Intelligent control method and system for full-automatic plate edge sealing equipment

Abstract

The invention provides an intelligent control method and system for full-automatic plate edge sealing equipment, and relates to the technical field of intelligent control, and the method comprises the steps: carrying out the median filtering of super-resolution edge coordinates, then carrying out the clockwise sorting, forming an ordered set, calculating curvature marking segmentation points, dividing edge segments, and carrying out the segmentation of the edge segments; randomly selecting points to fit a circle by taking each section as a unit, optimizing to obtain a final arc model, mapping to an original image coordinate system along sampling points of the arc model, and combining to form a super-resolution edge coordinate set after fitting; acquiring a real-time state parameter of the edge sealing equipment, and performing normalization processing on the state parameter to obtain a correction factor for correcting the measurement deviation; and correcting the fitted super-resolution edge coordinate set through a correction factor to obtain a corrected edge coordinate set. According to the invention, high-precision identification and edge sealing control on the edge of the plate are realized, and the edge sealing precision and efficiency are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent control, and particularly to an intelligent control method and system for a fully automatic plate edge banding device. Background Art

[0002] In the production process of panel furniture, the quality of edge banding not only directly affects the aesthetics of the furniture, but also relates to the durability and environmental performance of the furniture. Traditional plate edge banding devices mainly adopt mechanical or simple automatic control methods, which have many defects in practical applications and are difficult to meet the requirements of high precision and high efficiency in modern furniture production.

[0003] For example, due to the complex and variable edge shapes of profiled plates, traditional devices are difficult to accurately control the fitting degree between the edge banding strip and the plate edge, resulting in obvious gaps and black lines at the edge banding, which not only affects the aesthetics of the furniture, but also reduces the market competitiveness of the product. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide an intelligent control method and system for a fully automatic plate edge banding device, which realizes high-precision recognition and edge banding control of the plate edge, and improves the precision and efficiency of edge banding.

[0005] To solve the above technical problem, the technical solution of the present invention is as follows:

[0006] In a first aspect, an intelligent control method for a fully automatic plate edge banding device, the method comprising:

[0007] Step 1, constructing an octree structure according to the original plate image resolution and the preset octree depth and mapping pixel values, performing multi-layer decomposition on the nodes by a filter to obtain a sub-band set, performing morphological processing on the high-frequency sub-bands to obtain a binary edge map, connecting the edge points into curves by the Hough transform and interpolating to extract super-resolution edge coordinates;

[0008] Step 2, performing median filtering on the super-resolution edge coordinates and sorting them in clockwise order to form an ordered set, calculating the curvature to mark the segmentation points and dividing the edge segments, randomly selecting points for each segment to fit a circle and optimizing to obtain the final circular arc model, and mapping the sampled points along the circular arc model back to the original image coordinate system and merging to form a set of fitted super-resolution edge coordinates;

[0009] Step 3, obtaining the real-time state parameters of the edge banding device, performing normalization processing on the state parameters to obtain a correction factor for correcting measurement deviations;

[0010] Step 4, correcting the set of fitted super-resolution edge coordinates by the correction factor to obtain a corrected set of edge coordinates;

[0011] Step 5: Calculate the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions based on the corrected edge coordinate set, and obtain the dimension deviation score.

[0012] Step 6: Mark the sheets that do not meet the requirements according to the dimension deviation score, and control the equipment to perform corresponding processing.

[0013] Furthermore, construct an octree structure based on the resolution of the original sheet image and the preset octree depth, map the pixel values, and perform multi-layer decomposition on the nodes through filters to obtain a set of sub-bands, including:

[0014] Construct an octree structure according to the resolution of the original sheet image and the preset octree depth. Each node in the octree structure represents a pixel block in the original sheet image.

[0015] Map the pixel values of the original sheet image block by block into the nodes of the octree structure.

[0016] For each node of the octree, perform convolution operations through the low-pass filter and high-pass filter of the Haar wavelet respectively. After each layer of decomposition, a low-frequency sub-band and three high-frequency sub-bands are generated.

[0017] Use the low-frequency sub-band obtained from the current layer of decomposition as the input for the next layer of decomposition; redistribute the data of the low-frequency sub-band into the nodes of the octree, adjust the node size to match the resolution after the current decomposition, repeat the filtering operation on the remapped low-frequency sub-band to generate new low-frequency and high-frequency sub-bands. If the current layer number has not reached the preset layer number, increment the layer number and repeat the filtering operation; if it has reached, terminate the decomposition to obtain a set of sub-bands after multi-layer decomposition. The set of sub-bands includes the low-frequency sub-band of the final layer and the high-frequency sub-bands of each layer.

[0018] Furthermore, perform morphological processing on the high-frequency sub-bands to obtain a binary edge map, determine the edge points, and use the Hough transform to connect them into curves and interpolate to extract super-resolution edge coordinates, including:

[0019] Perform morphological processing on the high-frequency sub-bands. In the high-frequency sub-bands after morphological processing, determine the edge points through threshold segmentation.

[0020] Input the edge points into the Hough transform, identify collinear points and fit them into continuous curves, and merge adjacent curve segments to form a continuous edge curve.

[0021] On the continuous edge curve, obtain the interpolated edge point coordinates through linear interpolation.

[0022] Extract the interpolated edge point coordinates to form a super-resolution edge coordinate set.

[0023] Further, perform median filtering on the super-resolution edge coordinates, sort them in clockwise order to form an ordered set, and calculate curvature markers to divide the edge segments at the segmentation points, including:

[0024] Perform median filtering on the super-resolution edge coordinates, sort the filtered coordinate points in the clockwise direction of the edge curve to form an ordered set of edge coordinates;

[0025] Calculate the local curvature of the ordered coordinate points by cubic spline interpolation, traverse all points, and mark them as segmentation points when the curvature exceeds the threshold;

[0026] Divide the edge into multiple continuous segments according to the segmentation points.

[0027] Further, randomly select points for each segment to fit a circle and optimize to obtain the final arc model, map the sampled points along the arc model back to the original image coordinate system and merge them to form the fitted super-resolution edge coordinate set, including:

[0028] Extract the set of coordinate points within each segmentation segment as an independent fitting unit;

[0029] According to the set of coordinate points of each segmentation segment, randomly select 3 points from the coordinate points of the current segment, calculate the center and radius of the fitted circle, and calculate the distance from all points in the segment to the current arc model;

[0030] Mark the points with a distance less than the threshold as inliers, and re-optimize the center and radius by the least squares method using all inliers to obtain the final arc model for each segmentation segment;

[0031] For the final arc model of each segmentation segment, uniformly sample points along the arc path with a super-resolution step size;

[0032] Map the sampled points back to the original image coordinate system, and merge the dense coordinate point sets of all segments in the original sorting order to form the fitted super-resolution edge coordinate set.

[0033] Further, obtain the real-time state parameters of the edge-sealing device, perform normalization processing on the state parameters to obtain a correction factor for correcting measurement deviations, including:

[0034] Obtain the state parameters of the current edge-sealing device, and the state parameters include the numerical values of roller wear, temperature, pressure, motor speed, and sensor calibration error;

[0035] Perform normalization processing on each state parameter to obtain the normalized state parameter;

[0036] Determine the central value and sensitivity of the Gaussian function corresponding to each state parameter according to the normalized state parameters; calculate the Gaussian weight of each state parameter according to the central value and sensitivity of the Gaussian function corresponding to each state parameter, and normalize the Gaussian weight of each state parameter to obtain the normalized Gaussian weight of each state parameter;

[0037] For each state parameter, calculate the corresponding average value and standard deviation from the historical data;

[0038] Calculate the coefficient of variation according to the average value and standard deviation corresponding to each state parameter, and calculate the initial weight of each state parameter using the coefficient of variation;

[0039] Multiply the normalized Gaussian weight value by the corresponding initial weight to obtain the final weight corresponding to each state parameter;

[0040] Multiply each normalized state parameter by the corresponding final weight and accumulate to obtain the correction factor.

[0041] Furthermore, correct the fitted super-resolution edge coordinate set through the correction factor to obtain the corrected edge coordinate set, including:

[0042] Traverse each point in the fitted coordinate set, multiply the horizontal coordinate x and vertical coordinate y of each point by the correction factor to calculate the corrected coordinates.

[0043] Furthermore, according to the corrected edge coordinate set, calculate the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions, and obtain the dimension deviation score, including:

[0044] Calculate the curvature between adjacent points for the corrected coordinate set, identify the curvature mutation points, and divide the coordinate set into straight line segments and arc segments according to the curvature threshold;

[0045] For the straight line segments, accumulate the lengths of all straight line segments to obtain the total length and width of the sheet; for the arc segments, take the average value of the fillet radii as the overall fillet feature;

[0046] Calculate the length deviation and width deviation according to the total length and width of the sheet; calculate the fillet radius deviation according to the overall fillet feature;

[0047] Calculate the corresponding length relative deviation, width relative deviation and fillet radius relative deviation according to the length deviation, width deviation and fillet radius deviation respectively;

[0048] Normalize the length relative deviation, width relative deviation and fillet radius relative deviation, and perform weighted summation on the normalized deviations of each item to generate the dimension deviation score.

[0049] Further, according to the dimensional deviation score, mark the sheets that do not meet the requirements and control the equipment to perform corresponding processing, including:

[0050] Set a scoring threshold. If the score of the sheet is > the scoring threshold, it is determined as unqualified, and a mark, deviation type, and score value are added to the surface of the sheet.

[0051] Transfer the marked sheets to the rework station and call the preset cutting parameters, grinding parameters, or shape correction parameters.

[0052] In a second aspect, an intelligent control system for a fully automatic sheet edge banding device includes:

[0053] A determination module for constructing an octree structure and mapping pixel values according to the original sheet image resolution and preset octree depth, decomposing the nodes through a filter in multiple layers to obtain a set of subbands, performing morphological processing on the high-frequency subbands to obtain a binary edge map, connecting the edge points into curves using the Hough transform, and interpolating to extract the super-resolution edge coordinates.

[0054] A calculation module for performing median filtering on the super-resolution edge coordinates and sorting them in clockwise order to form an ordered set, calculating the curvature to mark the segmentation points and dividing the edges into segments, randomly selecting points for each segment to fit a circle and optimizing to obtain the final circular arc model, and mapping the sampled points along the circular arc model back to the original image coordinate system and merging them to form a set of fitted super-resolution edge coordinates.

[0055] A normalization module for obtaining the real-time state parameters of the edge banding device and performing normalization processing on the state parameters to obtain a correction factor for correcting the measurement deviation.

[0056] A correction module for correcting the set of fitted super-resolution edge coordinates through the correction factor to obtain a corrected set of edge coordinates.

[0057] An evaluation module for calculating the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions according to the corrected set of edge coordinates, and obtaining the dimensional deviation score.

[0058] A control module for marking the sheets that do not meet the requirements according to the dimensional deviation score and controlling the equipment to perform corresponding processing.

[0059] In a third aspect, a computing device includes:

[0060] One or more processors;

[0061] A storage device for storing one or more programs, which when executed by the one or more processors, cause the one or more processors to implement the described method.

[0062] Fourthly, a computer-readable storage medium stores a program which, when executed by a processor, implements the method described above.

[0063] The above solution of the present invention has at least the following beneficial effects.

[0064] By constructing an octree structure and mapping pixel values, and combining multi-layer decomposition of filters and morphological processing, the present invention can accurately extract the binary image of the edge of the board, and then use the Hough transform to connect the edge points and interpolate to extract the super-resolution edge coordinates. This process significantly improves the accuracy of edge recognition.

[0065] After median filtering and sorting the super-resolution edge coordinates, the segmentation points are marked by calculating the curvature to divide the edge segments, and the circular arc model is fitted with each segment as a unit. This step not only optimizes the set of edge coordinates, but also makes the fitting of the edge more conform to the actual shape, improving the fitting degree and aesthetics of the edge banding.

[0066] The real-time state parameters of the edge banding device are obtained and normalized to obtain a correction factor for correcting the measurement deviation, enabling the present invention to adjust the edge banding operation in real time, compensate for the errors caused by changes in the device state, and thus improve the accuracy and stability of the edge banding.

[0067] The fitted super-resolution edge coordinate set is corrected by the correction factor to obtain a more accurate edge coordinate set. Based on these coordinate sets, the present invention can accurately calculate the actual geometric dimensions of the board and their deviations.

[0068] According to the size deviation score, the present invention can automatically mark the boards that do not meet the requirements and control the device to perform corresponding processing. This intelligent processing flow not only improves the production efficiency, but also reduces the cost and errors of manual intervention, making the edge banding process more automated and intelligent.

[0069] In summary, through a series of steps such as high-precision edge recognition, edge fitting and coordinate optimization, real-time state parameter correction, edge coordinate correction and size calculation, and size deviation scoring and intelligent processing, the present invention significantly improves the quality and production efficiency of board edge banding. At the same time, this method also reduces the requirements for the skills of operators and reduces the impact of human factors on product quality. Description of the Drawings

[0070] Figure 1 is a schematic flow chart of an intelligent control method for a full-automatic board edge banding device provided by an embodiment of the present invention. Detailed Embodiments

[0071] Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be completely conveyed to those skilled in the art.

[0072] As Figure 1 shown, an intelligent control method for a full-automatic board edge banding device according to an embodiment of the present invention includes the following steps:

[0073] Step 1: Construct an octree structure according to the original board image resolution and the preset octree depth, map pixel values, decompose the nodes through a filter in multiple layers to obtain a set of subbands, perform morphological processing on the high-frequency subbands to obtain a binary edge map, connect the edge points into curves by the Hough transform after determining the edge points, and interpolate to extract the super-resolution edge coordinates;

[0074] Step 2: Perform median filtering on the super-resolution edge coordinates, sort them in clockwise order to form an ordered set, calculate the curvature to mark the segmentation points to divide the edge segments, randomly select points for each segment to fit a circle and optimize to obtain the final circular arc model, and map the sampled points along the circular arc model back to the original image coordinate system and merge them to form a set of fitted super-resolution edge coordinates;

[0075] Step 3: Obtain the real-time state parameters of the edge banding device, perform normalization processing on the state parameters to obtain a correction factor for correcting measurement deviations;

[0076] Step 4: Correct the set of fitted super-resolution edge coordinates by the correction factor to obtain a corrected set of edge coordinates;

[0077] Step 5: Calculate the actual geometric dimensions of the board and the deviation of the actual geometric dimensions according to the corrected set of edge coordinates, and obtain a dimensional deviation score;

[0078] Step 6: Mark the boards that do not meet the requirements according to the dimensional deviation score, and control the device to perform corresponding processing.

[0079] In the embodiment of the present invention, in step 1, by constructing an octree structure according to the original sheet image resolution and the preset octree depth, the image data can be efficiently organized and stored. Mapping the pixel values retains the detailed information in the image, providing a rich data source for edge recognition. Performing multi-layer decomposition of the nodes by a filter can extract different frequency components in the image. In particular, the high-frequency subbands contain rich edge information, and this step helps to more accurately identify the edges in the subsequent process. Performing morphological processing on the high-frequency subbands can highlight the edge features and suppress the noise, obtaining a clear binary edge map, which greatly simplifies the complexity of edge recognition and improves the recognition accuracy. Connecting the edge points into curves by the Hough transform and interpolating to extract the super-resolution edge coordinates can accurately locate the edge positions.

[0080] In step 2, performing median filtering on the super-resolution edge coordinates can effectively remove the noise points, making the edges smoother. Sorting them in clockwise order to form an ordered set is convenient for subsequent edge fitting and processing. Calculating the curvature and marking the segmentation points can accurately divide the edge segments. This step helps to better adapt to edges of different shapes and improve the fitting degree of edge sealing. Randomly selecting points in each segment to fit a circle and obtaining the final circular arc model through optimization can more accurately describe the shape of the edge. Sampling points along the circular arc model and mapping them back to the original image coordinate system for merging forms the set of super-resolution edge coordinates after fitting, providing more accurate data for the edge sealing operation. By obtaining the real-time state parameters of the edge sealing device, the running state and performance changes of the device can be understood in a timely manner. Normalizing the state parameters can eliminate the dimensional differences between different parameters, facilitating comparison and analysis. The calculated correction factor can be used to correct the measurement deviation and improve the accuracy of edge sealing.

[0081] In step 4, correcting the set of super-resolution edge coordinates after fitting by the correction factor can eliminate the errors caused by changes in the device state, obtaining a more accurate set of edge coordinates. This step helps to improve the accuracy and stability of edge sealing. In step 5, according to the corrected set of edge coordinates, the actual geometric dimensions of the sheet can be accurately calculated, the deviation of the actual geometric dimensions can be calculated, and the dimensional deviation score can be obtained, which can intuitively reflect the dimensional accuracy and edge sealing quality of the sheet. This step helps to timely discover and process unqualified products. In step 6, according to the dimensional deviation score, the sheets that do not meet the requirements are marked. This step improves the production efficiency and product quality. Controlling the device to perform corresponding processing, such as adjusting the edge sealing parameters, replacing the edge sealing materials, etc., can automatically adapt to the edge sealing requirements of different sheets, improving the flexibility and adaptability of edge sealing. At the same time, this also reduces the requirements for the skills of operators and reduces the influence of human factors on product quality.

[0082] In a preferred embodiment of the present invention, an octree structure is constructed according to the resolution of the original sheet image and the preset octree depth, and pixel values are mapped. The nodes are decomposed by a filter in multiple layers to obtain a set of subbands, including:

[0083] According to the resolution of the original sheet image and the preset octree depth, an octree structure is constructed. Each node in the octree structure represents a pixel block in the original sheet image, specifically including: determining the resolution of the original sheet image, for example, the width is W and the height is H; according to the preset octree depth D, calculating the size of the pixel block represented by each node. Assuming the image is evenly divided into a 2 D ×2 D grid, then the size of the pixel block represented by each node is Initialize the octree structure. Each node contains its position information (such as coordinates), the size of the pixel block, and a pointer to the child node. Recursively construct the octree. Starting from the root node, divide the image into smaller sub-blocks according to the size of the pixel block, and create a child node for each sub-block until the preset octree depth is reached.

[0084] Map the pixel values of the original sheet image into the nodes of the octree structure block by block, specifically including: traversing each pixel of the original sheet image, calculating which pixel block represented by an octree node it belongs to according to its coordinates; adding the pixel value of this pixel to the total pixel value of the corresponding node; if the node represents a pixel block, the values of all pixels within the pixel block can be selected for storage.

[0085] For each node of the octree, perform convolution operations through the low-pass filter and high-pass filter of the Haar wavelet respectively. After each layer of decomposition, a low-frequency subband and three high-frequency subbands are generated, specifically including: for each node in the octree (representing a pixel block), obtain its pixel value (if the node stores the values of all pixels within the pixel block, directly obtain it; if it stores a feature representation, the pixel block needs to be reconstructed first); apply the low-pass filter and high-pass filter of the Haar wavelet to the pixel block for convolution operations. The Haar wavelet filter is a 2×2 matrix. The low-pass filter is used to smooth the image, and the high-pass filter is used to detect edges. After the convolution operation, four subbands are obtained: a low-frequency subband (LL) and three high-frequency subbands (LH, HL, HH), which represent different frequency components of the image respectively; store the four subbands obtained from the convolution operation respectively, where the low-frequency subband (LL) will be used for the next layer of decomposition, and the high-frequency subbands (LH, HL, HH) are retained as the high-frequency information of the current layer.

[0086] Use the low-frequency subband obtained by decomposing the current layer as the input for the next layer of decomposition; redistribute the data of the low-frequency subband into the nodes of the octree, adjust the node size to match the resolution after the current decomposition, repeat the filtering operation on the remapped low-frequency subband to generate new low-frequency and high-frequency subbands. If the current layer number has not reached the preset layer number, increment the layer number and repeat the filtering operation; if it has reached, terminate the decomposition to obtain the set of subbands after multi-layer decomposition, where the set of subbands includes the low-frequency subband of the final layer and the high-frequency subbands of each layer, specifically including:

[0087] Extract pixel values from the low-frequency subband (LL) obtained by decomposing the current layer, and use these pixel values as the input for the next layer of decomposition, that is, as the pixel values of the nodes of the octree in the next layer; adjust the size of the octree nodes according to the resolution after the current decomposition. Since each layer of decomposition halves the resolution of the image (in both the horizontal and vertical directions), the size of the nodes also needs to be halved accordingly. Redistribute the pixel values of the low-frequency subband into the adjusted octree nodes, which requires reconstructing part or all of the octree to adapt to the new resolution; apply the low-pass filter and high-pass filter of the Haar wavelet to the remapped low-frequency subband again for convolution operations, and repeat the operations to generate a new low-frequency subband and three high-frequency subbands;

[0088] Maintain a counter for the current layer number. After each layer of decomposition is completed, check whether the current layer number has reached the preset layer number. If not, increment the layer number counter and return to continue the decomposition. If the preset layer number has been reached, terminate the decomposition process; during the decomposition process, save the high-frequency subbands (LH, HL, HH) generated in each layer. When the decomposition terminates, also save the low-frequency subband (LL) of the final layer. Combine all the saved subbands into a set of subbands as the input for subsequent processing. This process realizes the efficient representation and feature extraction of the original sheet metal image through steps such as constructing the octree structure, mapping pixel values, and multi-layer filter decomposition.

[0089] In the embodiments of the present invention, the octree structure can efficiently divide the three-dimensional space (or the hierarchical division of a two-dimensional image in this scenario), and each node represents a pixel block in the original sheet metal image. This division method helps to quickly locate and access specific regions in the image, improving the efficiency of subsequent processing. By presetting the octree depth, image representations at different scales can be constructed, which helps to consider both the local details and global features of the image in subsequent processing, improving the accuracy of edge recognition. Mapping the pixel values into the octree nodes enables the orderly organization of image data, which helps to quickly access and modify the pixel values in specific regions during subsequent processing, improving the processing efficiency. Block-by-block mapping preserves the detailed information in the image, providing a rich data source for subsequent edge recognition and super-resolution reconstruction. Through the convolution operation of the low-pass filter and the high-pass filter, the image can be decomposed into components of different frequencies. The low-frequency subband contains the main information and contours of the image, while the high-frequency subband contains the details and edge information of the image. The high-pass filter helps to highlight the edge features in the image, making subsequent edge recognition more accurate.

[0090] Each layer of decomposition generates a low-frequency subband and three high-frequency subbands, providing rich image information. By adjusting the number of decomposition layers, the richness of subband information and the degree of detail retention can be flexibly controlled, which helps to perform customized processing according to the requirements of different tasks. By gradually decomposing the low-frequency subband layer by layer, the feature representation of the image can be gradually refined, which helps to more accurately identify edge and detail information in subsequent processing. Since the low-frequency subband contains the main information of the image, decomposing it layer by layer can reduce the computational amount and improve the processing efficiency. As the number of decomposition layers increases, the resolution of the image gradually decreases. By adjusting the size of the octree nodes to match the current resolution after decomposition, an effective representation of the image data in the octree structure can be ensured. Redistributing the data of the low-frequency subband into the octree nodes maintains the consistency and integrity of the data. By repeating the filtering operation, the feature representation of the image can be continuously refined, improving the accuracy of edge recognition. Each layer of decomposition generates new low-frequency and high-frequency subbands, providing multi-layer feature information.

[0091] By presetting the number of layers to control the decomposition process, the required level of subband information can be ensured. Terminating the decomposition when the preset number of layers is reached can avoid the increase in computational amount and information loss caused by over-decomposition, which helps to maintain the balance between processing efficiency and accuracy. The set of subbands after multi-layer decomposition provides a rich image feature representation, including information such as contours, details, and edges.

[0092] In a preferred embodiment of the present invention, a binary edge map is obtained by morphological processing of the high-frequency subband. After determining the edge points, the Hough transform is used to connect them into curves and interpolate to extract the super-resolution edge coordinates, including:

[0093] Perform morphological processing on the high-frequency subband. In the high-frequency subband after morphological processing, determine the edge points through threshold segmentation, which specifically includes: processing the high-frequency subband through erosion operations. For example, a structuring element (such as a 3×3 rectangle or cross) can be used to perform erosion operations on the high-frequency subband to remove small noise points and highlight edge features. Perform threshold segmentation on the high-frequency subband after morphological processing, select an appropriate threshold, set the area with pixel values higher than the threshold to white (or 1), and the area lower than the threshold to black (or 0). Alternatively, the threshold can be automatically calculated through the Otsu algorithm. In the binary image, the white area represents the edge points. The pixel coordinates with a value of 1 can be recorded by traversing the image pixels as the edge points.

[0094] Input the edge points into the Hough transform to identify collinear points and fit them into continuous curves, and merge adjacent curve segments to form a continuous edge curve, which specifically includes: input the determined edge points into the Hough transform algorithm. The Hough transform can detect straight lines in the image, record the number of points on each possible straight line through an accumulator array, and select the straight line with a higher value in the accumulator array as the candidate straight line. For each candidate straight line, identify the edge points (collinear points) located on this straight line, and use these collinear points for curve fitting, such as fitting a straight line or polynomial curve using the least squares method; for adjacent curve segments, check their connectivity and direction consistency. If the angle between adjacent curve segments is less than a certain threshold and they are close enough in space, merge them into a continuous curve.

[0095] On the continuous edge curve, obtain the coordinates of the interpolated edge points through linear interpolation, which specifically includes: on the continuous edge curve, select the points to be interpolated. These points can be at any position on the curve, but usually the connection points between curve segments or key points on the curve are selected for interpolation; for each selected interpolation point, use its two adjacent known points for linear interpolation. The formula for linear interpolation is:

[0096]

[0097] where (x1, y1) and (x2, y2) are two adjacent known points, and (x, y) is the coordinate of the interpolation point. Through linear interpolation, the coordinates of the interpolated edge points can be obtained.

[0098] Extract the coordinates of the interpolated edge points to form a super-resolution edge coordinate set, which specifically includes: traverse the interpolated edge points, extract their coordinates. The coordinates can be expressed in the form of (x, y), where x and y are the horizontal and vertical coordinates of the edge points in the image respectively. Combine all the extracted edge point coordinates into a set, that is, the super-resolution edge coordinate set.

[0099] In the embodiments of the present invention, morphological processing (such as dilation, erosion, opening operation, closing operation, etc.) can effectively highlight the edge features in the high-frequency subbands, reduce noise interference, make the edges clearer. Through threshold segmentation, the high-frequency subbands after morphological processing can be converted into binary edge maps, making the determination of edge points simpler and more accurate. The combination of morphological processing and threshold segmentation can significantly improve the accuracy of edge detection. The Hough transform can identify collinear points among the edge points and thus fit them into continuous curves. Through the Hough transform, discrete edge points can be connected into continuous curves, better describing the shape and trend of the edges. For adjacent curve segments, they can be merged through a certain algorithm to form a more complete continuous edge curve, improving the coherence and accuracy of the edges; linear interpolation can generate more edge point coordinates on the continuous edge curve, thereby improving the resolution of the edges, making the edges smoother and more delicate. The interpolated edge point coordinates can better capture the detailed information of the edges, and the interpolated edge point coordinates are more dense and uniform; the super-resolution edge coordinate set contains richer edge information and can be used to improve the resolution and quality of the image. Through image reconstruction or edge rendering using the super-resolution edge coordinate set, a clearer and more delicate edge effect can be obtained, enhancing the visual effect of the image.

[0100] In a preferred embodiment of the present invention, median filtering is performed on the super-resolution edge coordinates and then sorted in clockwise order to form an ordered set, and curvature is calculated to mark segmentation points to divide edge segments, including:

[0101] Perform median filtering on the super-resolution edge coordinates, and sort the filtered coordinate points in the clockwise direction of the edge curve to form an ordered edge coordinate set. Specifically, it includes: determining the window size of the median filtering, usually an odd number, such as 3×3, 5×5, etc. The selection of the window size depends on the density of the edge coordinates and the noise level; for each edge coordinate point, take all the coordinate points (including this point itself) within a filtering window centered on it; sort the x coordinates and y coordinates of all the coordinate points within the filtering window respectively, and then take the median as the filtered coordinate of this point, and replace the original coordinate with the filtered coordinate to form the filtered edge coordinate set; select a suitable starting point, such as the point with the smallest x coordinate or the smallest y coordinate in the edge coordinate set, as the starting point for sorting. For each edge coordinate point, calculate the angle (polar angle) between the line connecting it and the starting point and the positive direction of the x axis, and the arctangent function (such as atan2) can be used to calculate the polar angle; sort all the edge coordinate points according to the size of the polar angle. If the polar angles are the same, secondary sorting can be performed according to the distance from the starting point, and the sorted edge coordinate points are combined into an ordered edge coordinate set.

[0102] Calculate the local curvature of ordered coordinate points by cubic spline interpolation, traverse all points, and mark the points as segmentation points when the curvature exceeds the threshold. Specifically, it includes: using the ordered edge coordinate set to construct a cubic spline interpolation function, which can fit a smooth curve passing through all ordered edge coordinate points. Calculate the first derivative and second derivative of the cubic spline interpolation function. The first derivative represents the slope of the curve, and the second derivative represents the curvature change rate of the curve. Calculate the local curvature at each coordinate point according to the second derivative and the first derivative. According to experience, set a suitable curvature threshold. When the curvature exceeds this threshold, consider this point as a segmentation point. Traverse all ordered edge coordinate points, calculate the curvature of each point, and if the curvature of a certain point exceeds the threshold, mark this point as a segmentation point. The specific application is as follows:

[0103] Arrange the set of ordered edge coordinate points \(\{(x_0,y_0),(x_1,y_1),\cdots,(x n ,y n )\) in clockwise or counterclockwise order; parameterize the curve as \(t\in[0,n]\), where \(n\) represents the number of edge coordinate points, and \(t = i\) corresponds to the \(i\)-th point \((x i ,y i )\);

[0104] Construct cubic spline functions \(S x (t)\) and \(S y (t)\) for \(x(t)\) and \(y(t)\) respectively, such that:

[0105] S x (t i ) = x i , S y (t i ) = y i \((i = 0,1,\cdots,n)\);

[0106] Interpolation method:

[0107] In each interval \([t i ,t i+1 \), define a cubic polynomial:

[0108] S(t)=a i +b i (t - t i )+c i (t - t i ) 2 +d i (t - t i ) 3 ;

[0109] Meet the continuity conditions:

[0110] Position continuity: S(ti ) = y i ;

[0111] First derivative continuous:

[0112] Second derivative continuous:

[0113] Boundary conditions (such as natural splines):

[0114] S″(t0) = S″(t n ) = 0;

[0115] Calculate the first derivative and the second derivative:

[0116] Taking S x (t) and S y (t) as an example:

[0117] First derivative:

[0118] S′ x (t) = b i + 2c i (t - t i ) + 3d i (t - t i ) 2 ;

[0119] S′ y (t) = b i + 2c i (t - t i ) + 3d i (t - t i ) 2 ;

[0120] Second derivative:

[0121] S″ x (t) = 2c i + 6d i (t - t i );

[0122] S″ y (t) = 2c i + 6d i (t - t i );

[0123] Calculate the local curvature, where the formula for the curvature κ(t) is:

[0124]

[0125] According to the specific application scenario, determine the threshold κ through statistical analysis h(For example: take the first 5% quantile of the curvature distribution, or manually adjust it to the optimal segmentation effect);

[0126] Traverse all points and mark the segmentation points. The traversal process is as follows:

[0127] For each point (x i , y i ), calculate its curvature κ at the corresponding parameter t = i i ; if κ i > κ h , mark (x i , y i ) as a segmentation point. Among them, the set of segmentation points is

[0128] where x(t) represents the x - coordinate of the ordered edge points expressed as a function of the parameter t; y(t) represents the y - coordinate of the ordered edge points expressed as a function of the parameter t; t i represents the parameter value corresponding to the i - th point. If index parameterization is adopted, t i = i (for example, t0 = 0, t1 = 1); t i+1 represents the parameter value corresponding to the i + 1 - th point, and the interval is [t i , t i+1 ; a i represents the constant term of the cubic polynomial within the interval [t i , t i+1 , and is equal to the value of S(t i ) (that is, x i or y i ); b i represents the coefficient of the first - order term and represents the first - order derivative value at t = t i ; c i represents the coefficient of the second - order term and represents half of the second - order derivative value at t = t i (because the second - order derivative is 2c i +6d i (t - t i )); d i represents the coefficient of the cubic term and controls the bending degree of the curve within the interval; t represents the parameterization variable, and its value range covers all ordered points. For example, t ∈ [0, n]; S′ x (t) represents the first - order derivative of x(t) with respect to the parameter t and represents the change rate in the x - direction; S′ y (t) represents the first - order derivative of y(t) with respect to the parameter t and represents the change rate in the y - direction; S″ x (t) represents the second - order derivative of x(t) with respect to the parameter t and represents the curvature change rate in the x - direction; S″ y (t) represents the second - order derivative of y(t) with respect to the parameter t and represents the curvature change rate in the y - direction.

[0129] The edge is divided into multiple continuous segments according to the segmentation points, which specifically includes: creating an empty segment list to store the divided edge segments, starting from the starting point of the ordered edge coordinate set, traversing all coordinate points, when encountering a segmentation point, adding the current segment (all coordinate points between the previous segmentation point and the current segmentation point) to the segment list and starting a new segment, after the traversal ends, adding the last segment (all coordinate points between the last segmentation point and the end point) to the segment list, each element in the segment list is a continuous edge segment, and these segments together constitute the divided edge segment set.

[0130] In the embodiment of the present invention, median filtering is an effective non-linear filtering method that can remove the noise points in the edge coordinates and retain the true structure of the edge, which is crucial for subsequent edge analysis and processing, because the noise points may interfere with the calculation of curvature and the accurate marking of segmentation points. Median filtering can also smooth the edge to a certain extent, making the edge more continuous and smooth, which helps to improve the accuracy and stability of subsequent curvature calculations. The ordered edge coordinate set makes subsequent processing (such as curvature calculation, segmentation point marking, etc.) more convenient and efficient, because the ordered coordinate set can ensure that the processing is carried out according to the actual trend of the edge, avoiding processing errors caused by the disorder of coordinate points; clockwise sorting can preserve the topological structure of the edge, that is, the starting point and end point of the edge and their connection relationship.

[0131] The cubic spline interpolation method can accurately fit the edge curve and calculate the local curvature. The cubic spline interpolation method can capture the subtle changes of the edge, including the sudden and gradual changes of curvature, which helps to more accurately identify the segmentation points of the edge. By setting the curvature threshold, the segmentation points of the edge can be automatically marked, and the edge can be divided into multiple continuous segments, which helps to more clearly understand the structure and characteristics of the edge. The selection of the curvature threshold can be adjusted according to the actual situation to ensure the accuracy and stability of the segmentation. By reasonably selecting the threshold, the problems of over-segmentation or under-segmentation can be avoided. After the edge is divided into multiple continuous segments, independent processing and analysis can be carried out for each segment, such as feature extraction, shape recognition, etc. Segment processing can significantly improve the processing efficiency.

[0132] In a preferred embodiment of the present invention, points are randomly selected for each segment to fit a circle and optimize to obtain the final circular arc model, and the sampled points along the circular arc model are mapped back to the original image coordinate system and merged to form the super-resolution edge coordinate set after fitting, including:

[0133] Extract the set of coordinate points within each segmented section as an independent fitting unit, which specifically includes: obtaining the edge segmentation section information divided in the previous steps. Each segmented section contains a set of consecutive coordinate points. For each segmented section, extract all the coordinate points within it to form an independent set of coordinate points, and store the set of coordinate points of each segmented section as an independent fitting unit.

[0134] According to the set of coordinate points of each segmented section, randomly select 3 points from the coordinate points of the current section, and calculate the center and radius of the fitted circle. For all points within the section, calculate their distances to the current circular arc model, which specifically includes: for each fitting unit (i.e., the set of coordinate points of the segmented section), randomly select 3 non-collinear points. Using these 3 points, calculate the center and radius of the fitted circle through geometric methods (such as the perpendicular bisector method). For all points within the fitting unit, calculate their distances to the center of the current fitted circle and compare them with the radius of the circle to obtain the distance of each point to the circular arc model. When specifically applied, it may specifically include:

[0135] Randomly select 3 points from the set of coordinate points of the fitting unit, denoted as A(x1,y1), B(x2,y2), C(x3,y3); ensure that these 3 points are non-collinear, which can be judged by calculating the cross product of the vectors and If the cross product is not zero, then the three points are non-collinear.

[0136] Calculate the equations of the perpendicular bisectors of line segments AB and AC:

[0137] The coordinates of the midpoint of line segment AB:

[0138] The slope of line segment AB: (when x1≠x2);

[0139] The slope of the perpendicular bisector of line segment AB (when k AB ≠0); if k AB =0, then the perpendicular bisector is a vertical line

[0140] The equation of the perpendicular bisector of line segment AB: Use the point-slope form where y and x represent the coordinate variables of any point on the perpendicular bisector; and represent the coordinates of the midpoint M AB of line segment AB, where: (the average value of the abscissas of the two endpoints of line segment AB); (the average value of the ordinates of the two endpoints of line segment AB).

[0141] Similarly, calculate the equation of the perpendicular bisector of segment AC.

[0142] Solve the perpendicular bisector equations of the line segments AB and AC and get the coordinates of the intersection (x0, y0), which is the center of the fitted circle.

[0143] Calculate the radius using the center (x0, y0) and an arbitrary point (such as point A)

[0144] For each point P(x i ,y i ), calculate its distance to the center of the circle (x0, y0):

[0145]

[0146] The calculated distance d i Compare with the radius r of the circle to get the distance difference (or residual) from each point to the arc model: Δd i =|d i -r|.

[0147] The points whose distance is less than the threshold are marked as inliers, and the center and radius are re-optimized by the least square method using all the inliers to obtain the final arc model of each segment. Specifically, it includes: according to experience, a suitable distance threshold is set, and the points whose distance is less than the threshold are marked as inliers. These points are considered to be points that meet the current arc model. All the inliers are used to re-fit the center and radius by the least square method to obtain a more accurate arc model. The least square method can minimize the sum of the squares of the distances from the inliers to the arc model, thereby obtaining the optimal center and radius. When applied specifically, it can include:

[0148] According to experience, a suitable distance threshold ∈ is set, which is used to judge whether the point conforms to the current arc model; for each point P in the fitting unit i (x i ,y i ), calculate its distance to the center of the current fitting circle (x0, y0) If d i <∈, then mark the point as an interior point; minimize the sum of the squares of the distances from the interior point to the arc model, that is, solve the following optimization problem:

[0149]

[0150] Among them, (x c ,y c ) is the new center coordinate, and r is the new radius.

[0151] Algebraic fitting method (based on the general equation of a circle):

[0152] General equation of a circle:

[0153] x 2 + y 2 + Dx + Ey + F = 0;

[0154] where the center coordinates are and the radius is

[0155] For each interior point (x i , y i ), substituting into the general equation of the circle gives:

[0156]

[0157] This can be regarded as a linear equation about D, E, and F.

[0158] Combine the equations of all interior points into a system of linear equations, and use the least squares method (such as solving by matrix or iterative optimization method) to solve this system of linear equations to obtain the optimal solutions of D, E, and F.

[0159] Calculate the center and radius:

[0160] Center coordinates:

[0161] Radius:

[0162] For the final circular arc model of each segmentation segment, uniformly sample points along the circular arc path with a super-resolution step size, specifically including:

[0163] For a circle or circular arc, use the parametric equation to generate sample points:

[0164]

[0165] where (x0, y0) are the center coordinates, r is the radius, θ is the parameter angle, set the angular step Δθ according to the super-resolution requirement, the smaller the step size, the denser the sample points, and the total number of sample points N is determined by the circular arc span Δθ total and the step size:

[0166]

[0167] Starting from the starting angle θ start , increment the step size Δθ in sequence to generate a parameter sequence:

[0168] θ i = θ start + i·Δθ (i = 0, 1, 2, …, N - 1);

[0169] Substitute into the parametric equation to calculate the coordinates of each sample point:

[0170] P i =(x0 + r * cos(θ i ), y0 + r * sin(θ i ));

[0171] where θ i is the parameter angle of the i-th sampling point, which is generated by starting from the starting angle θ start and incrementing by a fixed angular step Δθ each time. θ i determines the position of the sampling point on the circular arc; θ start is the parameter angle at the start of sampling, which usually corresponds to the angle of the starting point of the circular arc or circle relative to the center of the circle. In practical applications, θ start can be determined according to the starting position and direction of the circular arc; Δθ is the angular step, that is, the angular difference between adjacent sampling points, which determines the density of the sampling points. The smaller the step, the denser the sampling points and the better the super-resolution effect; i is the index of the sampling point, starting from 0 and incrementing, which is used to locate each sampling point in the parameter sequence; N is the total number of sampling points, which is determined by the total span of the circular arc (in degrees) and the angular step Δθ. For example, if the total span of the circular arc is 90° and Δθ = 1°, then N = 90; P i is the coordinate of the i-th sampling point, which is a two-dimensional point composed of an x coordinate and a y coordinate; the two parameters x0 and y0 are the x coordinate and y coordinate of the center of the circle respectively, which define the position of the circle or circular arc in the two-dimensional plane; r is the radius of the circle, which determines the size of the circle or circular arc; cos(θ i ) and sin(θ i ) are trigonometric functions that calculate the cosine value and sine value of the parameter angle θ i respectively.

[0172] Map the sampling points back to the original image coordinate system, and merge the dense coordinate point sets of all segments in the original sorting order to form a set of super-resolution edge coordinates after fitting, specifically including:

[0173] Coordinate system definition:

[0174] Local coordinate system: With the center of the circle (x0, y0) as the origin, the X-axis is along the direction from the center of the circle to the starting point.

[0175] Original image coordinate system: With the upper left corner of the image as the origin, the X-axis is to the right and the Y-axis is downward.

[0176] Translate the origin (x0, y0) of the local coordinate system to the corresponding position in the image coordinate system, and the translation vector is:

[0177] T = (x0, y0);

[0178] Coordinates after translation:

[0179] P′ i = P i + T = (x0 + rcos(θ i ), y0 + rsin(θ i ));

[0180] Rotation transformation (if there is a rotation between the local coordinate system and the image coordinate system):

[0181] Calculation of the rotation angle to determine the angle α between the X-axis of the local coordinate system and the X-axis of the image coordinate system. Usually, α can be calculated from the starting point P start and the center (x0, y0):

[0182] α = arctan2(y start - y0, x start - x0);

[0183] Rotation matrix:

[0184]

[0185] Coordinates after rotation:

[0186] P″ i = R(α)·(P i - T) + T;

[0187]

[0188] where x″ i and y″ i are respectively the x-coordinate and y-coordinate of the i-th sampled point in the original image coordinate system after transformation; the two parameters x0 and y0 are respectively the x-coordinate and y-coordinate of the center of the circle, defined in the original image coordinate system;

[0189] For each segmented segment, record its starting index k start and ending index k end in the original edge point set; initialize an empty list P merged ; traverse all segmented segments in the original order, and append the sampled points {P″ i} of each segmented segment to P merged in turn; since the sampled points are generated in increasing order of the parameter angle θ and are merged in the original segmented segment order, the merged point set naturally maintains the topological order of the original edge points.

[0190] In the embodiment of the present invention, the complex edge curve is segmented into multiple simple segments, and each segment can be independently fitted, reducing the difficulty and complexity of fitting. The set of coordinate points within each segmented segment retains the local features of the segment, which helps to more accurately fit the circular arc model of the segment. Three points are randomly selected for preliminary fitting, increasing the diversity and robustness of fitting, helping to avoid local optimal solutions, and calculating the distances from all points within the segment to the current circular arc model. By marking inliers through a distance threshold, it helps to remove noise points and outliers, improving the accuracy of fitting. Using all inliers to re-optimize the center and radius by the least squares method can obtain a more accurate circular arc model, better fitting the coordinate points within the segment. Sampling points uniformly at a super-resolution step size can generate more points on the circular arc, improving the resolution and fineness of the edge. The sampling points are evenly distributed on the circular arc, which helps to more accurately describe the shape and features of the circular arc. Mapping the sampling points back to the original image coordinate system makes the fitted edge coordinate set consistent with the original image, facilitating subsequent processing and analysis. Merging the dense coordinate point sets of all segments in the original sorting order forms a complete fitted super-resolution edge coordinate set, retaining the integrity and continuity of the edge.

[0191] In a preferred embodiment of the present invention, real-time state parameters of the edge-sealing device are obtained, and the state parameters are normalized to obtain a correction factor for correcting measurement deviations, including:

[0192] Obtain the state parameters of the current edge-sealing device. The state parameters include the amount of roller wear, temperature, pressure, motor speed, and the values of sensor calibration errors. Specifically, it includes: real-time collecting the state parameters of the device through various sensors (such as wear sensors, temperature sensors, pressure sensors, speed sensors, calibration error sensors) installed on the edge-sealing device; transmitting the collected data to the data processing unit (such as PLC, computer, or embedded system) by wired or wireless means. The data processing unit receives the sensor data and stores it in memory or a database for subsequent processing.

[0193] Normalize each state parameter to obtain the normalized state parameter. Specifically, it includes: for each state parameter, determine its normalization range, such as [0, 1], and use Map the state parameter to the normalization range, where x is the original state parameter value, x min and x max are the minimum and maximum values of the parameter respectively (which can be determined according to the device specifications or historical data). Apply the above formula to each state parameter to obtain the normalized state parameter x norm .

[0194] Based on the normalized state parameters, determine the central value and sensitivity of the Gaussian function corresponding to each state parameter; according to the central value and sensitivity of the Gaussian function corresponding to each state parameter, calculate the Gaussian weight of each state parameter, and normalize the Gaussian weight of each state parameter to obtain the normalized Gaussian weight of each state parameter, which specifically includes: based on the normalized state parameters, determine the central value of the Gaussian function corresponding to each parameter, and set the central value as the average value of the parameter within the normal working range. The sensitivity reflects the degree of influence of parameter changes on the correction factor, and the sensitivity value of each parameter can be determined through experiments or experience. Use the Gaussian function formula to calculate the Gaussian weight w of each state parameter i :

[0195]

[0196] where, x norm,i is the i-th normalized state parameter, μ i is the central value of the Gaussian function, and σ i is the standard deviation related to the sensitivity; normalize the Gaussian weights of all state parameters so that their sum is 1,

[0197]

[0198] where, w norm,i represents the normalized Gaussian weight, and n represents the number of Gaussian weights w i .

[0199] For each state parameter, calculate the corresponding average value and standard deviation from the historical data, which specifically includes: collect multiple sample values of each state parameter from the historical data, calculate the average value of each state parameter, and calculate the standard deviation of each state parameter.

[0200] According to the average value and standard deviation corresponding to each state parameter, calculate the coefficient of variation, and use the coefficient of variation to calculate the initial weight of each state parameter, which specifically includes: by calculating the ratio of the standard deviation to the average value to obtain the coefficient of variation of each state parameter, and use the coefficient of variation to calculate the initial weight of each state parameter. The larger the coefficient of variation, the greater the historical fluctuation of the parameter and the greater the influence on the correction factor. Therefore, a larger initial weight w init,i is given, where,

[0201]

[0202] where, C Vi represents the i-th coefficient of variation, C Vj represents the j-th coefficient of variation, and n represents the number of coefficients of variation.

[0203] Multiply the normalized Gaussian weight values by the corresponding initial weights to obtain the final weights corresponding to each state parameter; multiply each normalized state parameter by the corresponding final weight and accumulate them to obtain the correction factor.

[0204] In the embodiments of the present invention, by acquiring the state parameters of the edge banding device in real time (such as roller wear amount, temperature, pressure, motor speed, sensor calibration error, etc.), the change of the device state can be detected in time, so as to more accurately reflect the actual operating condition of the device. Normalize each state parameter to eliminate the influence of dimension and numerical range between different parameters, making each parameter comparable in subsequent calculations and improving the accuracy of the correction factor. According to the normalized state parameters, determine the center value and sensitivity of the Gaussian function corresponding to each state parameter, and then calculate the Gaussian weight. The Gaussian function can well describe the influence degree of the parameter on the correction factor, making the correction factor more in line with the operating characteristics of the actual device. Multiply the normalized Gaussian weight values by the corresponding initial weights to obtain the final weights corresponding to each state parameter. This weight adjustment mechanism can dynamically reflect the contribution degree of each parameter to the correction factor and enhance the adaptability of the correction factor. Calculate the average value and standard deviation of each state parameter from historical data, and use the coefficient of variation to calculate the initial weight of each state parameter, considering the historical fluctuation of the parameter, making the correction factor more robust. The larger the coefficient of variation, the greater the historical fluctuation of the parameter and the greater its influence on the correction factor, so a larger initial weight is given to it. The correction factor comprehensively considers the real-time state and historical data of the device, can more accurately correct the measurement deviation. By applying the correction factor, the operating parameters of the edge banding device can be controlled more effectively, reducing the influence of the measurement deviation on the edge banding quality, thereby improving the quality and consistency of the edge banding products. By real-time monitoring and analyzing the device state parameters, it is possible to predict possible failures or performance degradation of the device, perform maintenance in advance, and reduce downtime and maintenance costs.

[0205] In a preferred embodiment of the present invention, the corrected edge coordinate set is obtained by correcting the fitted super-resolution edge coordinate set with the correction factor, including:

[0206] Traverse each point in the fitted coordinate set, multiply the horizontal coordinate x and vertical coordinate y of each point by the correction factor respectively to calculate the corrected coordinates, specifically including:

[0207] The fitted super-resolution edge coordinate set is assumed to be where N is the total number of coordinate points, (x i , y i ) are the horizontal coordinate and vertical coordinate of the i-th point.

[0208] The correction factor is assumed to be CF and is used to correct the coordinates. Use a loop structure to iterate through each point in the fitted coordinate set. For each point (x i , y i ), perform the following calculations:

[0209] Corrected horizontal coordinate: x' i = x i × CF;

[0210] Corrected vertical coordinate: y' i = y i × CF;

[0211] where x' i and y' i are the corrected horizontal and vertical coordinates respectively. Store the corrected coordinates (x' i , y' i ) into a new coordinate set to form the corrected edge coordinate set where i represents the index variable used to iterate through each point in the corrected edge coordinate set, with a value range from 1 to N. During the loop or iteration process, i is used to identify the currently processed coordinate point. N represents the total number of coordinate points in the corrected edge coordinate set.

[0212] In the embodiments of the present invention, the fitted super-resolution edge coordinate set may have deviations due to factors such as measurement errors, inaccurate models, or data noise. By multiplying by the correction factor, these errors can be corrected, thereby improving the accuracy of the coordinates. The correction factor can further refine the coordinate values, making the edge coordinate set more accurate at the super-resolution level, which helps to improve the quality of subsequent image processing or analysis.

[0213] The correction factor can be dynamically adjusted according to different device states, measurement conditions, or application scenarios. This means that the correction process has stronger adaptability and can handle various complex situations. By introducing the correction factor, the model can better generalize to unseen data, improving the robustness and stability of the model. Iterating through the fitted coordinate set and directly multiplying by the correction factor is a simple and effective correction method. It does not require a complex iterative optimization process and can quickly obtain the corrected edge coordinate set. This method is easy to implement in the existing image processing or analysis process and does not require significant modification of the original algorithm.

[0214] The corrected edge coordinate set can more accurately locate the edge positions in edge detection tasks, improving the accuracy and reliability of edge detection. In image segmentation tasks, the corrected edge coordinate set helps to more precisely divide the image regions, enhancing the segmentation effect. For application scenarios requiring 3D reconstruction, the corrected edge coordinate set can provide more accurate 3D information, contributing to improving the reconstruction quality. As the basic data for subsequent image processing or analysis, the accuracy of the corrected edge coordinate set directly affects the difficulty and effect of subsequent processing. By correcting with a correction factor, the difficulty of subsequent processing can be reduced and the processing efficiency can be improved.

[0215] In a preferred embodiment of the present invention, based on the corrected edge coordinate set, the actual geometric dimensions of the sheet material and the deviation of the actual geometric dimensions are calculated, and a dimensional deviation score is obtained, including:

[0216] Calculate the curvature between adjacent points for the corrected coordinate set, identify the curvature mutation points, and divide the coordinate set into straight line segments and circular arc segments according to the curvature threshold. Specifically, for the corrected edge coordinate set Calculate the curvature at each point (x′ i , y′ i ). The curvature can be approximated by calculating the reciprocal of the radius of the circle formed by three adjacent points. The formula is as follows:

[0217]

[0218] where κ i is the curvature at point i. Set a curvature threshold κ th . When the curvature change between two adjacent points exceeds κ th , this point is considered a curvature mutation point; according to the curvature mutation points, divide the coordinate set into straight line segments and circular arc segments; the straight line segment is a continuous point set with a smaller curvature; the circular arc segment is a continuous point set with a larger curvature; (x′ i+1 - x′ i ) is the difference between the horizontal coordinate of the (i + 1)-th point and the horizontal coordinate of the i-th point in the corrected coordinate set, representing the change in the horizontal distance between two adjacent points; (y′ i+2 - y′ i+1 ) is the difference between the vertical coordinate of the (i + 2)-th point and the vertical coordinate of the (i + 1)-th point in the corrected coordinate set, representing the change in the vertical distance between the subsequent two adjacent points; (y′ i+1 - y′ i ) is the difference between the vertical coordinate of the (i + 1)-th point and the vertical coordinate of the i-th point in the corrected coordinate set, representing the change in the vertical distance between the current two adjacent points; (x′ u+2 - x′ u+1 ) is the difference between the horizontal coordinate of the (i + 2)-th point and the horizontal coordinate of the (i + 1)-th point in the corrected coordinate set.

[0219] For straight line segments, sum up the lengths of all straight line segments to obtain the total length and width of the sheet; for arc segments, take the average value of the fillet radii as the overall fillet feature, specifically including:

[0220] For each straight line segment, calculate its length. Assume the straight line segment is composed of points (x′ i , y′ i ) to (x′ j , y′ j ), then the length L is:

[0221]

[0222] Sum up the lengths of all straight line segments to obtain the total length of the sheet (assuming the sheet is rectangular, the total length is the sum of the lengths of all horizontal straight line segments, and the total width is the sum of the lengths of all vertical straight line segments); for each arc segment, calculate its fillet radius. Assume the arc segment is determined by three points (x′ i , y′ i ), (x′ i+1 , y′ i+1 ), (x′ i+2 , y′ i+2 ), then the fillet radius R can be calculated by the method of fitting a circle. Take the average value of the fillet radii of all arc segments as the overall fillet feature.

[0223] Calculate the length deviation and width deviation based on the total length and width of the sheet; calculate the fillet radius deviation based on the overall fillet feature, specifically including: Assume the theoretical length of the sheet is L theory , and the theoretical width is W theory , then the length deviation ΔL and width deviation ΔW are respectively:

[0224] ΔL = |L actual - L theory |;

[0225] ΔW = |W actual - W theory |;

[0226] where L actual and W actual are respectively the actual total length and total width of the sheet. Assume the theoretical fillet radius of the sheet is R theory , then the fillet radius deviation ΔR is:

[0227] ΔR = |R average - R theory |;

[0228] where R average is the average value of the fillet radii of all arc segments.

[0229] Calculate the corresponding relative length deviation, relative width deviation, and relative fillet radius deviation according to the length deviation, width deviation, and fillet radius deviation respectively, specifically including: The relative length deviation Re is:

[0230] The relative width deviation Rl is: The relative fillet radius deviation Ra is:

[0231] Perform normalization processing on the relative length deviation, relative width deviation, and relative fillet radius deviation, and perform weighted summation on the normalized deviations of each item to generate a dimensional deviation score.

[0232] In the embodiment of the present invention, by calculating the curvature between adjacent points and identifying the curvature mutation points, the coordinate set can be accurately divided into straight line segments and arc segments. This segmentation processing method helps to calculate the straight line length and arc radius of the plate more precisely, thereby improving the accuracy of dimensional measurement; calculating the deviation according to the total length, width, and fillet radius of the plate can intuitively reflect the difference between the plate size and the theoretical value; accumulating the lengths of all straight line segments to obtain the total length and width of the plate can comprehensively reflect the linear dimensional characteristics of the plate, and taking the average value of the fillet radii of the arc segments as the overall fillet feature can simplify the processing of fillet dimensions while retaining the main information of the fillet dimensions. Calculating the relative length deviation, relative width deviation, and relative fillet radius deviation can normalize the deviation values to a relative scale, facilitating the comparison and evaluation of deviations between plates of different sizes. Performing weighted summation on the normalized deviations of each item to generate a dimensional deviation score can comprehensively reflect the overall deviation of the plate size, providing a quantitative index for the quality grading and screening of the plate; through the dimensional deviation score, problems existing in the production process, such as equipment wear and improper process parameters, can be discovered in a timely manner, so as to adjust the production process in a timely manner, improve production efficiency. The dimensional deviation score can be used as an important quality control index to monitor the production quality of the plate and ensure that the product meets the design requirements. At the same time, by comparing the dimensional deviation scores of different batches or different production lines, the stability and consistency of the production line can be evaluated. This process can be integrated into an automated detection system to realize automatic measurement and deviation evaluation of the plate size, improve the detection efficiency and accuracy. Based on the dimensional deviation score, an intelligent decision-making system can be developed to automatically judge the quality grade of the plate and give corresponding treatment suggestions, such as rework, scrapping, or downgrading for use, etc.

[0233] In a preferred embodiment of the present invention, according to the dimensional deviation score, mark the plates that do not meet the requirements and control the equipment to perform corresponding processing, including:

[0234] Set a scoring threshold. If the score of the board is greater than the scoring threshold, it is determined as unqualified, and marks, deviation types, and score values are added to the surface of the board; the marked board is conveyed to the rework station, and the preset cutting parameters, grinding parameters, or shape correction parameters are called, specifically including:

[0235] According to the quality requirements of the board and industry standards, set a threshold for the dimensional deviation score. For example, set the scoring threshold to 0.1 (assuming the scoring range is between 0 and 1, and the larger the value, the greater the deviation); input the scoring threshold into the automated inspection and scoring system as the basis for determining whether the board is qualified; perform dimensional deviation scoring on each board. The scoring process includes calculating length deviation, width deviation, and fillet radius deviation, and performing normalization and weighted summation (as described above); compare the calculated dimensional deviation score with the set scoring threshold. If the score is greater than the scoring threshold, it is determined that the board is unqualified. For unqualified boards, generate marking information including deviation types (such as length deviation, width deviation, fillet radius deviation) and score values, and use an inkjet printer, laser marking machine, or other marking equipment to add the marking information to the surface of the board; use sensors or vision recognition systems to identify the boards with marks on the surface, and convey the marked boards to the designated rework station through conveyors, robotic arms, or other automated equipment; at the rework station, use scanning equipment or vision systems to read the deviation type marks on the surface of the board, and according to the deviation type, call the preset cutting parameters, grinding parameters, or shape correction parameters. For example, if the deviation type is length deviation, call the preset cutting parameters to adjust the length; if the deviation type is fillet radius deviation, call the preset shape correction parameters to correct the fillet, and use cutting equipment, grinding equipment, or shape correction equipment to perform rework on the board according to the called parameters.

[0236] When specifically applied, it specifically includes:

[0237] After a certain board undergoes dimensional deviation scoring, the score is 0.18 (greater than the set scoring threshold of 0.1), and it is determined as unqualified. The deviation types are length deviation and fillet radius deviation. Specifically, the actual length of the board is 8 mm longer than the theoretical length, and the fillet radius is 2 mm smaller than the theoretical radius. The parameter adjustment process:

[0238] Add a mark on the surface of the board: "Length deviation 8 mm, fillet radius deviation -2 mm, score 0.18"; the board is conveyed by the conveyor to a comprehensive rework station including cutting and shape correction equipment, and the rework station scans the marks on the surface of the board to identify that the length deviation is 8 mm.

[0239] Call the preset cutting parameters: the cutting length is 8 mm, and the cutting position is one end of the plate (determined according to the deviation direction). Use the cutting equipment to cut the plate according to the called parameters, remove the extra 8 mm length, and the rework station continues to identify that the fillet radius deviation is -2 mm.

[0240] Call the preset shape correction parameters: increase the shape correction radius by 2 mm, and the shape correction position is at all fillets of the plate.

[0241] Perform the shape correction operation:

[0242] Use the shape correction equipment to perform shape correction on the fillets of the plate according to the called parameters to make the fillet radius reach the theoretical value. After the rework is completed, the plate is inspected again to confirm that the length deviation and fillet radius deviation have been eliminated. After passing the inspection, the plate is released to the next process.

[0243] In the embodiment of the present invention, by setting a scoring threshold, it is possible to accurately identify the plates with size deviations exceeding the allowable range, prevent unqualified products from flowing into the next process or the final product, add marks, deviation types and scoring values on the surface of the plates, which is convenient for subsequent processing personnel to quickly understand the problems of the plates, so as to take targeted rework measures. Automatically transfer the marked plates to the rework station, and call the preset cutting parameters, grinding parameters or shape correction parameters to realize the automation of the rework process, improve the processing efficiency. The automated processing reduces the workload of manual identification and rework, reduces the risk of human errors, and at the same time improves the overall operation efficiency of the production line. By promptly reworking unqualified plates, the scrap rate is reduced, and the waste of raw materials and energy is reduced. The automated rework processing enables the equipment to operate more efficiently, reducing the idle time of the equipment caused by waiting for manual processing; the marks and scoring values on the surface of the plates provide detailed quality data, which is convenient for subsequent quality traceability and analysis, helps to discover potential problems in the production process. By analyzing and processing unqualified plates, the production process and quality control system can be continuously optimized, the overall quality level of the product can be improved, and the stability and consistency of the product quality are ensured through strict size deviation scoring and rework processing, improving customer satisfaction with the product.

[0244] The embodiment of the present invention also provides a fully automatic intelligent control system for a plate edge banding device, including:

[0245] A determination module, configured to construct an octree structure and map pixel values according to the resolution of the original plate image and the preset octree depth, perform multi-layer decomposition on the nodes by a filter to obtain a set of sub-bands, perform morphological processing on the high-frequency sub-bands to obtain a binary edge map, connect the edge points into curves by the Hough transform and interpolate to extract the super-resolution edge coordinates;

[0246] A calculation module, which is used to perform median filtering on the super-resolution edge coordinates, sort them in clockwise order to form an ordered set, calculate curvature markers to divide the edge segments, randomly select points for each segment to fit a circle and optimize to obtain the final arc model, and map the sampled points along the arc model back to the original image coordinate system to merge and form a set of super-resolution edge coordinates after fitting;

[0247] A normalization module, which is used to obtain the real-time state parameters of the edge-sealing device and perform normalization processing on the state parameters to obtain a correction factor for correcting measurement deviations;

[0248] A correction module, which is used to correct the set of super-resolution edge coordinates after fitting through the correction factor to obtain a set of corrected edge coordinates;

[0249] An evaluation module, which is used to calculate the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions according to the set of corrected edge coordinates, and obtain a dimension deviation score;

[0250] A control module, which is used to mark the sheets that do not meet the requirements according to the dimension deviation score and control the device to perform corresponding processing.

[0251] An embodiment of the present invention also provides a computing device, including: a processor and a memory storing a computer program. When the computer program is run by the processor, it executes the method described above. All implementation manners in the above method embodiments are applicable to this embodiment and can also achieve the same technical effects.

[0252] An embodiment of the present invention also provides a computer-readable storage medium storing instructions. When the instructions are run on a computer, the computer is made to execute the method described above. All implementation manners in the above method embodiments are applicable to this embodiment and can also achieve the same technical effects.

[0253] The above is the preferred embodiment of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent control method for a fully automatic sheet edge banding device, characterized in that, The method includes: Construct an octree structure according to the original sheet image resolution and the preset octree depth, and map the pixel values. Decompose the nodes through a filter in multiple layers to obtain a set of subbands. Perform morphological processing on the high-frequency subbands to obtain a binary edge map. After determining the edge points, use the Hough transform to connect them into curves and interpolate to extract the super-resolution edge coordinates; Perform median filtering on the super-resolution edge coordinates and sort them in clockwise order to form an ordered set. Calculate the curvature to mark the segmentation points and divide the edge segments. Use each segment as a unit to randomly select points to fit circles and optimize to obtain the final circular arc model. Map the sampled points along the circular arc model back to the original image coordinate system and merge them to form the fitted super-resolution edge coordinate set; Obtain the real-time state parameters of the edge banding device, and perform normalization processing on the state parameters to obtain a correction factor for correcting the measurement deviation; Correct the fitted super-resolution edge coordinate set through the correction factor to obtain the corrected edge coordinate set; According to the corrected edge coordinate set, calculate the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions, and obtain the dimension deviation score; According to the dimension deviation score, mark the sheets that do not meet the requirements, and control the device to perform corresponding processing.

2. The intelligent control method of a fully automatic sheet edge banding device according to claim 1, wherein, Construct an octree structure according to the original sheet image resolution and the preset octree depth, and map the pixel values. Decompose the nodes through a filter in multiple layers to obtain a set of subbands, including: Construct an octree structure according to the resolution of the original sheet image and the preset octree depth. Each node in the octree structure represents a pixel block in the original sheet image; Map the pixel values of the original sheet image into the nodes of the octree structure block by block; For each node of the octree, perform convolution operations through the low-pass filter and high-pass filter of the Haar wavelet respectively. After each layer of decomposition, a low-frequency subband and three high-frequency subbands are generated; Use the low-frequency subband obtained from the current layer of decomposition as the input for the next layer of decomposition; redistribute the data of the low-frequency subband into the nodes of the octree, adjust the node size to match the resolution after the current decomposition, repeat the filtering operation on the remapped low-frequency subband to generate new low-frequency and high-frequency subbands. If the current layer number has not reached the preset layer number, increment the layer number and repeat the filtering operation; if it has reached, terminate the decomposition to obtain the set of subbands after multiple layers of decomposition. The set of subbands includes the low-frequency subband of the final layer and the high-frequency subbands of each layer.

3. The intelligent control method of a fully automatic plate edge banding device according to claim 2, characterized in that, Perform morphological processing on the high-frequency subbands to obtain a binary edge map. After determining the edge points, use the Hough transform to connect them into curves and interpolate to extract the super-resolution edge coordinates, including: Perform morphological processing on the high-frequency subbands. In the high-frequency subbands after morphological processing, determine the edge points through threshold segmentation; Input the edge points into the Hough transform, identify the collinear points and fit them into continuous curves, and merge adjacent curve segments to form a continuous edge curve; On the continuous edge curve, obtain the coordinates of the interpolated edge points through linear interpolation; Extract the coordinates of the interpolated edge points to form a super-resolution edge coordinate set.

4. An intelligent control method for a fully automatic sheet edge banding device according to claim 3, characterized in that, Perform median filtering on the super-resolution edge coordinates and sort them in clockwise order to form an ordered set. Calculate the curvature to mark the segmentation points and divide the edge segments, including: Perform median filtering on the super-resolution edge coordinates, and sort the filtered coordinate points in the clockwise direction of the edge curve to form an ordered edge coordinate set; Calculate the local curvature of the ordered coordinate points by cubic spline interpolation method, traverse all points, and mark the points as segmentation points when the curvature exceeds the threshold; Divide the edge into multiple continuous segments according to the segmentation points.

5. An intelligent control method for a full-automatic sheet edge banding device according to claim 4, characterized in that, Randomly select points in each segment to fit a circle and optimize to obtain the final circular arc model. Map the sampled points along the circular arc model back to the original image coordinate system and merge them to form the fitted super-resolution edge coordinate set, including: Extract the set of coordinate points within each segmentation segment as an independent fitting unit; According to the set of coordinate points of each segmentation segment, randomly select 3 points among the coordinate points of the current segment, calculate the center and radius of the fitted circle, and calculate the distance from all points in the segment to the current circular arc model; Mark the points with a distance less than the threshold as inliers, and re-optimize the center and radius using all inliers by the least squares method to obtain the final circular arc model for each segmentation segment; For the final circular arc model of each segmentation segment, uniformly sample points along the circular arc path at the super-resolution step size; Map the sampled points back to the original image coordinate system, and merge the dense coordinate point sets of all segments in the original sorting order to form the fitted super-resolution edge coordinate set.

6. The intelligent control method of a fully automatic sheet edge banding device according to claim 5, characterized in that, Obtain the real-time state parameters of the edge-sealing device, and perform normalization processing on the state parameters to obtain a correction factor for correcting measurement deviation, including: Obtain the state parameters of the current edge-sealing device. The state parameters include the numerical values of roller wear, temperature, pressure, motor speed, and sensor calibration error; Perform normalization processing on each state parameter to obtain the normalized state parameter; According to the normalized state parameters, determine the central value and sensitivity of the Gaussian function corresponding to each state parameter; according to the central value and sensitivity of the Gaussian function corresponding to each state parameter, calculate the Gaussian weight of each state parameter, and perform normalization processing on the Gaussian weight of each state parameter to obtain the normalized Gaussian weight of each state parameter; For each state parameter, calculate the corresponding average value and standard deviation from the historical data; Calculate the coefficient of variation according to the average value and standard deviation corresponding to each state parameter, and use the coefficient of variation to calculate the initial weight of each state parameter; Multiply the normalized Gaussian weight value by the corresponding initial weight to obtain the final weight corresponding to each state parameter; Multiply each normalized state parameter by the corresponding final weight and accumulate to obtain the correction factor.

7. An intelligent control method for a fully automatic sheet edge banding device according to claim 6, characterized in that, Correct the fitted super-resolution edge coordinate set by the correction factor to obtain the corrected edge coordinate set, including: Traverse each point in the fitted coordinate set, multiply the horizontal coordinate x and vertical coordinate y of each point by the correction factor to calculate the corrected coordinates.

8. An intelligent control method for a fully automatic sheet edge banding device according to claim 7, characterized in that, According to the corrected edge coordinate set, calculate the actual geometric dimensions of the sheet material and the deviation of the actual geometric dimensions, and obtain the dimension deviation score, including: Calculate the curvature between adjacent points for the corrected coordinate set, identify the curvature mutation points, and divide the coordinate set into straight line segments and circular arc segments according to the curvature threshold; For straight line segments, accumulate the lengths of all straight line segments to obtain the total length and width of the sheet; for arc segments, take the average value of the fillet radii as the overall fillet feature; Calculate the length deviation and width deviation based on the total length and width of the sheet; calculate the fillet radius deviation based on the overall fillet feature; Calculate the corresponding length relative deviation, width relative deviation, and fillet radius relative deviation according to the length deviation, width deviation, and fillet radius deviation respectively; Normalize the length relative deviation, width relative deviation, and fillet radius relative deviation, and perform weighted summation on the normalized deviations of each item to generate a dimensional deviation score.

9. The intelligent control method of a fully automatic sheet edge banding device according to claim 8, wherein, According to the dimensional deviation score, mark the sheets that do not meet the requirements and control the equipment to perform corresponding processing, including: Set a scoring threshold. If the sheet score > the scoring threshold, it is determined as unqualified, and a mark, deviation type, and score value are added to the surface of the sheet; Transfer the marked sheets to the rework station and call the preset cutting parameters, grinding parameters, or shape correction parameters.

10. An intelligent control system for a fully automatic plate edge banding equipment, characterized in that, The system executes the method described in any one of claims 1 to 9, including: A determination module for constructing an octree structure and mapping pixel values according to the original sheet image resolution and the preset octree depth, performing multi-layer decomposition of the nodes by a filter to obtain a set of subbands, performing morphological processing on the high-frequency subbands to obtain a binary edge map, connecting the edge points into curves by the Hough transform, and interpolating to extract the super-resolution edge coordinates; A calculation module for performing median filtering on the super-resolution edge coordinates, sorting them in clockwise order to form an ordered set, calculating the curvature to mark the segmentation points to divide the edge segments, randomly selecting points for each segment to fit a circle and optimizing to obtain the final arc model, and mapping the sampled points along the arc model back to the original image coordinate system and merging them to form a set of fitted super-resolution edge coordinates; A normalization module for obtaining the real-time state parameters of the edge banding equipment and normalizing the state parameters to obtain a correction factor for correcting the measurement deviation; A correction module for correcting the set of fitted super-resolution edge coordinates by the correction factor to obtain a corrected set of edge coordinates; An evaluation module for calculating the actual geometric dimensions of the sheet and the deviation of the actual geometric dimensions according to the corrected set of edge coordinates, and obtaining a dimensional deviation score; A control module for marking the sheets that do not meet the requirements according to the dimensional deviation score and controlling the equipment to perform corresponding processing.

Images