Image data processing method for decorative panel inlay positioning

Abstract

The present application belongs to the technical field of image data processing, and discloses an image data processing method for decoration panel inlay positioning, comprising the following steps: S1, synchronously collecting two-dimensional image data and three-dimensional point cloud data of the decoration panel and the inlay; S2, performing grayscale correction, 3*3 kernel Gaussian filtering and adaptive threshold segmentation preprocessing on the two-dimensional image data, and adopting a statistical filtering algorithm to denoise and downsample the three-dimensional point cloud data, then performing coordinate registration with the preprocessed two-dimensional image data to establish a mapping relationship between pixel coordinates and physical coordinates; S3, inputting the preprocessed two-dimensional image data and the three-dimensional point cloud data into a CNN convolutional neural network model trained by a decoration panel inlay scene sample; S4, establishing an XYA reference coordinate library based on a decoration panel design drawing, calculating independent detection coordinates for four types of feature information respectively, and outputting final detection coordinates after fusion through a weighted average algorithm; and S5, calculating an uncompensated bias value of the final detection coordinates and the reference coordinates.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and more specifically to an image data processing method for the positioning of decorative panels. Background Technology

[0002] The surface inlay process of miniature decorative panels is a core step in the processing of high-end miniature home decorations and precision ornaments. The inlay precision directly determines the artistic value and reliability of the product. The core technical bottleneck of this process lies in the control of positioning accuracy—the decorative panels and inlays are tiny (the width and depth of the inlay grooves are mostly microscale), and it is necessary to simultaneously identify four types of features: decorative baseline, inlay groove outline, inlay shape, and texture alignment marks. Traditional image data processing methods are difficult to achieve accurate collaborative positioning of multiple features.

[0003] In existing technologies, image data processing often employs single 2D image recognition or simple 3D point cloud analysis, which has significant drawbacks: single 2D images are easily affected by the surface texture, color difference, and reflection of decorative panels, leading to deviations in the recognition of micro-scale features; single 3D point clouds are insufficient in recognizing two-dimensional features such as planar texture alignment marks, and point cloud data is prone to containing discrete noise; at the same time, there is a lack of standardized multi-feature coordinate fusion mechanisms, and the influence of environmental factors such as temperature and humidity on positioning is not considered, resulting in the final positioning offset value exceeding the preset accuracy threshold (±0.005mm), which in turn causes problems such as glue overflow and insufficient bonding strength, failing to meet the high-precision requirements of micro-decorative panel inlay.

[0004] To address this core technical issue, there is an urgent need for an image data processing method that integrates two-dimensional images and three-dimensional point cloud data, possesses multi-feature collaborative recognition and precise coordinate calculation capabilities, and can fundamentally improve the accuracy and stability of decorative panel inlay positioning. Summary of the Invention

[0005] The purpose of this invention is to provide an image data processing method that integrates two-dimensional images and three-dimensional point cloud data, and has the ability to perform multi-feature collaborative recognition and accurate coordinate calculation, in order to solve the problem of poor accuracy and stability in the above-mentioned decorative panel inlay positioning.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for image data processing for decorative panel inlay positioning includes the following steps:

[0008] S1. Synchronously acquire two-dimensional image data and three-dimensional point cloud data of decorative panels and inlays. The two-dimensional image data are surface images of decorative panels and inlays acquired from different perspectives, and the three-dimensional point cloud data are the outline shape data of decorative panels and inlays.

[0009] S2. Perform grayscale correction, 3×3 kernel Gaussian filtering and adaptive threshold segmentation preprocessing on the two-dimensional image data. Use statistical filtering algorithm to denoise and downsample the three-dimensional point cloud data. Then perform coordinate registration with the preprocessed two-dimensional image data to establish the mapping relationship between pixel coordinates and physical coordinates.

[0010] S3. Input the preprocessed two-dimensional image data and three-dimensional point cloud data into the CNN convolutional neural network model trained on the decorative panel inlay scene sample, and fuse three-dimensional depth and two-dimensional texture features to simultaneously identify four types of feature information: decorative baseline, inlay groove outline, inlay shape and texture alignment mark.

[0011] S4. Establish an XYA reference coordinate library based on the decorative panel design drawings, calculate independent detection coordinates for the four types of feature information, and output the final detection coordinates after fusion by a weighted average algorithm.

[0012] S5. Calculate the uncompensated offset value between the final detection coordinates and the reference coordinates, and correct it by substituting it into the temperature compensation formula in combination with the real-time operating temperature of the equipment to obtain the compensated offset value.

[0013] S6. Use the root mean square error algorithm to verify the positioning accuracy of the compensated offset value. If it exceeds the preset accuracy threshold, return to step S1 to re-collect data until the positioning accuracy meets the preset requirements.

[0014] Preferably, when acquiring two-dimensional image data in step S1, based on the real-time detection results of the surface material, texture density and reflectivity of the decorative panel, four different wavelength supplementary light sources—450nm blue light, 530nm green light, 650nm red light and 780nm near-infrared light—are dynamically switched, and the light intensity of the light source is adaptively adjusted within a preset range according to a fixed gradient.

[0015] Preferably, the statistical filtering denoising algorithm formula used for the 3D point cloud data in step S2 is: , among which, among which, Let p be the set of 3D point cloud data points retained after filtering. This is a set separator, indicating that subsequent conditions are met. Let p be the number of points in the neighborhood of p. Let p and its neighboring points Euclidean distance, The mean distance between neighboring points. This represents the distance from the standard deviation.

[0016] The statistical filtering denoising algorithm removes discrete noise points that deviate from the mean by more than three times the standard deviation.

[0017] Preferably, the coordinate registration in step S2 adopts the iterative nearest point algorithm, and a dynamic weight factor is introduced during the registration process to optimize the iterative convergence speed. The formula for calculating the dynamic weight factor is: ,in, For the first Weighting factor for the next iteration The initial weighting factor has a value between 0.8 and 1.0. This is the attenuation coefficient, with a value ranging from 0.05 to 0.1. This represents the number of iterations.

[0018] Coordinate registration unifies the coordinate system of 3D point cloud data with the pixel coordinate system of 2D image data to the physical coordinate system, with a registration error ≤0.002mm and a registration time ≤0.5s.

[0019] Preferably, in step S3, the shallow convolutional layers of the CNN convolutional neural network model extract basic features of edges and textures through 3×3 convolutional kernels, the middle convolutional layers enhance local feature responses through dilated convolutions, and the deep convolutional layers fuse three-dimensional depth information and two-dimensional texture features through a cross-dimensional attention mechanism to achieve complementary enhancement of the two types of data. Furthermore, the CNN convolutional neural network model has a built-in feature confidence evaluation mechanism. When the confidence of a single type of feature recognition does not reach the preset confidence threshold, the backup feature extraction channel is automatically called to re-extract and calculate the feature for that type.

[0020] Preferably, the feature fusion loss function formula for the CNN convolutional neural network model in step S3 is: ,in, The total loss function for feature fusion is... Cross-entropy loss for 2D image feature recognition The mean square error loss for 3D point cloud feature recognition. The contrast loss for fusion of 2D and 3D features. , , Here are the loss weighting coefficients, and .

[0021] Preferably, the XYA reference coordinate library in step S4 includes four types of features: X-axis reference coordinates, Y-axis reference coordinates, rotation reference angle around the Z-axis, and allowable deviation range. The fusion formula of the weighted average algorithm is: , , ,in, , , These are the X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis of the final detected data after fusion.

[0022] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the decorative baseline features;

[0023] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the mosaic groove contour features;

[0024] , , These are the independent detection of the X-axis coordinate, Y-axis coordinate, and rotation angle around the Z-axis for the shape features of the inlay;

[0025] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for texture alignment marker features;

[0026] Simultaneously calculate the contribution rate of the deviation between each feature detection coordinate and the reference coordinate: , , ,in, , , They represent the first The contribution rate of the deviation of the class feature in the X-axis, Y-axis, and rotation angle directions around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. The corresponding serial numbers for the four types of features are: 1 for the decorative baseline, 2 for the inlay groove outline, 3 for the inlay shape, and 4 for the texture alignment mark. These are the traversal indices for the four types of features. This indicates that the absolute values ​​of the corresponding deviations of the four types of features are summed.

[0027] Preferably, the formula for calculating the uncompensated bias value in step S5 is: , , The temperature compensation formula is: , , ,in,

[0028] , , These are the preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis in the product design drawings.

[0029] , , These are the uncompensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively.

[0030] , , These are the temperature-compensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively.

[0031] This refers to the real-time operating temperature of the machine. 25 represents the coefficient of thermal expansion of the decorative panel material, and 25 represents the standard reference temperature, i.e., the product design reference temperature.

[0032] Simultaneously, a humidity compensation factor is introduced to perform a secondary correction on the bias value after temperature compensation: , , , ,in, As a humidity compensation factor, Humidity influence coefficient This represents the actual ambient humidity; 50% is the standard reference humidity. , , These are the X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis after secondary correction.

[0033] Preferably, the root mean square error algorithm formula in step S6 is: ,in,

[0034] RMSE is the root mean square error of the positioning system, used to quantify positioning accuracy; 0.01 is the angle error conversion factor, which converts the rotation angle offset into an equivalent length offset to ensure the uniformity of error calculation dimensions.

[0035] The preset positioning accuracy threshold is 0.005mm. When RMSE>0.005mm, return to step S1 to trigger re-acquisition and positioning, and optimize the supplementary light source parameters and point cloud acquisition density.

[0036] Preferably, the process also includes step S7: after the positioning accuracy verification is qualified, the original two-dimensional positioning image, three-dimensional contour data image, reference coordinates, final detection coordinates, compensated offset value, processing parameters of each step, environmental temperature and humidity data, and feature recognition confidence level are associated with and stored in a distributed database along with the QR code of the decorative panel product; at the same time, the compensated offset value is transmitted to the actuator in real time to realize the linkage control between positioning and subsequent processes. During the linkage process, a parameter abnormality feedback mechanism is set. When the deviation of the linkage parameter reaches the abnormal threshold, an alarm is triggered and the execution of subsequent processes is suspended. Operation is resumed after confirmation or correction.

[0037] By adopting the above technical solution, the present invention has the following advantages compared with the prior art:

[0038] 1. This invention provides an image data processing method for positioning decorative panel inlays. By acquiring and fusing two-dimensional images and three-dimensional point clouds, and combining the multi-feature synchronous recognition capability of the CNN convolutional neural network model, it achieves accurate recognition of four core features, including decorative baselines and inlay groove contours. With the addition of weighted average coordinate fusion, temperature and humidity compensation, and root mean square error verification, the positioning accuracy is strictly controlled within a preset threshold, thus solving the problem of excessive positioning deviation in traditional methods.

[0039] 2. This invention provides an image data processing method for decorative panel inlay positioning. It adopts a CNN convolutional neural network model trained with massive inlay scene samples, extracts basic features at the shallow level, and fuses 3D depth and 2D texture information at the deep level. It can simultaneously and accurately identify four types of core features, solving the feature recognition blind spot problem of single 2D or 3D processing. At the same time, through feature confidence evaluation and backup extraction channel design, it ensures the consistency and accuracy of the four types of feature recognition and avoids positioning errors caused by the failure of single feature recognition.

[0040] 3. This invention provides an image data processing method for decorative panel inlay positioning. In the data acquisition stage, the image quality is optimized by dynamic supplementary lighting. In the coordinate correction stage, a temperature and humidity compensation factor is introduced to effectively counteract the interference of surface reflection and environmental temperature and humidity changes on positioning, and adapt to decorative panel inlay scenarios with different materials and surface characteristics.

[0041] 4. This invention provides an image data processing method for decorative panel inlay positioning. It integrates multi-feature detection coordinates through a weighted average algorithm, prioritizing the positioning priority of core reference features. At the same time, it introduces a temperature compensation formula to eliminate coordinate deviations caused by thermal expansion and contraction of the machine, making the coordinate calculation results more consistent with the actual working state of the equipment. Furthermore, it establishes a standardized root mean square error accuracy verification process, converting angle errors into length errors to achieve dimensional uniformity. It presets a high-precision threshold, and automatically re-collects data when the accuracy is not met, ensuring that the final positioning accuracy meets the stringent requirements of micro-decorative panel inlay. Attached Figure Description

[0042] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0043] 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.

[0044] Example

[0045] Please refer to Figure 1 As shown, this invention discloses an image data processing method for decorative panel inlay positioning, comprising the following steps:

[0046] S1. Simultaneously acquire two-dimensional image data and three-dimensional point cloud data of decorative panels and inlays. The two-dimensional image data are surface images of decorative panels and inlays acquired from different perspectives, and the three-dimensional point cloud data are the outline and shape data of decorative panels and inlays.

[0047] S2. Perform grayscale correction, 3×3 kernel Gaussian filtering and adaptive threshold segmentation preprocessing on the two-dimensional image data. Use statistical filtering algorithm to denoise and downsample the three-dimensional point cloud data. Then perform coordinate registration with the preprocessed two-dimensional image data to establish the mapping relationship between pixel coordinates and physical coordinates.

[0048] S3. Input the preprocessed two-dimensional image data and three-dimensional point cloud data into the CNN convolutional neural network model trained on the decorative panel inlay scene sample, and fuse three-dimensional depth and two-dimensional texture features to simultaneously identify four types of feature information: decorative baseline, inlay groove outline, inlay shape and texture alignment mark.

[0049] S4. Establish an XYA reference coordinate library based on the decorative panel design drawings, calculate independent detection coordinates for the four types of feature information, and output the final detection coordinates after fusion by a weighted average algorithm.

[0050] S5. Calculate the uncompensated offset value between the final detection coordinates and the reference coordinates, and correct it by substituting it into the temperature compensation formula in combination with the real-time operating temperature of the equipment to obtain the compensated offset value.

[0051] S6. Use the root mean square error algorithm to verify the positioning accuracy of the compensated offset value. If it exceeds the preset accuracy threshold, return to step S1 to collect data again until the positioning accuracy meets the preset requirements.

[0052] S7. After the positioning accuracy verification is qualified, the original two-dimensional positioning image, three-dimensional contour data image, reference coordinates, final detection coordinates, compensated offset value, processing parameters of each step, environmental temperature and humidity data, and feature recognition confidence level are associated with and stored in the distributed database along with the QR code of the decorative panel product. At the same time, the compensated offset value is transmitted to the actuator in real time to realize the linkage control between positioning and subsequent processes. During the linkage process, a parameter abnormality feedback mechanism is set. When the deviation of the linkage parameter reaches the abnormal threshold, an alarm is triggered and the execution of subsequent processes is suspended. Operation is resumed after confirmation or correction.

[0053] Prerequisites for implementing this embodiment:

[0054] Hardware configuration: Two-dimensional image acquisition uses dual industrial cameras to acquire images from the vertical direction (front view) and the oblique direction (side view) respectively; three-dimensional point cloud acquisition uses a laser triangulation scanner to ensure the measurement accuracy of contour and topography data and the ability to capture details; the data processing unit adopts an architecture processor that supports parallel computing and real-time data processing to meet the needs of synchronous processing of multi-source data.

[0055] Software environment: The CNN convolutional neural network model is trained based on a deep learning framework. The training samples contain a large amount of data from different scenes, covering decorative panels of various materials, surface textures, inlay slots of various specifications, and inlay parts of various shapes. The number of model parameters meets the requirements for high-precision feature recognition, and the time consumption of a single feature recognition after deployment meets the requirements for real-time processing.

[0056] Preset parameters: The intensity adjustment range and gradient step size of the supplementary lighting source are set according to the actual scene requirements; the number of neighborhood points in the statistical filtering denoising algorithm is adapted to the point cloud data density; the initial weight factor and attenuation coefficient of the iterative nearest point algorithm are optimized according to the registration convergence efficiency; the weight coefficients of the feature fusion loss function of the CNN model satisfy the constraint that the summation is 1; the humidity influence coefficient, positioning accuracy threshold, and linkage parameter abnormality threshold are all set according to the accuracy requirements of the decorative panel inlay process. This embodiment takes the positioning scenario of a micro-decorative panel (including micro-inlay slots and inlay parts) as an example for specific explanation.

[0057] Step S1: Simultaneous acquisition of dual data

[0058] After the device is started, the multi-axis moving mechanism drives the acquisition module to a preset distance above the decorative panel to be positioned, and sends a position signal to trigger data acquisition:

[0059] Two-dimensional image acquisition: The surface characteristic detection unit acquires the panel surface material, texture density, and reflectivity in real time. Based on a preset switching logic, it automatically selects a supplementary light source with an appropriate wavelength, and the light intensity is adjusted to the optimal value according to a fixed gradient. The main-view camera and the side-view camera simultaneously acquire surface images to ensure spatiotemporal consistency. When acquiring two-dimensional image data, based on the real-time detection results of the decorative panel surface material, texture density, and reflectivity, multiple sets of supplementary light sources with different wavelengths are dynamically switched, and the light intensity of the light source is adaptively adjusted within a preset range according to a fixed gradient.

[0060] 3D point cloud acquisition: The laser triangulation scanner is started to simultaneously acquire the contour data of the decorative panel inlay groove (including geometric parameters such as width, depth, and groove wall tilt angle) and inlay (including shape parameters such as edge chamfer radius). The point cloud data output frequency meets the requirements for complete capture of contour details.

[0061] When acquiring two-dimensional image data, based on the real-time detection results of the surface material, texture density and reflectivity of the decorative panel, four different wavelength supplementary light sources—450nm blue light, 530nm green light, 650nm red light and 780nm near-infrared light—are dynamically switched, and the light intensity of the light source is adaptively adjusted within a preset range according to a fixed gradient.

[0062] Step S2: Data preprocessing and coordinate registration

[0063] Two-dimensional image preprocessing: The acquired 2D image is sequentially subjected to grayscale correction (to eliminate grayscale deviation caused by uneven illumination), 3×3 kernel Gaussian filtering (to smooth image noise and preserve edge features), and adaptive threshold segmentation (the threshold is dynamically adjusted according to local texture) to finally obtain a clear binary image of the feature region, eliminating the interference of surface defects on subsequent recognition.

[0064] 3D Point Cloud Denoising and Downsampling: A statistical filtering algorithm is used to denoise the 3D point cloud data, removing discrete noise points (mainly environmental dust and laser reflection interference points) that deviate from the mean by three times the standard deviation. A downsampling algorithm is used to retain the core contour data, improving processing efficiency while maintaining accuracy. The formula for the statistical filtering denoising algorithm used for the 3D point cloud data is as follows: , among which, among which, Let p be the set of 3D point cloud data points retained after filtering. This is a set separator, indicating that subsequent conditions are met. Let p be the number of points in the neighborhood of p. Let p and its neighboring points Euclidean distance, The mean distance between neighboring points. This represents the distance from the standard deviation.

[0065] The statistical filtering denoising algorithm removes discrete noise points that deviate from the mean by more than three times the standard deviation.

[0066] Coordinate registration: The iterative nearest-point algorithm is used to register the preprocessed 3D point cloud data with the 2D image data. During the registration process, a dynamic weight factor is introduced to optimize the iterative convergence speed. The formula for calculating the dynamic weight factor is: ,in, For the first Weighting factor for the next iteration The initial weighting factor has a value between 0.8 and 1.0. This is the attenuation coefficient, with a value ranging from 0.05 to 0.1. This represents the number of iterations.

[0067] Coordinate registration unifies the coordinate system of 3D point cloud data with the pixel coordinate system of 2D image data to the physical coordinate system, with a registration error ≤0.002mm and a registration time ≤0.5s.

[0068] Step S3: Simultaneous recognition of multiple features

[0069] The registered 2D image data and 3D point cloud data are input into the trained CNN convolutional neural network model, and the model performs feature extraction according to a three-level architecture of shallow, medium, and deep layers.

[0070] Shallow convolutional layers: Extract basic features such as decorative baseline edges and tessellation groove textures using 3×3 convolutional kernels;

[0071] Intermediate convolutional layers: dilated convolutions are used to enhance the response of local features such as the shape of the mosaicks and texture alignment markers;

[0072] Deep convolutional layers: By using a cross-dimensional attention mechanism module, weights are assigned to 3D depth information (such as tessellation depth and tessellation height) and 2D texture features (such as texture alignment mark patterns) respectively, and then they are fused to achieve complementary enhancement of the two types of data.

[0073] The model has a built-in feature confidence evaluation mechanism. When the confidence of a single feature recognition does not reach the preset confidence threshold, the backup feature extraction channel is automatically called to re-extract and calculate the feature for that type, ensuring the accuracy and reliability of the four types of feature recognition.

[0074] The feature fusion loss function formula for the CNN convolutional neural network model in step S3 is as follows: ,in, The total loss function for feature fusion is... Cross-entropy loss for 2D image feature recognition The mean square error loss for 3D point cloud feature recognition. The contrast loss for fusion of 2D and 3D features. , , Here are the loss weighting coefficients, and .

[0075] Step S4: Coordinate Fusion and Deviation Contribution Rate Calculation

[0076] XYA Reference Coordinate Library Establishment: Based on the decorative panel design drawings, an XYA reference coordinate library containing four types of features is established. This library includes the X-axis reference coordinates, Y-axis reference coordinates, rotation reference angles around the Z-axis, and allowable deviation ranges for the four types of features. The allowable deviation ranges are set according to the precision requirements of the inlay process.

[0077] Independent detection coordinate calculation: Calculate the independent detection coordinates for the four types of features identified in step S3 to obtain the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the decorative baseline, inlay groove outline, inlay shape, and texture alignment mark.

[0078] Weighted average fusion: A weighted average algorithm is used to fuse the independent detection coordinates of the four types of features. The fusion formula for the weighted average algorithm is as follows: , , ,in, , , These are the X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis of the final detected data after fusion.

[0079] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the decorative baseline features;

[0080] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the mosaic groove contour features;

[0081] , , These are the independent detection of the X-axis coordinate, Y-axis coordinate, and rotation angle around the Z-axis for the shape features of the inlay;

[0082] , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for texture alignment marker features;

[0083] Simultaneously calculate the contribution rate of the deviation between each feature detection coordinate and the reference coordinate: , , ,in, , , They represent the first The contribution rate of the deviation of the class feature in the X-axis, Y-axis, and rotation angle directions around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. The corresponding serial numbers for the four types of features are: 1 for the decorative baseline, 2 for the inlay groove outline, 3 for the inlay shape, and 4 for the texture alignment mark. These are the traversal indices for the four types of features. This indicates that the absolute values ​​of the corresponding deviations of the four types of features are summed.

[0084] Step S5: Offset Calculation and Temperature / Humidity Compensation

[0085] The formula for calculating the uncompensated bias value is: , , The temperature compensation formula is: , , ,in,

[0086] , , These are the preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis in the product design drawings.

[0087] , , These are the uncompensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively.

[0088] , , These are the temperature-compensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively.

[0089] This refers to the real-time operating temperature of the machine. 25 represents the coefficient of thermal expansion of the decorative panel material, and 25 represents the standard reference temperature, i.e., the product design reference temperature.

[0090] Simultaneously, a humidity compensation factor is introduced to perform a secondary correction on the bias value after temperature compensation: , , , ,in, As a humidity compensation factor, Humidity influence coefficient This represents the actual ambient humidity; 50% is the standard reference humidity. , , These are the X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis after secondary correction.

[0091] Step S6: Positioning accuracy verification and re-acquisition mechanism

[0092] The root mean square error (RMSE) algorithm is used to verify the positioning accuracy of the compensated offset value. The formula for the RMSE algorithm is as follows: ,in,

[0093] RMSE is the root mean square error of the positioning system, used to quantify positioning accuracy; 0.01 is the angle error conversion factor, which converts the rotation angle offset into an equivalent length offset to ensure the uniformity of error calculation dimensions.

[0094] The preset positioning accuracy threshold is 0.005mm. When RMSE > 0.005mm, the system returns to step S1 to trigger re-acquisition and positioning, and optimizes the supplementary lighting parameters and point cloud acquisition density to ensure better data quality in the re-acquisition. If RMSE is within the preset accuracy threshold range, the positioning accuracy is acceptable, and re-acquisition is not required.

[0095] Step S7: Data storage and linkage control

[0096] Data storage: After the positioning accuracy verification is passed, the original 2D positioning image, 3D contour data image, XYA reference coordinates, final detection coordinates, compensated offset value, processing parameters for each step, environmental temperature and humidity data, and feature recognition confidence level are associated with the QR code of the decorative panel product and stored in a distributed database. The data storage duration meets the production traceability requirements and supports multi-dimensional queries by product number, production date, material type, etc.

[0097] Linkage control: The compensated bias value is transmitted in real time to the subsequent actuators (such as dispensing actuators) via an industrial bus. The actuators adjust their motion trajectory according to the bias value, achieving seamless linkage between positioning and subsequent processes. A parameter anomaly feedback mechanism is set up during the linkage process to monitor linkage parameter deviations in real time. When the deviation reaches the linkage parameter anomaly threshold, an alarm is immediately triggered and subsequent processes are paused. Operation resumes only after manual confirmation or parameter correction, ensuring product processing quality.

[0098] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An image data processing method for positioning and embedding decorative panels, characterized in that, Includes the following steps: S1. Synchronously acquire two-dimensional image data and three-dimensional point cloud data of decorative panels and inlays. The two-dimensional image data are surface images of decorative panels and inlays acquired from different perspectives, and the three-dimensional point cloud data are the outline shape data of decorative panels and inlays. S2. Perform grayscale correction, 3×3 kernel Gaussian filtering and adaptive threshold segmentation preprocessing on the two-dimensional image data. Use statistical filtering algorithm to denoise and downsample the three-dimensional point cloud data. Then perform coordinate registration with the preprocessed two-dimensional image data to establish the mapping relationship between pixel coordinates and physical coordinates. S3. Input the preprocessed two-dimensional image data and three-dimensional point cloud data into the CNN convolutional neural network model trained on the decorative panel inlay scene sample, and fuse three-dimensional depth and two-dimensional texture features to simultaneously identify four types of feature information: decorative baseline, inlay groove outline, inlay shape and texture alignment mark. S4. Establish an XYA reference coordinate library based on the decorative panel design drawings, calculate independent detection coordinates for the four types of feature information, and output the final detection coordinates after fusion by a weighted average algorithm. S5. Calculate the uncompensated offset value between the final detection coordinates and the reference coordinates, and correct it by substituting it into the temperature compensation formula in combination with the real-time operating temperature of the equipment to obtain the compensated offset value. The formula for calculating the uncompensated bias value in step S5 is as follows: , , The temperature compensation formula is: , , ,in, , , These are the X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis of the final detected data after fusion. , , These are the preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis in the product design drawings. , , These are the uncompensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively. , , These are the temperature-compensated X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis, respectively. This is the real-time operating temperature of the machine. 25 represents the coefficient of thermal expansion of the decorative panel material, and 25 represents the standard reference temperature, i.e., the product design reference temperature. Simultaneously, a humidity compensation factor is introduced to perform a secondary correction on the bias value after temperature compensation: , , , ,in, As a humidity compensation factor, Humidity influence coefficient This represents the actual ambient humidity; 50% is the standard reference humidity. , , These are the X-axis offset value, Y-axis offset value, and rotation offset angle around the Z-axis after secondary correction, respectively. S6. Use the root mean square error algorithm to verify the positioning accuracy of the compensated offset value. If it exceeds the preset accuracy threshold, return to step S1 to re-collect data until the positioning accuracy meets the preset requirements.

2. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: When acquiring two-dimensional image data in step S1, based on the real-time detection results of the surface material, texture density and reflectivity of the decorative panel, four different wavelength supplementary light sources—450nm blue light, 530nm green light, 650nm red light and 780nm near-infrared light—are dynamically switched, and the light intensity of the light source is adaptively adjusted within a preset range according to a fixed gradient.

3. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: The statistical filtering denoising algorithm formula used for the 3D point cloud data in step S2 is as follows: ,in, Let p be the set of 3D point cloud data points retained after filtering. This is a set separator, indicating that subsequent conditions are met. Let p be the number of points in the neighborhood of p. Let p and its neighboring points Euclidean distance, The mean distance to neighboring points. This represents the distance from the standard deviation. The statistical filtering denoising algorithm removes discrete noise points that deviate from the mean by more than three times the standard deviation.

4. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: The coordinate registration in step S2 adopts the iterative nearest point algorithm. During the registration process, a dynamic weight factor is introduced to optimize the iterative convergence speed. The formula for calculating the dynamic weight factor is: ,in, For the first Weighting factor for the next iteration The initial weighting factor has a value between 0.8 and 1.

0. This is the attenuation coefficient, with a value ranging from 0.05 to 0.

1. This represents the number of iterations. Coordinate registration unifies the coordinate system of 3D point cloud data with the pixel coordinate system of 2D image data to the physical coordinate system, with a registration error ≤0.002mm and a registration time ≤0.5s.

5. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: In step S3, the shallow convolutional layers of the CNN convolutional neural network model extract basic features of edges and textures through 3×3 convolutional kernels, the middle convolutional layers enhance local feature responses through dilated convolutions, and the deep convolutional layers fuse three-dimensional depth information and two-dimensional texture features through a cross-dimensional attention mechanism to achieve complementary enhancement of the two types of data. Furthermore, the CNN convolutional neural network model has a built-in feature confidence evaluation mechanism. When the confidence of a single type of feature recognition does not reach the preset confidence threshold, the backup feature extraction channel is automatically called to re-extract and calculate the feature for that type.

6. The image data processing method for decorative panel inlay positioning as described in claim 5, characterized in that: The feature fusion loss function formula for the CNN convolutional neural network model in step S3 is as follows: ,in, The total loss function for feature fusion is... Cross-entropy loss for 2D image feature recognition The mean square error loss for 3D point cloud feature recognition. The contrast loss for fusion of 2D and 3D features. , , Here are the loss weighting coefficients, and .

7. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: The XYA reference coordinate library mentioned in step S4 includes four types of features: X-axis reference coordinates, Y-axis reference coordinates, rotation reference angle around the Z-axis, and allowable deviation range. The fusion formula of the weighted average algorithm is as follows: , , ,in, , , These are the X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis of the final detected data after fusion. , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the decorative baseline features; , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for the mosaic groove contour features; , , These are the independent detection of the X-axis coordinate, Y-axis coordinate, and rotation angle around the Z-axis for the shape features of the inlay; , , These are the independent detection X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis for texture alignment marker features; Simultaneously calculate the contribution rate of the deviation between each feature detection coordinate and the reference coordinate: , , ,in, , , They represent the first The contribution rate of the deviation of the class feature in the X-axis, Y-axis, and rotation angle directions around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first Independent detection of class features: X-axis coordinates, Y-axis coordinates, and rotation angle around the Z-axis. , , The first The class features are defined in the XYA reference coordinate library with preset X-axis reference coordinates, Y-axis reference coordinates, and rotation reference angle around the Z-axis. The corresponding serial numbers for the four types of features are: 1 for the decorative baseline, 2 for the inlay groove outline, 3 for the inlay shape, and 4 for the texture alignment mark. These are the traversal indices for the four types of features. This indicates that the absolute values ​​of the corresponding deviations of the four types of features are summed.

8. The image data processing method for decorative panel inlay positioning as described in claim 7, characterized in that: The root mean square error algorithm formula described in step S6 is as follows: RMSE is the root mean square error of the positioning system, used to quantify the positioning accuracy; 0.01 is the angle error conversion coefficient, which converts the rotation angle offset into an equivalent length offset to ensure the uniformity of error calculation dimensions. The preset positioning accuracy threshold is 0.005mm. When RMSE>0.005mm, return to step S1 to trigger re-acquisition and positioning, and optimize the supplementary light source parameters and point cloud acquisition density.

9. The image data processing method for decorative panel inlay positioning as described in claim 1, characterized in that: The process also includes step S7, after the positioning accuracy verification is qualified, storing the original two-dimensional positioning image, three-dimensional contour data image, reference coordinates, final detection coordinates, compensated offset value, processing parameters of each step, environmental temperature and humidity data, and feature recognition confidence level in association with the QR code of the decorative panel product in a distributed database; at the same time, the compensated offset value is transmitted to the actuator in real time to realize the linkage control of positioning and subsequent processes. During the linkage process, a parameter abnormality feedback mechanism is set. When the deviation of the linkage parameter reaches the abnormal threshold, an alarm is triggered and the execution of subsequent processes is suspended. Operation is resumed after confirmation or correction.

Images