local image patches. These defects may not be visually apparent from a macroscopic perspective, but can be accurately captured through quantitative analysis of local second-order moment features. Furthermore, combining adjacent anomalous image patches into connected regions helps reduce false alarms caused by image noise or random factors, improving the accuracy and reliability of defect identification. Simultaneously, the adjustable statistical distribution and deviation threshold parameters of normal WPC wood flooring texture features enhance the adaptability and configurability of the visual recognition model, allowing optimization based on the characteristics of different types of WPC wood flooring or varying detection requirements. This significantly improves the accuracy and generalization ability of defect identification while maintaining detection efficiency.

[0100] In some embodiments of this application, the acoustic features include the amplitude attenuation rate of the echo signal, and the thermal features include the attenuation rate of the surface temperature; the step of verifying whether the potential defect region is a real defect based on the acoustic features and thermal features, and generating a verification result preferably includes: when the amplitude attenuation rate exceeds a preset attenuation threshold and the attenuation rate is lower than a preset rate threshold, generating a verification result that the potential defect region is a real defect.

[0101] The amplitude attenuation rate of an echo signal refers to the degree to which the amplitude of the echo signal decreases relative to the initial transmitted signal amplitude after a sound wave penetrates the wood-plastic composite (WPC) material and reflects back from its internal structure. This attenuation rate quantifies the energy loss encountered by the sound wave as it propagates within the material, and is typically determined by the material's density, homogeneity, and the presence of internal defects (such as voids, cracks, delamination, etc.). When defects exist within the material, the sound wave propagation path is disrupted, leading to increased energy scattering and absorption, thus making the amplitude attenuation of the echo signal more significant.

[0102] The surface temperature decay rate refers to the rate at which the surface temperature of wood-plastic composite (WPC) flooring decreases over time after localized heating. When defects exist within the WPC flooring, such as voids or foreign objects, these defects alter the material's heat conduction path and heat diffusion characteristics. Specifically, voids and other defects typically have lower thermal conductivity, hindering the effective transfer of heat from the surface to the interior. Consequently, the surface temperature in the defective area decays at a slower rate compared to the normal area after heating ceases.

[0103] The preset attenuation threshold and preset rate threshold were determined through statistical analysis and experimental verification based on acoustic and thermal response data from a large number of normal plastic-plastic flooring samples. These thresholds represent the typical range of acoustic amplitude and thermal attenuation of normal plastic-plastic flooring materials in a defect-free state. When the detected amplitude attenuation rate exceeds the attenuation threshold, it indicates abnormal sound wave energy loss, possibly indicating internal structural discontinuities. Simultaneously, when the detected surface temperature attenuation rate is lower than the rate threshold, it indicates abnormal heat diffusion, possibly indicating internal defects with high thermal resistance. Only when both conditions are met simultaneously is a defect identified as a genuine defect, thus avoiding misjudgments that may result from judging based on a single feature.

[0104] This application improves the accuracy of defect identification by explicitly defining the acoustic characteristic as the amplitude attenuation rate of the echo signal and the thermal characteristic as the attenuation rate of the surface temperature, and by employing a joint judgment mechanism based on "AND" logic. This method can effectively distinguish between visual anomalies caused by non-defect factors such as uneven surface texture and genuine internal structural defects, thereby reducing false alarm rates and avoiding the incorrect handling of qualified products. Furthermore, through quantified physical characteristics and clear threshold standards, the defect verification process becomes more objective and automated, reducing the subjectivity and uncertainty of manual judgment and providing a solid technical guarantee for the quality control of wood-plastic composite flooring.

[0105] In another embodiment of this application, the acoustic features include at least one of the main frequency offset of the echo signal, the waveform distortion index, and the phase change amount, and the thermal features include at least one of the multi-point temperature gradient change rate, the thermal diffusion anisotropy index, and the surface temperature fluctuation frequency.

[0106] The step of verifying whether the potential defect region is a real defect based on the acoustic and thermal characteristics, and generating a verification result, preferably includes:

[0107] Based on the acoustic and thermal features, a multidimensional feature vector is constructed.

[0108] The multidimensional feature vector is input into a trained pattern recognition model, and the probability value of the potential defect region being a real defect is output; wherein, the pattern recognition model is a Bayesian classifier or a fuzzy logic reasoning model.

[0109] When the probability value exceeds a preset confidence threshold, a verification result is generated indicating that the potential defect area is a real defect.

[0110] The dominant frequency offset of an echo signal refers to the change in the dominant frequency relative to the normal material region when a sound wave encounters a defect while penetrating a material. This change can reflect the size, depth, and internal structure of the defect. The waveform distortion index refers to the degree of difference between the waveform of the echo signal and the normal signal waveform. It can be characterized by calculating the waveform's skewness, kurtosis, or mean square error compared to a standard waveform. Its purpose is to quantify the impact of defects on the sound wave propagation path. The phase change refers to the phase delay or lead of the echo signal relative to the normal signal. This is usually related to local variations in physical parameters such as density and elastic modulus within the material.

[0111] The multi-point temperature gradient change rate can be understood as the difference in the rate of temperature change over time at different locations within a potential defect region after local heating, revealing anomalies in the internal heat conduction paths of the material. The thermal diffusivity anisotropy index refers to the difference in a material's thermal diffusivity in different directions. For wood-plastic composite flooring with anisotropic structures, defects can lead to significant changes in local thermal diffusivity. The surface temperature fluctuation frequency refers to the frequency characteristics of the surface temperature response in a potential defect region under periodic thermal excitation; different types of defects may cause different frequency response modes.

[0112] The multidimensional feature vector can include multiple dimensions such as dominant frequency offset, waveform distortion index, phase change, multi-point temperature gradient change rate, thermal diffusion anisotropy index, and surface temperature fluctuation frequency. This multidimensional feature vector can more comprehensively and meticulously characterize the physical properties of potential defect regions.

[0113] Bayesian classifiers classify potential defect regions by calculating the posterior probability that they belong to real defects given a feature vector. Fuzzy logic inference models, on the other hand, define fuzzy rules and membership functions to perform fuzzy inference on input features, outputting a probability value that a potential defect region is a real defect. During the training phase, these models utilize a large amount of known defect and non-defect sample data to establish a complex mapping relationship between features and defect authenticity. When the probability value output by the pattern recognition model exceeds a preset confidence threshold, the potential defect region is determined to be a real defect. This confidence threshold can be adjusted according to actual application requirements and acceptable false positive and false negative rates.

[0114] Through the above technical solutions, this application can more comprehensively and deeply analyze the physical characteristics of potential defect areas, effectively overcoming the limitations of traditional simple feature extraction and threshold judgment methods in complex defect identification. The introduction of multi-dimensional feature vectors and pattern recognition models makes the defect identification process more intelligent and precise, significantly improving the accuracy and robustness of identifying various defects in wood-plastic composite flooring, especially subtle, complex, or ambiguous defects. Furthermore, by outputting probability values ​​and combining them with confidence thresholds, the system can provide a more flexible decision-making mechanism, adapting to different quality control standards and application scenarios, thereby improving the overall reliability and efficiency of wood-plastic composite flooring quality inspection.

[0115] The following is a specific example to illustrate this.

[0116] Suppose a potential defect area is initially identified on a wood-plastic composite (WPC) flooring production line using a visual recognition model. To verify its authenticity, a local detection device is positioned above this area to perform acoustic and thermal detection. Specifically, the acoustic probe emits ultrasonic waves and receives the echo signals. If the area has internal voids or delamination, the echo signal may exhibit a significant frequency shift (e.g., the frequency shifts from 5MHz to 4.5MHz), increased waveform distortion (e.g., the correlation coefficient between the waveform and the standard waveform is less than 0.8), and phase change (e.g., a phase lag of 30 degrees). Simultaneously, the thermal probe locally heats the area and monitors its surface temperature changes. If internal defects exist, the heat conduction path is obstructed, potentially leading to abnormal multi-point temperature gradient change rates (e.g., the temperature difference decay rate between the center point and the edge points is much lower than in normal areas), increased thermal diffusivity anisotropy (indicating large differences in the rate of heat diffusion in different directions), and changes in the surface temperature fluctuation frequency (e.g., under periodic heating, the response frequency is inconsistent with the excitation frequency). The extracted features, such as the dominant frequency offset, waveform distortion index, phase change, multi-point temperature gradient change rate, thermal diffusivity anisotropy index, and surface temperature fluctuation frequency, are combined into a multi-dimensional feature vector. This vector is then input into a Bayesian classifier pre-trained with a large number of real defects and normal samples to calculate the probability that the potential defect region is a real defect. For example, if the calculated probability is 0.95 and the preset confidence threshold is 0.8, the system will generate a verification result that the potential defect region is a real defect. In this way, even internal defects that are difficult to detect with the naked eye can be identified with high precision.

[0117] In a specific embodiment of this application, the step of adjusting and calibrating the recognition logic parameters of the visual recognition model based on the verification results preferably includes:

[0118] The generated verification results are bound to the corresponding potential defect area images to form labeled training samples;

[0119] When the preset calibration triggering conditions are met, the visual recognition model is retrained or incrementally learned using the training samples in order to adjust and calibrate the parameters of the visual recognition model.

[0120] The calibration triggering condition includes at least the following: the cumulative number of training samples collected reaches a preset threshold.

[0121] The verification results are then bound to the corresponding potential defect area images. These clearly labeled image data constitute labeled training samples. These samples form the basis for the visual recognition model to learn and improve. The calibration trigger condition refers to the conditions that initiate the adjustment and calibration of the visual recognition model parameters. Besides the cumulative number of collected training samples reaching a preset threshold, calibration trigger conditions may include, but are not limited to: the visual recognition model's recognition accuracy falling below a preset threshold, the system running time reaching a preset cycle, or the detection of a specific number of false alarms or missed alarms. When any calibration trigger condition is met, the system will initiate the model update process. In practical applications, retraining or incrementally learning the visual recognition model using the training samples means updating the parameters of the visual recognition model using newly collected labeled training samples. Retraining involves training the model from scratch using all training samples, including historical data and newly collected data, to ensure the model can fully learn all known patterns. Incremental learning, on the other hand, involves making minor parameter adjustments using only newly collected training samples based on the existing model to adapt to new data distributions or defect features. This method is generally more efficient and suitable for real-time or near-real-time model update scenarios. Through these two methods, the recognition logic of the visual recognition model can be dynamically adjusted and calibrated, enabling it to better adapt to changes in the actual production environment.

[0122] This application provides high-quality real data for the learning of visual recognition models by transforming the verification results of physical detection into labeled training samples, effectively avoiding performance degradation caused by data bias or environmental changes.

[0123] In a more specific embodiment of this application, the step of extracting feature information of potential defect areas verified as real defects, associating it with a pre-established set of durability rules characterizing the deterioration behavior of materials under outdoor environmental stress, predicting the service life range and / or durability risk level of the wood-plastic flooring, and outputting the prediction results preferably includes:

[0124] Visual morphological features and physical field signal features of potential defect regions verified as real defects are extracted as feature information; wherein, the visual morphological features include at least one of length, width, and curvature, and the physical field signal features include at least one of the acoustic features of echo signal amplitude attenuation rate and the thermal features of surface temperature attenuation rate.

[0125] The extracted feature information and preset outdoor environmental conditions are input together into a pre-established durability rule set that characterizes the degradation behavior of materials under outdoor environmental stress; wherein, the durability rule set stores correlation relationships, and the correlation relationships define the mapping between different combinations of visual morphological features, physical field signal features and outdoor environmental conditions and the corresponding degradation rate.

[0126] The durability rule set is used to match the corresponding correlation based on the input feature information and outdoor environmental conditions to obtain the degradation rate corresponding to the potential defect area;

[0127] Based on the deterioration rate, the visual morphological characteristics of the potential defective areas, and the preset product failure threshold, the estimated time interval required for the wood-plastic flooring to reach a failure state is calculated.

[0128] The estimated time interval is used as the estimated service life interval of the wood-plastic flooring, and mapped to the corresponding durability risk level;

[0129] The output includes the predicted service life range and / or durability risk level.

[0130] First, feature information needs to be extracted. Visual morphological features refer to the geometric properties of the defect obtained through image analysis, such as the defect's length, width, or curvature. These features visually reflect the defect's macroscopic size and shape. Physical field signal features are the material response characteristics obtained through non-contact physical detection. Examples include the amplitude attenuation rate of the echo signal in acoustic features, which reflects the energy loss of sound waves propagating in the defect area; and the surface temperature attenuation rate in thermal features, which indicates changes in the thermal conductivity of the defect area. These features together constitute a comprehensive quantitative description of the defect.

[0131] Furthermore, the extracted feature information, along with preset outdoor environmental conditions (such as average temperature, humidity, and ultraviolet radiation intensity), is input into a pre-established durability rule set. This durability rule set is a knowledge base or model that stores a large number of relationships. These relationships are established based on long-term observations of the degradation behavior of wood-plastic composite flooring under various outdoor environmental stresses, experimental data analysis, and materials science principles. They precisely define the mapping relationship between different combinations of visual morphological features, physical field signal features, and outdoor environmental conditions and material degradation rates.

[0132] Therefore, the durability rule set intelligently matches the most suitable correlation to the current situation based on the specific input feature information and outdoor environmental conditions, thereby outputting the degradation rate corresponding to the potential defect area. This degradation rate quantifies the rate at which the defect expands or worsens under preset environmental conditions.

[0133] Based on this, by combining the obtained degradation rate, the visual morphological characteristics of potential defect areas (such as the initial size or depth of the defect), and the preset product failure threshold (such as the degree of defect at which the wood-plastic composite flooring is considered to have failed), the estimated time interval required for the wood-plastic composite flooring to reach the failure state can be calculated.

[0134] The estimated time interval required for the wood-plastic composite flooring to reach failure as described in this application is calculated using the following core inputs: the degradation rate (e.g., the estimated crack propagation rate, in millimeters per year) obtained from a durability rule set, the current defect state (e.g., the current crack length) extracted from visual morphological features, and a preset product failure threshold (e.g., the critical crack length for determining product failure, often defined as 50% of the flooring thickness). The basic principle of the calculation is kinematics, which divides the difference between the target state and the current state by the rate of change. Specifically, the estimated time interval is obtained by dividing the difference between the failure threshold and the current defect size by the degradation rate. For example, the formula for calculating the estimated lifespan (T) of crack propagation is: T = (failure crack length - current crack length) / estimated crack propagation rate. The final output is an "estimated time interval" rather than a single value, which is achieved by considering the uncertainty of the parameters. The most common approach is that the degradation rate matched by the system itself comes with a confidence range based on experimental statistics. By substituting the upper and lower limits of this rate into the above formula, the corresponding upper and lower limits of the lifespan can be calculated, thus forming a time interval.

[0135] Ultimately, the calculated estimated time interval is directly used as the estimated service life range of the wood-plastic composite flooring. To provide a more intuitive assessment, this service life range is also mapped to a corresponding durability risk level, such as "low risk," "medium risk," or "high risk," to facilitate quick understanding and decision-making by users. Finally, the system will output the prediction results, which include the estimated service life range and / or the durability risk level.

[0136] This application's solution integrates multi-source heterogeneous defect feature information (including visual morphological features and physical field signal features) and combines it with external environmental conditions, inputting it into a pre-established durability rule set for deep correlation analysis. This rule set can accurately deduce the degradation rate of defects based on these inputs, and then, combined with product failure thresholds, quantitatively predict the remaining service life of the WPC flooring. This process elevates traditional defect detection from a static "problem discovery" to a dynamic "future prediction," making the assessment of WPC flooring quality more forward-looking and practical.

[0137] In a further embodiment of this application, the physical detection further includes spectral detection, and the response characteristics further include chemical composition characteristics obtained based on spectral detection;

[0138] Before the step of calculating the estimated time interval required for the wood-plastic flooring to reach a failure state based on the deterioration rate, the visual morphological characteristics of the potential defect area, and a preset product failure threshold, the following steps are preferably also included:

[0139] Based on the characteristics of the chemical components, a preset chemical correction mapping table is consulted to obtain the corresponding chemical correction coefficients;

[0140] Multiply the degradation rate obtained by matching by the chemical correction factor to obtain the corrected degradation rate;

[0141] The degradation rate used to subsequently calculate the estimated time interval required for the wood-plastic composite flooring to reach a failure state is a corrected degradation rate.

[0142] Spectroscopic detection refers to the technology of obtaining information about the chemical composition of a material by analyzing its absorption, reflection, or scattering characteristics of light at specific wavelengths. Chemical composition characteristics can be understood as the internal chemical composition and proportions of the wood-plastic composite flooring material, such as wood flour content, polymer type, and types of additives. These characteristics directly affect the material's anti-aging, UV resistance, and hydrolysis resistance properties in outdoor environments.

[0143] Before calculating the estimated time interval required for the WPC flooring to reach its failure state, this application adds a step of correcting the degradation rate based on chemical composition characteristics. Specifically, firstly, a pre-defined chemical correction mapping table is consulted based on the obtained chemical composition characteristics. This chemical correction mapping table pre-stores the correspondence between different chemical composition characteristics and corresponding chemical correction coefficients. For example, the presence or absence of certain additives, or changes in the ratio of wood flour to plastic, may correspond to different correction coefficients. After the query, a chemical correction coefficient matching the current chemical composition characteristics of the WPC flooring is obtained. Subsequently, the degradation rate obtained by matching through the durability rule set is multiplied by this chemical correction coefficient to obtain the corrected degradation rate. This corrected degradation rate will serve as the basis for subsequently calculating the estimated time interval required for the WPC flooring to reach its failure state.

[0144] The reason this application incorporates spectral detection and chemical composition characteristics, and corrects for the degradation rate, is that the actual degradation process of WPC (wood-plastic composite) flooring is not only affected by its macroscopic defect morphology and physical responses (such as acoustics and heat), but more profoundly, the chemical composition of the material itself plays a decisive role in the degradation rate. For example, under the same external stress, WPC flooring with different formulations may exhibit significant differences in its UV resistance, oxidation resistance, and water absorption swelling rate, which are directly reflected in its chemical composition. Obtaining chemical composition characteristics through spectral detection can more comprehensively and fundamentally reflect the material's inherent degradation potential. Based on this, using a chemical correction mapping table to correct the initially calculated degradation rate can incorporate the material's inherent chemical properties into the degradation model, thus making the predicted degradation rate closer to reality. It is precisely because the influence of the material's chemical composition on the degradation process is considered that the prediction of the service life and durability risk of WPC flooring is more accurate.

[0145] The following is a specific example to illustrate this.

[0146] Suppose that during the inspection of a batch of wood-plastic composite flooring, potential defect areas are identified using a visual recognition model, and corresponding acoustic and thermal characteristics are obtained through acoustic and thermal detection. Based on these characteristics, a degradation rate, for example, 0.5 mm per year, is initially obtained through a durability rule set.

[0147] Furthermore, the chemical composition characteristics of the potential defect region were obtained through spectral detection, for example, the content of a certain antioxidant was detected to be lower than the standard value. According to the preset chemical correction mapping table, the chemical correction factor corresponding to the low content of the antioxidant was found to be 1.2.

[0148] At this point, multiplying the initial degradation rate of 0.5 mm / year by the chemical correction factor of 1.2 yields a corrected degradation rate of 0.6 mm / year. This corrected 0.6 mm / year will be used as the basis for calculating the estimated time interval required for the wood-plastic composite flooring to reach failure.

[0149] For example, if the product failure threshold is a defect depth of 5 mm, and the visual morphological characteristics (such as initial depth) of the potential defect area are 1 mm, then based on the corrected degradation rate, the estimated time required to reach failure is (5-1) mm / 0.6 mm / year ≈ 6.67 years. Without chemical correction, the estimated time is (5-1) mm / 0.5 mm / year = 8 years. Clearly, by introducing chemical correction, potential degradation risks can be identified earlier, providing a more conservative and safer lifespan prediction.

[0150] In practical applications, the physical detection is based on a local detection device, which includes an integrated component and a motion platform for driving the integrated component. The integrated component includes an acoustic probe, a thermal probe, and a spectral probe.

[0151] The process for obtaining the response features includes:

[0152] Control the motion platform to position the integrated component directly above the potential defect area;

[0153] The acoustic probe, thermal probe, and spectral probe are driven synchronously to emit sound waves, perform local heating, and emit excitation light into the potential defect area, respectively, and receive echo signals, acquire temperature change sequences, and collect scattering spectral signals accordingly.

[0154] The acoustic features are extracted from the echo signal, the thermal features are extracted from the temperature change sequence, and the chemical component features are extracted from the scattering spectrum signal.

[0155] The local detection device is designed to perform precise, multimodal, non-contact detection of specific areas on the surface of wood-plastic composite (WPC) flooring. The core of this device lies in its integrated component and motion platform. The integrated component is a unit that compactly integrates multiple physical detection probes, designed to achieve simultaneous or rapid continuous detection of the same potential defect area. The acoustic probe is used to emit and receive sound waves to obtain information about the material's internal structure; the thermal probe is used to heat the local area and monitor its temperature changes to assess the material's thermal properties; and the spectral probe is used to emit excitation light and collect scattered spectral signals to analyze the material's chemical composition. The motion platform controls the movement and positioning of the integrated component on the WPC flooring surface, ensuring that the integrated component is accurately aligned with each identified potential defect area.

[0156] In the response feature acquisition process, firstly, the integrated component is precisely positioned directly above the target potential defect area by controlling the motion platform. This step ensures that subsequent physical probing can be focused on the specific area to be analyzed. Subsequently, the acoustic probe, thermal probe, and spectral probe within the integrated component are driven synchronously. Specifically, the acoustic probe emits sound waves towards the potential defect area and receives its echo signal; the thermal probe locally heats the area and acquires its temperature change sequence; the spectral probe emits excitation light and collects the scattered spectral signal of the area. After detection, acoustic features are extracted from the received echo signal, thermal features are extracted from the acquired temperature change sequence, and chemical composition features are extracted from the acquired scattered spectral signal. These features collectively constitute a comprehensive response feature of the potential defect area, providing multi-dimensional data support for subsequent defect verification and durability prediction.

[0157] This application, through the introduction of a local detection device, achieves precise localization and multimodal simultaneous detection of potential defect areas, effectively avoiding the localization errors and low detection efficiency that may exist in traditional detection methods. As a result, more accurate and comprehensive response characteristics can be obtained, providing a more reliable data foundation for subsequent verification of real defects and prediction of service life.

[0158] like Figure 2 As shown, this application also discloses a visual inspection system for the quality of wood-plastic composite flooring, comprising:

[0159] The image acquisition module 210 is used to acquire a surface image of the wood-plastic composite flooring and, based on a visual recognition model, identify potential defect areas in the surface image.

[0160] The feature acquisition module 220 is used to perform non-contact physical detection on the potential defect area to obtain the response features of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response features include acoustic features and thermal features;

[0161] The defect identification module 230 is used to verify whether the potential defect area is a real defect based on the acoustic and thermal characteristics, and to generate a verification result.

[0162] The visual calibration module 240 is used to adjust and calibrate the recognition logic parameters of the visual recognition model based on the verification results.

[0163] The result output module 250 is used to extract feature information of potential defect areas that have been verified as real defects, associate them with a pre-established set of durability rules that characterize the deterioration behavior of materials under outdoor environmental stress, predict the service life range and / or durability risk level of the wood-plastic flooring, and output the prediction results.

[0164] The image acquisition module 210 can be configured with an industrial camera to capture high-resolution surface images of the wood-plastic composite (WPC) flooring surface. These images are then transmitted to an image processing unit for preprocessing. A visual recognition model, deployed in the image acquisition module 210, analyzes the preprocessed surface images, identifies areas that statistically differ from the predefined texture features of normal WPC flooring, and marks these areas as potential defect areas. As an optional implementation, the image acquisition module 210 can also integrate a multispectral imaging device to acquire images containing richer spectral information, thereby providing the visual recognition model with more dimensional feature data.

[0165] The feature acquisition module 220 integrates an acoustic probe and a thermal probe. The acoustic probe can employ an ultrasonic sensor to emit high-frequency sound waves towards the potential defect area and receive the echo signals to extract acoustic features. The thermal probe can consist of an infrared thermal imager and a local heating device. By locally heating the potential defect area and monitoring its surface temperature change over time in real time, thermal features are extracted. The feature acquisition module 220 can also be equipped with a precision motion control platform, such as an XYZ-axis robotic arm driven by a servo motor, to accurately position the acoustic and thermal probes directly above the potential defect area, thereby achieving localized, high-precision physical detection.

[0166] The defect identification module 230 receives acoustic and thermal feature data from the feature acquisition module 220 and has built-in identification logic. This identification logic can be implemented based on preset threshold rules or through a trained classifier model (such as a support vector machine or neural network). This classifier model can comprehensively analyze multi-dimensional features and output the probability of the presence of defects.

[0167] The visual calibration module 240 achieves adaptive optimization of the model by establishing a feedback loop. Specifically, images of potential defect regions that have been verified as real defects or non-defects, along with their corresponding verification results, are stored as labeled training samples. When the cumulative number of collected training samples reaches a preset threshold, the visual calibration module 240 triggers a retraining or incremental learning process for the visual recognition model. During this process, the internal parameters of the visual recognition model, such as the weights of the feature extraction algorithm, the decision boundary of the classifier, or the recognition threshold, are adjusted and optimized based on these new, physically verified samples, thereby improving the model's accuracy and robustness in recognizing novel defects.

[0168] The result output module 250 receives confirmation information of real defects from the defect identification module 230 and further extracts their visual morphological features and physical field signal features. These features, along with preset outdoor environmental conditions, are input into the durability rule set. The durability rule set can be a database, an expert system, or a prediction model, storing the mapping relationship between different defect types, degrees, and material degradation rates. Based on the matched degradation rate, combined with the visual morphological features of the defects and preset product failure thresholds, the result output module 250 calculates the estimated time interval required for the WPC flooring to reach a failure state and maps it to the corresponding durability risk level. Finally, the result output module 250 outputs the prediction results, including the service life interval and / or durability risk level, in the form of a report, display, or data interface, providing a basis for production decisions and product quality traceability.

[0169] It should be noted that the information interaction and execution process between the above modules are based on the same concept as the method embodiments of this application. For details on their specific functions and technical effects, please refer to the method embodiments section, which will not be repeated here.

[0170] The preferred embodiments of this application have been described in detail above, but this application is not limited to the above embodiments. Those skilled in the art can make various equivalent modifications or substitutions without departing from the spirit of this application.

Claims

1. A visual inspection method for the quality of wood-plastic composite flooring, characterized in that, Includes the following steps: Acquire surface images of the wood-plastic composite flooring and identify potential defect areas in the surface images based on a visual recognition model; Non-contact physical detection is performed on the potential defect area to obtain the response characteristics of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response characteristics include acoustic characteristics and thermal characteristics; Based on the acoustic and thermal characteristics, the system verifies whether the potential defect region is a real defect and generates verification results. Based on the verification results, the recognition logic parameters of the visual recognition model are adjusted and calibrated. The feature information of potential defect areas verified as real defects is extracted and correlated with a pre-established set of durability rules that characterize the degradation behavior of materials under outdoor environmental stress. The service life range and / or durability risk level of the wood-plastic flooring are predicted, and the prediction results are output.

2. The visual inspection method for the quality of wood-plastic composite flooring according to claim 1, characterized in that, The step of acquiring a surface image of the wood-plastic composite flooring and identifying potential defect areas in the surface image based on a visual recognition model includes: Using an industrial camera, the surface of the wood-plastic composite flooring is photographed to obtain a surface image of the wood-plastic composite flooring; The surface image undergoes preprocessing including grayscale conversion, noise removal, and contrast enhancement. Based on a visual recognition model, an anomaly detection method based on local texture differences is used to analyze the preprocessed surface image and identify potential defect areas that have statistical differences from the preset normal wood-plastic composite flooring texture features.

3. The visual inspection method for the quality of wood-plastic composite flooring according to claim 2, characterized in that, The steps of using an anomaly detection method based on local texture differences to analyze the preprocessed surface image and identify potential defect areas that statistically differ from the preset normal texture features of wood-plastic composite flooring include: The preprocessed surface image is divided into multiple local image blocks; Calculate the local second-order moment features for each of the local image patches, wherein the local second-order moment features include at least one of energy, entropy, and moment of inertia; Calculate the deviation of the local second-order moment feature from the preset statistical distribution of normal wood-plastic composite wood flooring texture features; Local image blocks whose deviation values ​​exceed a preset deviation threshold are marked as abnormal image blocks; The abnormal image blocks that are spatially adjacent are combined to form at least one connected region, and each connected region is identified as a potential defect region. The adjustable parameters included in the recognition logic of the visual recognition model include the statistical distribution of the texture features of the normal wood-plastic composite flooring and / or the deviation threshold.

4. The visual inspection method for the quality of wood-plastic composite flooring according to claim 1, characterized in that, The acoustic features include the amplitude attenuation rate of the echo signal, and the thermal features include the attenuation rate of the surface temperature. The step of verifying whether the potential defect region is a real defect based on the acoustic and thermal characteristics, and generating a verification result, includes: When the amplitude attenuation rate exceeds a preset attenuation threshold and the attenuation rate is lower than a preset rate threshold, a verification result is generated that the potential defect area is a real defect.

5. The visual inspection method for the quality of wood-plastic composite flooring according to claim 1, characterized in that, The acoustic features include at least one of the following: the dominant frequency offset of the echo signal, the waveform distortion index, and the phase change; the thermal features include at least one of the following: the multi-point temperature gradient change rate, the thermal diffusion anisotropy index, and the surface temperature fluctuation frequency. The step of verifying whether the potential defect region is a real defect based on the acoustic and thermal characteristics, and generating a verification result, includes: Based on the acoustic and thermal features, a multidimensional feature vector is constructed. The multidimensional feature vector is input into a trained pattern recognition model, and the probability value of the potential defect region being a real defect is output; wherein, the pattern recognition model is a Bayesian classifier or a fuzzy logic reasoning model. When the probability value exceeds a preset confidence threshold, a verification result is generated indicating that the potential defect area is a real defect.

6. The visual inspection method for the quality of wood-plastic composite flooring according to claim 1, characterized in that, The step of adjusting and calibrating the recognition logic parameters of the visual recognition model based on the verification results includes: The generated verification results are bound to the corresponding potential defect area images to form labeled training samples; When the preset calibration triggering conditions are met, the visual recognition model is retrained or incrementally learned using the training samples in order to adjust and calibrate the parameters of the visual recognition model. The calibration triggering condition includes at least the following: the cumulative number of training samples collected reaches a preset threshold.

7. The visual inspection method for the quality of wood-plastic composite flooring according to claim 1, characterized in that, The steps of extracting feature information of potential defect areas verified as real defects, associating it with a pre-established set of durability rules characterizing the degradation behavior of materials under outdoor environmental stress, predicting the service life range and / or durability risk level of the wood-plastic flooring, and outputting the prediction results include: Visual morphological features and physical field signal features of potential defect regions verified as real defects are extracted as feature information; wherein, the visual morphological features include at least one of length, width, and curvature, and the physical field signal features include at least one of the acoustic features of echo signal amplitude attenuation rate and the thermal features of surface temperature attenuation rate. The extracted feature information and preset outdoor environmental conditions are input together into a pre-established durability rule set that characterizes the degradation behavior of materials under outdoor environmental stress; wherein, the durability rule set stores correlation relationships, and the correlation relationships define the mapping between different combinations of visual morphological features, physical field signal features and outdoor environmental conditions and the corresponding degradation rate. The durability rule set is used to match the corresponding correlation based on the input feature information and outdoor environmental conditions to obtain the degradation rate corresponding to the potential defect area; Based on the deterioration rate, the visual morphological characteristics of the potential defective areas, and the preset product failure threshold, the estimated time interval required for the wood-plastic flooring to reach a failure state is calculated. The estimated time interval is used as the estimated service life interval of the wood-plastic flooring, and mapped to the corresponding durability risk level; The output includes the predicted service life range and / or durability risk level.

8. The visual inspection method for the quality of wood-plastic composite flooring according to claim 7, characterized in that, The physical detection also includes spectral detection, and the response characteristics also include chemical composition characteristics obtained based on spectral detection; Before the step of calculating the estimated time interval required for the wood-plastic flooring to reach a failure state based on the deterioration rate, the visual morphological characteristics of the potential defect area, and a preset product failure threshold, the following steps are also included: Based on the characteristics of the chemical components, a preset chemical correction mapping table is consulted to obtain the corresponding chemical correction coefficients; Multiply the degradation rate obtained by matching by the chemical correction factor to obtain the corrected degradation rate; The degradation rate used to subsequently calculate the estimated time interval required for the wood-plastic composite flooring to reach a failure state is a corrected degradation rate.

9. A visual inspection method for the quality of wood-plastic composite flooring according to claim 8, characterized in that, The physical detection is based on a local detection device, which includes an integrated component and a motion platform for driving the integrated component. The integrated component includes an acoustic probe, a thermal probe, and a spectral probe. The process for obtaining the response features includes: Control the motion platform to position the integrated component directly above the potential defect area; The acoustic probe, thermal probe, and spectral probe are driven synchronously to emit sound waves, perform local heating, and emit excitation light into the potential defect area, respectively, and receive echo signals, acquire temperature change sequences, and collect scattering spectral signals accordingly. The acoustic features are extracted from the echo signal, the thermal features are extracted from the temperature change sequence, and the chemical component features are extracted from the scattering spectrum signal.

10. A visual inspection system for the quality of wood-plastic composite flooring, characterized in that, include: The image acquisition module is used to acquire surface images of the wood-plastic composite flooring and, based on a visual recognition model, identify potential defect areas in the surface images. The feature acquisition module is used to perform non-contact physical detection on the potential defect area to obtain the response features of the potential defect area; wherein, the physical detection includes acoustic detection and thermal detection, and the response features include acoustic features and thermal features; The defect identification module is used to verify whether the potential defect area is a real defect based on the acoustic and thermal characteristics, and to generate a verification result. The visual calibration module is used to adjust and calibrate the recognition logic parameters of the visual recognition model based on the verification results. The results output module is used to extract feature information of potential defect areas that have been verified as real defects, associate them with a pre-established set of durability rules that characterize the deterioration behavior of materials under outdoor environmental stress, predict the service life range and / or durability risk level of the wood-plastic flooring, and output the prediction results.

Images