A wood board edge sealing intelligent defect detection system

Abstract

The application discloses a kind of wood board edge banding intelligent defect detection systems, user management module includes user login submodule and user registration submodule;Edge banding defect detection module includes model selection submodule, detection submodule, data statistics submodule and detection result submodule;Database includes model file submodule, data storage submodule and user data submodule;User login submodule and user registration submodule are connected with user data submodule;Model selection submodule input connection model file submodule, output connection detection submodule;Detection submodule output connection data statistics submodule and detection result submodule;Detection result submodule output connection data storage submodule.The application not only effectively integrates the improved target detection model, but also provides automated, intelligent technical support for quality control in wood product manufacturing through modular, engineered software design, laying a solid foundation for subsequent system deployment and industrial applications.

Description

Technical Field

[0001] This invention belongs to the field of wood board defect detection technology, and relates to an intelligent defect detection system for wood board edge banding. Background Technology

[0002] In the modern furniture manufacturing and engineered wood processing industry, edge banding is a crucial step that determines the appearance, texture, moisture resistance, and structural stability of the boards. Due to the high speed of production lines and the variety of board colors and patterns, edge banding is prone to defects such as chipping, glue overflow, separation, and glue gaps. For a long time, the inspection of wood board edge banding quality has relied heavily on manual visual inspection. This not only leads to low inspection efficiency and high labor intensity, but also makes it difficult to achieve quantitative and standardized quality control because the inspection standards are greatly affected by subjective factors.

[0003] With the evolution of computer vision technology, deep learning-based target detection models (such as the YOLO series) have made automatic defect identification possible. However, in the specific industrial vertical scenario of wood panel edge banding, existing technologies still face three severe challenges: First, interference from complex textures. The natural mineral lines and annual ring textures of wood are extremely similar to scratches and glue seams in the feature space. Conventional convolutional neural networks often struggle to distinguish normal textures from physical defects when extracting features, resulting in a high false detection rate. Second, the special nature of defect morphology. Edge banding defects often appear as extremely narrow strips or tiny dots. After multiple downsampling, their spatial information is easily lost, causing small targets to be missed. The most critical problem lies in the disconnect between "detection" and "management." Existing academic models are mostly isolated algorithmic logics, lacking supporting industrial-grade software systems. In actual production lines, enterprises not only need to know "where the defects are," but also need a complete closed-loop system integrating identity authentication, multi-mode switching (video / dynamic stream), real-time data statistics, and graded evaluation functions to transform detection results into traceable quality archives. Without a deep integration of algorithms and software architecture, it will be difficult to truly solve the "last mile" problem in the intelligent transformation of furniture manufacturing. Summary of the Invention

[0004] To solve the above problems, the technical solution of the present invention is: an intelligent defect detection system for wood panel edge banding, comprising a user management module, an edge banding defect detection module, and a database connected in sequence, wherein, The user management module includes a user login submodule and a user registration submodule; The edge banding defect detection module includes a model selection submodule, a detection submodule, a data statistics submodule, and a detection result submodule. The database includes a model file submodule, a data storage submodule, and a user data submodule; The user login submodule and user registration submodule are both connected to the user data submodule; the input of the model selection submodule is connected to the model file submodule, and the output is connected to the detection submodule; the output of the detection submodule is connected to the data statistics submodule and the detection result submodule; the output of the detection result submodule is connected to the data storage submodule.

[0005] Preferably, the user management module uses QDialog as the main window container and QStackedWidget to switch between the login interface and the registration interface.

[0006] Preferably, the database is an SQLite database.

[0007] Preferably, the model selection submodule calls the defect detection model from the model file submodule and sends it to the detection submodule, which then performs defect detection using the defect detection model.

[0008] Preferably, the detection submodule includes image detection, video detection, and real-time detection.

[0009] Preferably, the data statistics submodule performs real-time quantity aggregation and dynamic updates of the defect types identified during the detection process.

[0010] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention, by constructing an improved YOLO-MDEW model and developing a corresponding system for detecting and evaluating defects in wood panel edge banding, has achieved significant benefits both technically and in industrial applications. First, it achieves accurate perception and robust identification of minute, slender defects against complex texture backgrounds. Addressing the extreme aspect ratios and susceptibility to wood grain interference inherent in wood panel edge banding defects, this invention introduces a category-aware attention mechanism and a multi-scale feature fusion module at the algorithm's underlying level. Experimental data shows that the improved YOLO-MDEW model achieves over 62.6% accuracy in the mAP50 index, with significantly enhanced perception sensitivity compared to the baseline model, especially when handling minute glue seams as narrow as 0.5mm. This improvement not only reduces false alarms caused by fluctuations in ambient light or changes in board color, but also ensures an extremely low false negative rate during high-speed production line operation, providing highly confident underlying data for subsequent quality grading.

[0011] Secondly, a three-in-one intelligent quality management closed loop integrating "detection, statistics, and evaluation" has been constructed. Unlike previous single models that could only output coordinate frames, this system integrates real-time counting and visualization distribution modules. Through real-time synchronization between the detection thread and the statistical structure, the system can intuitively display the distribution of various defects (such as edge chipping at 23.41% and breakage at 12.05%) in the form of dynamic pie charts. This quantitative data feedback enables production managers to gain real-time insight into the health status of the production line, for example, by promptly inferring abnormalities in glue pot temperature or coating pressure through an increase in residual glue rate. This upgrade from "individual identification" to "systematic evaluation" significantly enhances the company's ability to optimize process parameters.

[0012] Third, the system achieves deep optimization and adaptation between algorithm efficiency and industrial hardware performance. During its design, the system fully considered hardware redundancy in industrial settings and optimized operators for high-performance GPUs such as the RTX 4090, achieving an inference speed of 62.5 FPS. This means the system can not only process static images but also perfectly adapt to real-time video stream detection and online production line monitoring modes. This millisecond-level response speed ensures that the automated sorting mechanism can act instantly based on the evaluation results, truly realizing online identification and automatic rejection of defective materials, significantly improving the automation rate and overall yield of the production line.

[0013] Fourth, the end-to-end digital recording ensures the rigor and security of quality traceability. The system's built-in identity authentication module and database storage logic guarantee the ownership and authenticity of each batch of test data. The location, image snapshot, and assessment level of each defect are structured and stored in the backend database, providing the company with an authoritative and intuitive chain of original evidence when responding to subsequent customer quality complaints or conducting production quality audits. Simultaneously, the system supports one-click switching between image, video, and real-time stream modes, greatly enhancing the software's versatility and adaptability across different production stages (such as incoming inspection, in-line inspection, and outgoing inspection). Attached Figure Description

[0014] Figure 1 This is a structural block diagram of the intelligent defect detection system for wood panel edge banding according to a specific embodiment of the present invention; Figure 2 This is a diagram showing the interface operation effect of the intelligent defect detection system for wood panel edge banding in the image detection mode, according to a specific embodiment of the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0016] Conversely, this invention encompasses any substitutions, modifications, equivalent methods, and solutions made within the spirit and scope of the invention as defined in the claims. Furthermore, to provide a better understanding of the invention, certain specific details are described in detail below. However, those skilled in the art will fully understand the invention even without these detailed descriptions.

[0017] See Figure 1 This is a structural block diagram of the intelligent defect detection system for wood panel edge banding of the present invention, which includes a user management module 10, an edge banding defect detection module 20, and a database 30 connected in sequence. The user management module 10 includes a user login submodule 11 and a user registration submodule 12; the edge banding defect detection module 20 includes a model selection submodule 21, a detection submodule 22, a data statistics submodule 23, and a detection result submodule 24; the database 30 includes a model file submodule 31, a data storage submodule 32, and a user data submodule 33. Among them, the user login submodule 11 and the user registration submodule 12 are both connected to the user data submodule 33; the input of the model selection submodule 21 is connected to the model file submodule 31, and the output is connected to the detection submodule 22; the output of the detection submodule 22 is connected to the data statistics submodule 23 and the detection result submodule 24; the output of the detection result submodule 24 is connected to the data storage submodule 32.

[0018] The user management module 10 uses QDialog as the main window container and QStackedWidget to switch between the login and registration interfaces. Database 30 is an SQLite database. The model selection submodule 21 retrieves the defect detection model from the model file submodule 31 and sends it to the detection submodule 22, which performs defect detection using this model. The detection submodule 22 includes image detection, video detection, and real-time detection. The data statistics submodule 23 performs real-time aggregation and dynamic updates of the defect types identified during the detection process.

[0019] This system has the following main advantages: (1) Integrated deep learning detection process. The system integrates modules for the entire process, including data import, model loading, detection inference and result display, and provides an intuitive user interface, which improves operational efficiency and usability.

[0020] (2) Parametric operation and visualization design. Users do not need a deep learning background to select models and optimize performance through graphical parameter adjustment, which simplifies the system usage process.

[0021] (3) Good scalability and data sharing capabilities. The system design supports modular updates and algorithm expansion, and can be connected to new detection models or data sources in the future, with good maintainability and secondary development potential.

[0022] Overall, this system can maintain efficient and stable detection performance in diverse and complex wood panel edge banding inspection scenarios, providing reliable technical support for the automation and intelligentization of wood product production lines.

[0023] In a specific embodiment, this system adopts a front-end and back-end separation design concept, clearly distinguishing image defect detection, data management, user management and interface display, so as to improve the maintainability, scalability and response speed of the system.

[0024] The backend is primarily responsible for core business logic, data processing, and user management, including the following functions: Defect detection service: The backend loads a YOLOv8 model to perform image detection tasks, outputting standardized results including defect category, confidence score, and coordinates. The detection process is executed in a separate thread, ensuring the system's stability and real-time performance when processing large batches or high-resolution images.

[0025] Data statistics and analysis: The backend maintains cumulative defect data and real-time statistical information, enabling the accumulation of the number of defects for each type and the calculation of percentages. The statistical results provide data support for pie charts and interface labels, and can be updated in real time with the detection results.

[0026] Historical data management: All detection results are uniformly recorded in the history manager by the backend, including image information, detection time, defect type, confidence level, and coordinate information. The backend provides an interface to support pagination, filtering, and deletion of historical records, achieving data persistence and traceability.

[0027] User Management: The backend provides user registration, login, and permission management functions. Each user's operations and historical records can be saved independently, ensuring data security and isolation in a multi-user environment. Access control can distinguish between ordinary users and administrators, and supports setting management permissions for historical records and models.

[0028] Frontend The front-end is responsible for user interaction and data visualization, and its main functions include: User interface: The front-end is built using the Qt framework, providing controls for image selection, detection initiation, threshold input, user login, and registration. The interface interacts with the back-end through a signal-slot mechanism to achieve asynchronous task processing and ensure smooth operation.

[0029] Defect statistics and visualization: The front end displays statistical data provided by the back end through tables, labels, and pie charts. Pie charts only display legend information (category and percentage) to avoid an overly cluttered interface; tables display historical data in paginated format, supporting filtering and clearing operations for easy user querying and analysis.

[0030] User management interaction: The front end provides user login, registration, password modification, and permission management interfaces, and synchronizes data through a backend interface. Regular users can view and manage their own historical records and monitoring tasks; administrators have management permissions for user data and system settings.

[0031] Data interaction and synchronization: The front-end obtains detection results, statistical data, and user information by calling the back-end interface, achieving real-time synchronization between the interface and the back-end data. After each detection is completed, the front-end automatically refreshes the interface display and updates the cumulative statistics and pie chart.

[0032] Front-end and back-end interaction mechanism: The front-end initiates an image detection request, the back-end executes the YOLOv8 model and returns the detection results; the back-end updates the results to the statistics module and the history management module 10, and the front-end refreshes the interface by obtaining data through the interface; the user management function is implemented by the front-end calling the back-end interface to realize login verification, access control and personal data maintenance; the front-end is only responsible for display and operation, and does not directly handle calculation tasks, thus decoupling the interface from business logic. User Management Module 10 is a crucial entry point for the wood panel edge banding defect detection system. It primarily handles user authentication and access control, ensuring the security and standardization of system operations. This module utilizes PyQt5 to build a graphical interface, stores user information in an SQLite database, and supports both login and registration functions, meeting the usage needs of different user roles (administrators / regular users).

[0033] In terms of interface design, the user management module 10 uses QDialog as the main window container and QStackedWidget to switch between the login and registration interfaces. The overall interface style is simple and clear, with a reasonable element layout, mainly including a title area, view switching button area, user input area, and operation button area. Users can freely switch between the corresponding forms using the "Login" and "Register" buttons.

[0034] The user login submodule 11 provides two input fields: username and password. After the user completes the input, the system will perform an integrity check to ensure that the username and password are not empty. The system will then hash and encrypt the entered password and compare it with the encrypted passwords stored in database 30. When the username and password match, the system will retrieve the current user's basic information (username, user type, etc.) and successfully log in to the main interface; if the match fails, the system will provide a corresponding prompt to remind the user to check the input.

[0035] The user registration submodule 12's interface allows new users to create accounts. This interface includes input fields for username, password, and confirm password, and offers the option of either an administrator or a regular user role. The system will sequentially verify the registration information as follows: username and password cannot be empty; the two entered passwords must match; the password length must meet the minimum security requirements; and the username cannot be the same as an existing account in database 30.

[0036] After all verifications are successful, the system will encrypt the password and write it along with the user's role into the database, thus creating the account. Upon successful registration, the user will be notified and the system will automatically redirect back to the login screen.

[0037] The login module has a built-in SQLite database 30 management function. Upon first run, it automatically creates the user database 30 file and user data tables as needed. The system also checks if the administrator account exists; if database 30 is empty, it automatically creates a default administrator account (username: admin, password: admin123) to ensure the system can be used normally in its initial state.

[0038] The system's interface adopts a modular layout design based on PyQt, and is divided into an upper navigation area, a middle page display area, and a right-side function control and data display area, forming a graphical detection platform with a clear structure, well-defined logic, and convenient operation.

[0039] The top area of ​​the interface serves as the global navigation bar, occupying approximately 20% of the window's height. It primarily contains buttons for switching between the "Detection Interface" and "History" pages. The navigation bar is divided into left and right sections, visually separated by a vertical dividing line, allowing users to quickly switch between different functional modules. Additionally, the right side of the navigation bar displays the currently logged-in user's information and provides a "Logout" button for user management and access control.

[0040] The core content of the main interface is managed through QStackedWidget, corresponding to the detection operation interface and the detection history interface respectively. The detection interface adopts a left-right split layout: the left area is the display window for images (or video frames, real-time camera footage), responsible for displaying input content and real-time inference results; the right area is the function control area, containing operation buttons for model loading, image detection, video detection, camera detection, etc., and providing a confidence threshold input box to support user-defined detection parameters. In the statistical display section, the system has set up two visualization modules: "Data Statistics" and "Defect Distribution." The former updates the detection quantity of various defects in real time in the form of a label list, while the latter dynamically displays the proportional distribution of defect categories through a pie chart, making the detection results more readable and intuitive.

[0041] The history page displays the system's saved inspection results data. This interface provides a table view showing detailed information such as inspection time, defect type, confidence level, and coordinates. It supports pagination and record clearing, facilitating user review and management of inspected data. The tables are non-editable, allow full row selection, and feature alternating background colors, enhancing the clarity and professionalism of the data display.

[0042] The system provides a model file selection interface, allowing users to choose a .pt format YOLO object detection model weight file from any local directory. After a model is selected, the system automatically invokes the YOLO inference engine, loads the model weights into memory, and synchronously displays the loading status on the interface (e.g., "Loading model…", "Model loaded successfully"). If a path error, permission issue, or corrupted weight file occurs during loading, the system will capture the exception and provide an error message, ensuring the stability and controllability of the system operation.

[0043] To enhance user experience, the system introduces the QSettings mechanism for local persistent storage. The model selection submodule 21 automatically saves the model's path after each successful model loading. The next time the user opens the system or executes the "Select Model" function again, the file selection window will default to the most recently used path directory, reducing the need to repeatedly search for model files and improving the efficiency of the detection process.

[0044] After the model is successfully loaded, the detection submodule 22 allows users to freely select three detection modes according to their actual needs: (1) Image detection: Any single image can be selected from a specified folder for defect analysis; (2) Video detection: It can browse and process video files in a specified folder and detect them frame by frame; (3) Real-time detection: The system will automatically call available cameras to perform continuous image acquisition and real-time inference.

[0045] After file selection or camera activation, the detection task will start automatically. The system interface uses a left-right split layout to intuitively present the detection process: the left window always displays the original input content, including imported images, video frame-by-frame content, or live camera footage; the right window synchronously displays the model's detection results. The system draws clear bounding boxes for each identified defect target and labels the corresponding category name and prediction confidence next to the box. At the same time, the information display window below the left function area dynamically refreshes the current detection statistics, displaying the number of various defects in a list format. This achieves synchronous presentation of original input, detection visualization, and statistical data, allowing users to comprehensively understand the detection process and results.

[0046] Figure 2 The interface demonstrates the operation of the image detection mode: the left window fully displays the imported wood board image, while the right window simultaneously presents the result image after model inference, with all identified edge sealing defects accurately labeled. The statistics window in the lower left corner of the interface updates defect data information in real time based on the detection content. The regional layout achieves a unified display of the original image, detection labels, and statistical data, helping users quickly grasp the overall detection situation.

[0047] During continuous video playback and processing, the statistics window dynamically updates the number of identified defect categories based on the model's inference results, providing users with immediate feedback. Once the video detection task is complete, the system automatically displays a "Video Detection Complete" message, clearly indicating the task status to the user. By combining dynamic updates of the detection process with task completion notifications, the system provides a clear overview of the entire video detection process.

[0048] Regarding system stability and functional correctness verification, test results showed that the detection module operated smoothly, the annotation results were clear and reliable, and all statistical data were successfully saved, fully demonstrating the system's robustness and accuracy in data processing and storage. The complete testing process confirmed the system's good applicability in different detection modes and diverse application scenarios. These test results clearly verified that all functions and performance meet design requirements, laying a solid foundation for further functional expansion and performance optimization.

[0049] The edge banding defect detection module 20 incorporates a complete data statistics submodule 23, used for real-time quantity aggregation and dynamic updates of defect types identified during the detection process. When the user selects an image, video, or real-time view for detection, the detection result submodule 24 automatically classifies and counts each target in the detection results and synchronously displays the statistical results in the information window in the lower left corner of the interface. While outputting results, the detection thread immediately updates the internal counting structure and presents the quantity of each type of defect in a list format on the interface, achieving synchronous feedback between detection and statistics. This design ensures the efficiency, timeliness, and stability of the data statistics process, providing a reliable foundation for subsequent data analysis and quality assessment. To enable users to more intuitively understand the overall structure of the detection data, this system constructs a defect distribution visualization module and displays the proportion of each type of defect in the overall detection results using a pie chart. After each detection is completed, the system automatically updates the pie chart based on the cumulative statistical results, plotting different defect categories according to their quantity proportions, allowing users to clearly see the frequency of occurrence and relative proportion of each type of defect.

[0050] In the pie chart, each type of defect is represented by a different sector. Users can directly view the percentage of each type of defect from the legend, such as "Cracked Edge: 23.41%" and "Broken Edge: 12.75%". This visualization method can not only quickly reflect the distribution of defect types in the inspection sample, but also be used to analyze the prevalence of various defects in the production environment, providing data support for quality management and process improvement.

[0051] In a specific embodiment, a historical statistics module is also included as an independent function of the system, specifically responsible for managing and displaying detailed records of each inspection. The module stores information including serial number, image name, inspection time, defect type, confidence level, and coordinate information. The interface displays historical data through a table view and provides pagination control, filtering, and clearing functions, enabling visual management and convenient operation of large amounts of historical data. After inspection is completed, each defect record is written to the historical data manager, and the table display is updated synchronously. The table is implemented using QTableView and QStandardItemModel, supporting single-row selection, non-editability, and alternating row color display. The pagination control manages browsing large amounts of data, displaying a fixed number of records per page, and users can turn pages using "previous page" and "next page" buttons. The clear history function requires user confirmation to avoid accidental operation.

[0052] This invention focuses on the overall design and functional implementation of a wood panel edge banding defect detection system, detailing the system's requirements analysis, software architecture, core modules, and key technology implementation. First, based on actual production scenarios, clear requirements were identified for the system's compatibility, reliability, and ease of operation. Leveraging the advantages of deep learning technology, an intelligent detection scheme centered on an improved YOLO-MDEW model was established. Subsequently, a front-end / back-end separation architecture was adopted, decoupling functional modules such as user management, defect detection, data statistics, and historical records, thereby improving the system's maintainability and scalability.

[0053] In terms of implementation, the system utilizes PyQt to build an intuitive and user-friendly graphical interface, supporting three detection modes: image, video, and camera. It also integrates model loading, real-time inference, result visualization, and data statistics into a unified process. The login module employs an SQLite database and a password hashing mechanism to ensure user authentication security. The detection module relies on the YOLOv8 inference engine to ensure high accuracy and robustness even in complex environments. The history and statistics module supports the traceability, analysis, and management of detection data through structured storage and paginated display.

[0054] The wood board edge banding defect detection system designed and implemented in this invention not only effectively integrates the improved target detection model, but also provides automated and intelligent technical support for quality control in wood product manufacturing through modular and engineered software design, laying a solid foundation for subsequent system deployment and industrial application.

[0055] In addition to the embodiments described above, the present invention may have other implementations. All technical solutions formed by equivalent substitution or equivalent transformation are within the scope of protection claimed by the present invention.

[0056] The present invention has been described in detail above, but its specific implementation is not limited thereto. Various modifications or alterations can be made by those skilled in the art without departing from the spirit and scope of the claims of this application.

Claims

1. A smart defect detection system for wood panel edge banding, characterized in that, This includes a user management module, an edge banding defect detection module, and a database, which are connected in sequence. The user management module includes a user login submodule and a user registration submodule; The edge banding defect detection module includes a model selection submodule, a detection submodule, a data statistics submodule, and a detection result submodule. The database includes a model file submodule, a data storage submodule, and a user data submodule; The user login submodule and user registration submodule are both connected to the user data submodule; the input of the model selection submodule is connected to the model file submodule, and the output is connected to the detection submodule; the output of the detection submodule is connected to the data statistics submodule and the detection result submodule; the output of the detection result submodule is connected to the data storage submodule.

2. The system according to claim 1, characterized in that, The user management module uses QDialog as the main window container and QStackedWidget to switch between the login interface and the registration interface.

3. The system according to claim 1, characterized in that, The database is an SQLite database.

4. The system according to claim 1, characterized in that, The model selection submodule calls the defect detection model from the model file submodule and sends it to the detection submodule, which then performs defect detection using the defect detection model.

5. The system according to claim 1, characterized in that, The detection submodule includes image detection, video detection, and real-time detection.

6. The system according to claim 1, characterized in that, The data statistics submodule performs real-time quantity aggregation and dynamic updates of the defect types identified during the detection process.

Images