A label defect identification, cutting and collection method and system

Abstract

The application discloses a label defect identification, cutting and collecting method and system, places cloth into a cloth feeding roller of a cutting machine and fixes the cloth, sets an outer contour of a label to be cut, keeps the cloth open with a predetermined tension, controls a laser head to move to a starting position of a target label, takes an image of the target label, extracts a contour line of the target label image, carries out label identification and identification on the target label, controls the laser head to cut the cloth, cuts qualified labels and keeps waste labels, and arranges and collects qualified cloth corresponding to the qualified labels.

Description

Technical Field

[0001] This invention relates to the field of label identification technology, and in particular to a method and system for identifying, cutting, and collecting label defects. Background Technology

[0002] All the small tags in the entire piece of fabric are cut out, mainly using a laser cutting machine. However, most publicly available laser cutting machines can only cut the fabric, and then the cut fabric needs to be manually identified, collected, and replaced with the next piece of fabric.

[0003] In addition, the latest fabric laser cutting machine technology uses conveyor belts or traction devices to change fabric, and can also collect the cut fabric by vibration or air blowing.

[0004] While these technologies enable semi-automated cutting of small labels, manual arrangement and stacking of fabric is still required, along with visual inspection and removal of defective products. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for identifying, cutting, and collecting label defects, thereby enabling automated identification and cutting of fabric labels and their classification.

[0006] To address the aforementioned technical problems, embodiments of the present invention provide a method for identifying, cutting, and collecting label defects, comprising:

[0007] S1, Place the fabric into the feed roll of the cutting machine and fix the fabric in place;

[0008] S2, set the outer outline of the label to be cut;

[0009] S3, the fabric is spread out and the fabric is kept under predetermined tension;

[0010] S4, control the laser head to perform an alignment operation. By controlling the laser head to move omnidirectionally on the fabric and combining visual detection, the laser head is moved to the starting position of the target label.

[0011] S5, take an image of the target label to obtain a target label image, and then perform label recognition and identification on the target label by extracting the outline of the target label image;

[0012] S6, repeat S3-S5 to obtain smooth trajectory lines and recognition results of one or more columns of labels, merge the trajectory lines of qualified labels, control the laser head to cut the fabric, cut off the qualified labels, and keep the defective labels.

[0013] S7, neatly arrange and collect the qualified fabrics corresponding to the qualified labels.

[0014] Wherein, S3 includes:

[0015] The fabric is clamped at both ends by clamping devices.

[0016] The distance between the two rollers is controlled by a push rod motor to control the distance between the two clamping devices, thereby controlling the tension of the fabric.

[0017] S3 further includes:

[0018] S31, Obtain the current tension image of the long roll of fabric;

[0019] S32, perform enhancement processing on the tensioned image, the enhancement processing includes denoising, binarization and morphological processing;

[0020] S33, use the Canny operator to perform edge detection and extract the outer contour of the label under the current camera;

[0021] S34, transform the contour points of the outer contour into Cartesian contour points by using the Cartesian coordinate system;

[0022] S35, convert the Cartesian contour points into Fourier descriptors using Fourier series;

[0023] S36, extract features from the Fourier descriptor, the features including average radius, concavity and convexity, and symmetry;

[0024] S37, compare the features with the standard contour descriptor, evaluate the similarity and difference, and obtain the hash value of the difference by calculating the Euclidean distance;

[0025] S38. Repeat steps S31-S37 to identify multiple labels, calculate the average value using the least squares method, and determine and output whether the fabric is in a relaxed or taut state.

[0026] Wherein, S4 includes;

[0027] The X-axis and Y-axis motors of the laser head are controlled to move the laser head omnidirectionally on the feeding material. Combined with visual detection, a PID control algorithm is used to move the laser head to the starting position of the target label.

[0028] Wherein, S5 includes:

[0029] The target label image is captured and obtained, and its contour is extracted using the Canny operator;

[0030] After converting the contour lines of the contour into Cartesian coordinate values, a continuous trajectory line is formed, and the cusps in the trajectory line are removed.

[0031] The trajectory line is interpolated using spline curves to make it a smooth curve;

[0032] The labels within the outline are visually inspected and identified. Label anomalies are identified through deep learning and image detection. These anomalies include thread breakage, stains, yarn breakage, and misalignment.

[0033] The target label is a small label, and the detection of the small label further includes:

[0034] Enhance the target label image, perform white balance on the colors, and increase contrast;

[0035] The target label image is processed to obtain a Fourier descriptor for the target label image;

[0036] Pooling is performed on the Fourier descriptor of the target label image to preserve image features and reduce image size;

[0037] After normalizing and generating the co-occurrence probability matrix using the gray-level co-occurrence matrix, the uniformity and texture coarseness of the target label image in the corresponding channel and direction are calculated.

[0038] The target label image is fed into a pre-trained deep learning neural network for inference, and the trained defect class is obtained through target detection.

[0039] Merge the detection, repeat the above steps, collect and count the labels of multiple rows, perform statistical calculations based on the Fourier descriptor and the co-occurrence probability matrix, obtain the group with the largest difference and compare it with the standard label. If it is less than the threshold, all columns are judged to be qualified labels. Otherwise, mark the last one as defective and continue to find the label with the second largest difference, repeat the above qualified determination process.

[0040] To detect the cutting position of the fabric, in one embodiment, a gray-level co-occurrence matrix is ​​used to calculate the uniformity and texture coarseness of the target label image in the corresponding channel and direction, including:

[0041] The target label image is converted into a single-channel image;

[0042] The co-occurrence matrix is ​​calculated by selecting a direction perpendicular to the woven label direction and the distance between pixels, based on the woven label direction of the fabric.

[0043] Pixel values ​​are divided into 25 levels, and the co-occurrence frequency of each pixel with orientation and adjacent pixels at the single-channel level is counted.

[0044] Construct a 25x25 co-occurrence matrix, where each element CLCM(i,j) represents the number of times pixel levels i and j occur in pairs, given the chosen direction and distance;

[0045] The co-occurrence matrix is ​​normalized to obtain the co-occurrence probability matrix, where each element CLCM(i,j) represents the probability that i and j appear in pairs.

[0046] The image energy is calculated using the co-occurrence probability matrix and the following formula, reflecting the uniformity and texture coarseness of the target label image in the specified channel and direction.

[0047]

[0048] Repeat the above steps to calculate the red, green, and blue channels, and obtain three sets of cached data.

[0049] This application does not limit the process of listing the fabrics. In one embodiment, step S7 includes:

[0050] The fabric is moved forward to the arranging device, which is controlled by an X-axis motor, a Y-axis motor and a Z-axis motor. The head of the arranging device is equipped with an air pump that matches the number of labels in each row.

[0051] Based on the location information of the fabric and the location of the waste, control the air pump to move to the target point;

[0052] Move downwards and activate the air pump valves at all qualified label locations to attract and move the qualified labels to the listing area where the packaging boxes or baskets to be stored are placed.

[0053] Turn on the air pump, and after the label falls, move it downwards to press the label firmly.

[0054] By using a sorting device and an air pump, the fabric and labels are adsorbed and released, and placed in designated positions, thus achieving automated cutting and automatic label placement.

[0055] This application does not limit the control of the air pump. In one embodiment, controlling the air pump to move to the target point means controlling the air pump to adjust its position on the sliding guide rail, including:

[0056] Add the same number of air pumps to the sliding guide rail according to the number of labels in each row;

[0057] The spacing between adjacent air pumps is set according to the interval between the labels;

[0058] Once the cut label is moved to the listing area, the air pump moves down, sucks up the cut label, and then lifts it up.

[0059] The Y-axis motor of the air pump drives the label on the adsorption to move to the top of the box;

[0060] The air pump is lowered to press the adsorbed label into the box;

[0061] Return to the initial position to complete the collection and listing of a whole row of tags.

[0062] In addition, embodiments of this application also provide a label defect identification, cutting, and collection system, including:

[0063] A fabric fixing device includes a feeding roll and a clamping component, wherein the feeding roll is used to store fabric and the clamping component clamps the fabric;

[0064] A tension adjustment device is used to adjust the spacing between the clamping components that clamp the fabric, thereby adjusting the tension of the fabric.

[0065] The alignment device is used to control the X-axis motor and Y-axis motor of the laser head, so that the laser head moves in all directions on the feed material. Combined with visual detection, the PID control algorithm is used to move the laser head to the starting position of the target label.

[0066] The identification device is used to capture an image of the target label, obtain a target label image, and perform label recognition and identification on the target label by extracting the outline of the target label image.

[0067] A cutting device is used to cut the fabric corresponding to the qualified label identified by the identification device through the laser head, cut off the qualified label, and keep the defective label;

[0068] A sorting device is used to neatly sort and collect qualified fabrics corresponding to qualified labels.

[0069] The label defect identification, cutting, and collection method and system provided in this invention have the following advantages compared with the prior art:

[0070] The label defect identification, cutting, and collection method and system provided in this invention involves placing fabric into the feed roll of a cutting machine and fixing the fabric. The outer contour of the label to be cut is set, the fabric is stretched and maintained at a predetermined tension, and a laser head is controlled to move to the starting position of the target label to capture an image of the target label. After extracting the outline of the target label image, the target label is identified and authenticated. The laser head is then controlled to cut the fabric, cutting off qualified labels and retaining defective labels. The qualified fabric corresponding to the qualified labels is then neatly arranged and collected. By combining machine vision and artificial intelligence to locate and identify the cut labels, and controlling the two-dimensional laser to cut only qualified labels, defective labels can be quickly avoided. Qualified fabric is then collected and neatly arranged, achieving a fully automated cutting process from fabric replacement, identification, cutting to collection and arrangement. This reduces the cutting and collection of defective labels and the subsequent manual identification of defects, greatly improving production efficiency. Attached Figure Description

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0072] Figure 1 A schematic diagram of the step flow structure in one embodiment of the label defect identification, cutting and collection method provided by the present invention;

[0073] Figure 2 A schematic diagram of the fabric tension detection steps in one embodiment of the label defect identification, cutting and collection method provided by the present invention;

[0074] Figure 3 A schematic diagram of a structure in one embodiment of the label defect identification, cutting and collection system provided by the present invention;

[0075] Figure 4 A first-view structural schematic diagram of a physical object in one embodiment of the label defect identification, cutting and collection system provided by the present invention;

[0076] Figure 5 A second-view structural schematic diagram of an embodiment of the label defect identification, cutting and collection system provided by the present invention;

[0077] Among them, the feeding drum-1, the pulling devices-2 and 10, the laser head-13, the laser head X-axis motor-4, the laser head Y-axis motor-5, the camera-3, the air pump X-axis motor-6, the air pump Y-axis motor-7, the air pump suction head-8, the guide wheel-9, the receiving drum-11, the clamping device-12, the clamping device at the feeding end-14, the stacking area at the end-15, the packaging box to be placed-16, the sliding guide rail-17, and the display screen controller-18. Detailed Implementation

[0078] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0079] Please refer to Figure 1-5 , Figure 1 A schematic diagram of the step flow structure in one embodiment of the label defect identification, cutting and collection method provided by the present invention; Figure 2 A schematic diagram of the fabric tension detection steps in one embodiment of the label defect identification, cutting and collection method provided by the present invention; Figure 3 A schematic diagram of a structure in one embodiment of the label defect identification, cutting and collection system provided by the present invention; Figure 4 A first-view structural diagram of a physical object in one embodiment of the label defect identification, cutting, and collection system provided by the present invention. Figure 5 This is a second-view structural schematic diagram of an embodiment of the label defect identification, cutting, and collection system provided by the present invention.

[0080] In one specific embodiment, the label defect identification, cutting, and collection method includes:

[0081] S1, Place the fabric into the feed roll of the cutting machine and fix the fabric in place;

[0082] S2, set the outer outline of the label to be cut;

[0083] S3, the fabric is spread out and the fabric is kept under predetermined tension;

[0084] S4, control the laser head to perform an alignment operation. By controlling the laser head to move omnidirectionally on the fabric and combining visual detection, the laser head is moved to the starting position of the target label.

[0085] S5, take an image of the target label to obtain a target label image, and then perform label recognition and identification on the target label by extracting the outline of the target label image;

[0086] S6, repeat S3-S5 to obtain smooth trajectory lines and recognition results of one or more columns of labels, merge the trajectory lines of qualified labels, control the laser head to cut the fabric, cut off the qualified labels, and keep the defective labels.

[0087] S7, neatly arrange and collect the qualified fabrics corresponding to the qualified labels.

[0088] The fabric is placed into the feed roll of the cutting machine and fixed in place. The outer contour of the label to be cut is set, the fabric is stretched and kept under predetermined tension, and the laser head is controlled to move to the starting position of the target label to capture an image of the target label. After extracting the outline of the target label image, the target label is identified and authenticated. The laser head is then controlled to cut the fabric, cutting off qualified labels and retaining defective labels. The qualified fabric corresponding to the qualified labels is neatly arranged and collected. By combining machine vision and artificial intelligence to locate and identify the cut labels, the two-dimensional laser is controlled to cut only qualified labels, quickly avoiding defective labels. Qualified fabric is collected and neatly arranged, realizing a fully automated cutting process from fabric replacement, identification, cutting to collection and arrangement. This reduces the cutting and collection of defective labels and the subsequent manual identification of defects, greatly improving production efficiency.

[0089] This application does not limit the process of controlling the tension of the fabric. In one embodiment, S3 includes:

[0090] The fabric is clamped at both ends by clamping devices.

[0091] The distance between the two rollers is controlled by a push rod motor to control the distance between the two clamping devices, thereby controlling the tension of the fabric.

[0092] In one embodiment, the automatic material changing process consists of clamping devices at both ends and a visual inspection system to detect the tightness of the fabric. The two clamping devices control the distance between the two rollers via a push-rod motor to control the fabric tension. When the tension is too high, the clamping device at the take-up end closes, and the clamping device at the release end moves backward, easing the fabric. When the tension is too low, the clamping devices at both ends close, and the fabric is taken in by controlling a single-sided roller to maintain appropriate tension.

[0093] The determination of tension is not limited. To avoid errors caused by manual judgment alone, and to improve the accuracy and standardization of tension determination and reduce errors, in one embodiment, S3 further includes:

[0094] S31, Obtain the current tension image of the long roll of fabric;

[0095] S32, perform enhancement processing on the tensioned image, the enhancement processing includes denoising, binarization and morphological processing;

[0096] S33, use the Canny operator to perform edge detection and extract the outer contour of the label under the current camera;

[0097] S34, transform the contour points of the outer contour into Cartesian contour points by using the Cartesian coordinate system;

[0098] S35, convert the Cartesian contour points into Fourier descriptors using Fourier series;

[0099] S36, extract features from the Fourier descriptor, the features including average radius, concavity and convexity, and symmetry;

[0100] S37, compare the features with the standard contour descriptor, evaluate the similarity and difference, and obtain the hash value of the difference by calculating the Euclidean distance;

[0101] S38. Repeat steps S31-S37 to identify multiple labels, calculate the average value using the least squares method, and determine and output whether the fabric is in a relaxed or taut state.

[0102] This application includes, but is not limited to, the fabric condition detection described above. Other methods can also be used for detection, such as directly collecting the tension of the clamping device on the fabric, but different tensions are applied to different fabrics to achieve different tension states.

[0103] This application does not limit the positioning control of the laser head. In order to improve accuracy, in one embodiment, S4 includes:

[0104] The X-axis and Y-axis motors of the laser head are controlled to move the laser head omnidirectionally on the feeding material. Combined with visual detection, a PID control algorithm is used to move the laser head to the starting position of the target label.

[0105] In this application, the motor controlling the laser head includes, but is not limited to, two-dimensional control; it can also be three-dimensional control. For position control, in addition to PID control algorithm, other algorithms can also be used.

[0106] This application requires label inspection, including position detection and damage detection. The specific inspection process is not limited. In one embodiment, step S5 includes:

[0107] The target label image is captured and obtained, and its contour is extracted using the Canny operator;

[0108] After converting the contour lines of the contour into Cartesian coordinate values, a continuous trajectory line is formed, and the cusps in the trajectory line are removed.

[0109] The trajectory line is interpolated using spline curves to make it a smooth curve;

[0110] The labels within the outline are visually inspected and identified. Label anomalies are identified through deep learning and image detection. These anomalies include thread breakage, stains, yarn breakage, and misalignment.

[0111] To further enable the detection of specific tags, in one embodiment, the target tag is a small tag, and the detection of the small tag further includes:

[0112] Enhance the target label image, perform white balance on the colors, and increase contrast;

[0113] The target label image is processed to obtain a Fourier descriptor for the target label image;

[0114] Pooling is performed on the Fourier descriptor of the target label image to preserve image features and reduce image size;

[0115] After normalizing and generating the co-occurrence probability matrix using the gray-level co-occurrence matrix, the uniformity and texture coarseness of the target label image in the corresponding channel and direction are calculated.

[0116] The target label image is fed into a pre-trained deep learning neural network for inference, and the trained defect class is obtained through target detection.

[0117] Merge the detection, repeat the above steps, collect and count the labels of multiple rows, perform statistical calculations based on the Fourier descriptor and the co-occurrence probability matrix, obtain the group with the largest difference and compare it with the standard label. If it is less than the threshold, all columns are judged to be qualified labels. Otherwise, mark the last one as defective and continue to find the label with the second largest difference, repeat the above qualified determination process.

[0118] By inspecting small tags, efficiency and applicability are improved.

[0119] This application does not limit the specific calculation process of using the gray-level co-occurrence matrix. In one embodiment, the gray-level co-occurrence matrix is ​​used to calculate the uniformity and texture coarseness of the target label image in the corresponding channel and direction, including:

[0120] The target label image is converted into a single-channel image;

[0121] The co-occurrence matrix is ​​calculated by selecting a direction perpendicular to the woven label direction and the distance between pixels, based on the woven label direction of the fabric.

[0122] Pixel values ​​are divided into 25 levels, and the co-occurrence frequency of each pixel with orientation and adjacent pixels at the single-channel level is counted.

[0123] Construct a 25x25 co-occurrence matrix, where each element CLCM(i,j) represents the number of times pixel levels i and j occur in pairs, given the chosen direction and distance;

[0124] The co-occurrence matrix is ​​normalized to obtain the co-occurrence probability matrix, where each element CLCM(i,j) represents the probability that i and j appear in pairs.

[0125] The image energy is calculated using the co-occurrence probability matrix and the following formula, reflecting the uniformity and texture coarseness of the target label image in the specified channel and direction.

[0126]

[0127] Repeat the above steps to calculate the red, green, and blue channels, and obtain three sets of cached data.

[0128] This application does not limit the process of listing the fabrics. In one embodiment, step S7 includes:

[0129] The fabric is moved forward to the arranging device, which is controlled by an X-axis motor, a Y-axis motor and a Z-axis motor. The head of the arranging device is equipped with an air pump that matches the number of labels in each row.

[0130] Based on the location information of the fabric and the location of the waste, control the air pump to move to the target point;

[0131] Move downwards and activate the air pump valves at all qualified label locations to attract and move the qualified labels to the listing area where the packaging boxes or baskets to be stored are placed.

[0132] Turn on the air pump, and after the label falls, move it downwards to press the label firmly.

[0133] This application includes, but is not limited to, the above-described methods of listing fabrics.

[0134] This application does not limit the specific control structure and control method of the air pump. In one embodiment, controlling the air pump to move to the target point involves controlling the air pump to adjust its position on the sliding guide rail, including:

[0135] Add the same number of air pumps to the sliding guide rail according to the number of labels in each row;

[0136] The spacing between adjacent air pumps is set according to the interval between the labels;

[0137] Once the cut label is moved to the listing area, the air pump moves down, sucks up the cut label, and then lifts it up.

[0138] The Y-axis motor of the air pump drives the label on the adsorption to move to the top of the box;

[0139] The air pump is lowered to press the adsorbed label into the box;

[0140] Return to the initial position to complete the collection and listing of a whole row of tags.

[0141] In one embodiment, the label defect identification, cutting, and collection method includes:

[0142] Fabric placement. The long roll of fabric is manually placed into the feed roll of the cutting machine and secured to the take-up roll by clamping devices, the cutting table, and the rear clamping device.

[0143] Label outline setting. Set the outer outline of the label to be cut in the software. Set the label's length and width in pixels, determine the initial position, and then click "Run". The device will automatically begin cutting.

[0144] Fabric feeding and changing. The automatic fabric changing process consists of clamping devices at both ends and a visual inspection system to detect the tightness. The two clamping devices control the distance between the two rollers via a push rod motor to control the fabric tension. When the tension is too high, the clamping device at the take-up end closes, and the clamping device at the feed end moves backward, easing the fabric. When the tension is too low, the clamping devices at both ends close, and the feed is taken in by controlling a single roller to maintain appropriate tension.

[0145] Meanwhile, a visual method is proposed to assess the tightness of fabric.

[0146] 1) First, the image is enhanced, including denoising, binarization and morphological processing, to reduce noise and improve accuracy.

[0147] 2) Use the Canny operator to perform edge detection and extract the outer contour of the label under the current camera.

[0148] 3) Convert the contour points to Cartesian coordinates.

[0149] 4) Convert these contour points into Fourier descriptors using Fourier series, and calculate them using the following formula.

[0150]

[0151] Where F(n) is the nth Fourier series, N is the total number of contour points, and x k and y k It is a Cartesian point on the outline.

[0152] 6) Extract features from the Fourier descriptor, such as average radius, concavity / convexity, and symmetry.

[0153] 7) Compare the extracted features with the standard contour descriptor, evaluate the similarity and differences, and calculate the Euclidean distance to obtain the hash value of the difference.

[0154] 8) Repeat steps 1-7 to identify multiple tags and use the least squares method to calculate the average value to determine whether the fabric is in a relaxed or taut state.

[0155] Based on visual assessment of tightness, the clamping devices at both ends are controlled to loosen or tighten. Simultaneously, differences from the standard contour descriptor are used to determine if there are large areas of incorrect labels.

[0156] Laser head alignment. By controlling the X and Y axis motors of the laser head, the laser head is moved omnidirectionally on the fabric. Combined with visual inspection, a PID control algorithm is used to move the laser head to the starting position of the target label.

[0157] Label recognition and identification, along with contour extraction planning, are performed. Images are captured using a camera, and contours are extracted using the Canny operator. The contours are converted to Cartesian coordinates, forming a coherent trajectory, and sharp points are removed. Spline curves are used to interpolate the trajectory, making it smoother. Simultaneously, labels within the contours are visually inspected and identified, using deep learning and image detection to recognize issues such as thread breaks, stains, yarn breaks, and misaligned yarns.

[0158] A detection method for small labels is proposed, and the specific method is as follows:

[0159] Step 1: Image enhancement, including white balance adjustment and increased contrast. A Gaussian kernel is then used to filter the image, smoothing it and reducing noise.

[0160] Step 2: Using the visual inspection method from Step 3, obtain the Fourier descriptor of the image. Initially locate the label area and check for defects that differ significantly from the standard label. Filter out large defects such as large-area defects, errors, and occlusions.

[0161] Step 3: Introduce an image pooling algorithm to pool the input image, preserving image features and reducing image size.

[0162] Average pooling is used. A center-extracted convolutional kernel is employed, and pooling is performed with a stride of 2 units.

[0163]

[0164] Step 4: Use the Gray-Level Co-occurrence Matrix (GLCM)

[0165] The calculation steps are as follows:

[0166] 1) Convert the image to a single-channel image.

[0167] 2) Calculate the co-occurrence matrix by selecting a direction perpendicular to the fabric label and the distance between pixels.

[0168] 3) Divide pixel values ​​into 25 levels. For each pixel, count the co-occurrence frequency at the single-channel level with respect to orientation and adjacent pixels.

[0169] 4) Construct a 25x25 co-occurrence matrix, where each element CLCM(i,j) represents the number of times pixel levels i and j appear in pairs under the selected direction and distance.

[0170] 5) Then, normalize the matrix to obtain the co-occurrence probability matrix, where each element CLCM(i,j) represents the probability that i and j occur in pairs.

[0171] 6) Substitute the matrix into the following formula to calculate the image energy, which reflects the uniformity and texture coarseness of the image in that channel and direction.

[0172]

[0173] 7) Repeat the above steps to calculate the red, green, and blue channels and obtain three sets of cached data.

[0174] Step 5: Feed the image into the pre-trained deep learning neural network for inference, and obtain the trained defect categories through object detection, such as dirt, misalignment, and open lines.

[0175] The YOLO model improved using the focal loss function. The focal loss function is formed by improving the cross-entropy loss function, and its specific form is as follows:

[0176]

[0177]

[0178]

[0179] Where p t It represents the degree of proximity of category y, p t The larger the value, the more accurate the classification; γ>0 is a moderating factor. The value represents the predicted probability; y represents the label corresponding to 1 or 0.

[0180] Step 6: Merge and detect. Repeat steps 1-5 to collect and statistically analyze the labels across multiple rows. Perform statistical calculations based on the Fourier descriptor of the contours and the co-occurrence probability matrix. Find the group with the largest difference and compare it to the standard label. If the difference is less than a threshold, all columns are considered acceptable labels. Otherwise, mark the last label as defective and continue searching for the label with the second largest difference. If the difference is too large, mark it as defective and continue until the requirements are met.

[0181] Selective cropping. Repeat the above steps to obtain smooth trajectory lines and recognition results for one or more columns of labels. Based on the recognition results, merge only the trajectory lines of qualified labels.

[0182] By controlling the laser head to cut, qualified labels are cut off, while defective labels are retained.

[0183] The qualified fabric is neatly arranged and collected. The fabric is moved forward by a traction device to the arranging unit. The arranging unit is controlled by X, Y, and Z axis motors, and its head is equipped with an air pump matching the number of tags per row. The location information of the fabric and the location of defective fabric are obtained through steps three and four. The air pump is moved to the target point and then downwards, activating the air pump valves at all qualified tag locations to attract and move the qualified tags to the arranging area. In the arranging area, packaging boxes or baskets for storage are placed. Then the air pump is turned on, and after the tags fall, it moves downwards again to compress the tags.

[0184] The air pump can be moved along the sliding guide rails to adjust the spacing. The detailed steps are as follows:

[0185] 1. Add the same number of air pumps to the sliding guide rail according to the number of labels in each row.

[0186] 2. Set the spacing between each air pump according to the spacing between the labels.

[0187] 3. Once the cut labels are moved to the display area, the air pump moves down to pick up the cut labels. Then it lifts them up.

[0188] 4. The Y-axis motor of the air pump drives the label on the adsorption plate to move to the top of the box.

[0189] 5. Lower the air pump to press the label into the box.

[0190] 6. Return to the initial position to complete the collection and listing of a whole row of tags.

[0191] Repeat steps two through five until the entire roll of fabric has undergone a fully automated process of identification, cutting, screening, listing, and collection.

[0192] In addition, embodiments of this application also provide a label defect identification, cutting, and collection system, including:

[0193] The fabric fixing device 100 includes a feeding roll and a clamping component. The feeding roll is used to store fabric, and the clamping component clamps the fabric.

[0194] The tension adjustment device 200 is used to adjust the spacing between the clamping components that clamp the fabric, thereby adjusting the tension of the fabric.

[0195] Alignment device 300 is used to control the X-axis motor and Y-axis motor of the laser head, so that the laser head moves in all directions on the feed material. Combined with visual detection, the laser head is moved to the starting position of the target label using a PID control algorithm.

[0196] The identification device 400 is used to capture an image of the target label, obtain a target label image, and perform label recognition and identification on the target label after extracting the outline of the target label image.

[0197] The cutting device 500 is used to cut the fabric corresponding to the qualified label identified by the identification device through the laser head, cut off the qualified label, and keep the defective label.

[0198] The arranging device 600 is used to neatly arrange and collect qualified fabrics corresponding to qualified labels.

[0199] Since the label defect identification, cutting and collection system is the same system as the label defect identification, cutting and collection method described above, it has the same beneficial effects.

[0200] In one embodiment, such as Figure 4 and Figure 5The fabric fixing device includes a feeding drum 1 and a clamping component, as well as a pulling device 2 and 10 for moving the fabric at the feeding end. The laser head 13 is equipped with a laser head X-axis motor 4 and a laser head Y-axis motor 5, and a camera 3 is set up to take pictures. The air pump uses the air pump X-axis motor 6 and the air pump Y-axis motor 7 to adsorb the material through the air pump suction head 8, and guides the material movement through the set guide wheel 9. The receiving drum 11, the clamping device 12, the clamping device 14 set at the feeding end, the stacking area 15 at the end, the packaging box 16 to be put in, the movement of the air pump is controlled by the sliding guide rail 17, and the status is displayed and the device status is controlled by the display screen controller 18.

[0201] In summary, the label defect identification, cutting, and collection method and system provided in this invention involves placing fabric into the feed roll of a cutting machine and fixing the fabric. The outer contour of the label to be cut is set, the fabric is stretched and maintained at a predetermined tension, and the laser head is controlled to move to the starting position of the target label to capture an image of the target label. After extracting the outline of the target label image, the target label is identified and authenticated. The laser head is then controlled to cut the fabric, cutting off qualified labels and retaining defective labels. The qualified fabric corresponding to the qualified labels is then neatly arranged and collected. By combining machine vision and artificial intelligence to locate and identify the cut labels, and controlling the two-dimensional laser to cut only qualified labels, defective labels can be quickly avoided. Qualified fabric is then collected and neatly arranged, achieving a fully automated cutting process from fabric replacement, identification, cutting to collection and arrangement. This reduces the cutting and collection of defective labels and the subsequent manual identification of defects, greatly improving production efficiency.

[0202] The label defect identification, cutting, and collection method and system provided by this invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. It should be noted that those skilled in the art can make several improvements and modifications to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the protection scope of the claims of this invention.

Claims

1. A method for identifying, cutting, and collecting label defects, characterized in that, A label defect identification, trimming, and collection system includes: S1, Place the fabric into the feed roll of the cutting machine and fix the fabric in place; S2, set the outer outline of the label to be cut; S3, the fabric is spread out and the fabric is kept under predetermined tension; S4, control the laser head to perform an alignment operation. By controlling the laser head to move omnidirectionally on the fabric and combining visual detection, the laser head is moved to the starting position of the target label. S5, take an image of the target label to obtain a target label image, and then perform label recognition and identification on the target label by extracting the outline of the target label image; S6, repeat S3-S5 to obtain smooth trajectory lines and recognition results of one or more columns of labels, merge the trajectory lines of qualified labels, control the laser head to cut the fabric, cut off the qualified labels, and keep the defective labels. S7, neatly arrange and collect the qualified fabrics corresponding to the qualified labels; S3 includes: The fabric is clamped at both ends by clamping devices. The distance between the two rollers is controlled by a push rod motor to control the distance between the two clamping devices, thereby controlling the tension of the fabric. Step S3 further includes: S31, Obtain the current tension image of the fabric; S32, perform enhancement processing on the tensioned image, the enhancement processing includes denoising, binarization and morphological processing; S33, use the Canny operator to perform edge detection and extract the outer contour of the label under the current camera; S34, transform the contour points of the outer contour into Cartesian contour points by using the Cartesian coordinate system; S35, convert the Cartesian contour points into Fourier descriptors using Fourier series; S36, extract features from the Fourier descriptor, the features including average radius, concavity and convexity, and symmetry; S37, compare the features with the standard contour descriptor, evaluate the similarity and difference, and obtain the hash value of the difference by calculating the Euclidean distance; S38, Repeat steps S31-S37 to identify multiple tags, calculate the average value using the least squares method, and determine and output whether the fabric is in a relaxed or taut state. S5 includes: The target label image is captured and obtained, and its contour is extracted using the Canny operator; After converting the contour lines of the contour into Cartesian coordinate values, a continuous trajectory line is formed, and the cusps in the trajectory line are removed. The trajectory line is interpolated using spline curves to make it a smooth curve; Visual inspection and identification are performed on the labels within the outline. Label anomalies are identified through deep learning and image detection. The label anomalies include broken thread ends, stains, broken yarns, and misaligned shuttles. The label defect identification, cutting, and collection system includes: A fabric fixing device includes a feeding roll and a clamping component, wherein the feeding roll is used to store fabric and the clamping component clamps the fabric; A tension adjustment device is used to adjust the spacing between the clamping components that clamp the fabric, thereby adjusting the tension of the fabric. The alignment device controls the X-axis and Y-axis motors of the laser head, enabling the laser head to move omnidirectionally on the fabric. Combined with visual detection, a PID control algorithm is used to move the laser head to the starting position of the target label. The identification device is used to capture an image of the target label to obtain an image of the target label. After extracting the outline of the target label image, the device performs label identification and authentication on the target label. The label identification and authentication includes identifying label anomalies through deep learning and image detection. The label anomalies include broken threads, stains, broken yarns, and misaligned yarns. Based on the identification results, the device determines whether the target label is a qualified label or a defective label. A cutting device is used to cut the fabric corresponding to the qualified label identified by the identification device through the laser head, cut off the qualified label, and keep the defective label; A sorting device is used to neatly sort and collect qualified fabrics corresponding to qualified labels.

2. The label defect identification, cutting, and collection method as described in claim 1, characterized in that, S4 includes; The X-axis and Y-axis motors of the laser head are controlled to move the laser head omnidirectionally on the fabric. Combined with visual detection, a PID control algorithm is used to move the laser head to the starting position of the target label.

3. The label defect identification, cutting, and collection method as described in claim 1, characterized in that, The target label is a small label, and the detection of the small label further includes: Enhance the target label image, perform white balance on the colors, and increase contrast; The target label image is processed to obtain a Fourier descriptor for the target label image; Pooling is performed on the Fourier descriptor of the target label image to preserve image features and reduce image size; After normalizing and generating the co-occurrence probability matrix using the gray-level co-occurrence matrix, the uniformity and texture coarseness of the target label image in the corresponding channel and direction are calculated. The target label image is fed into a pre-trained deep learning neural network for inference, and the trained defect class is obtained through target detection. Merge the detection, repeat the above steps, collect and count the labels of multiple rows, perform statistical calculations based on the Fourier descriptor and the co-occurrence probability matrix, obtain the group with the largest difference and compare it with the standard label. If it is less than the threshold, all columns are judged to be qualified labels. Otherwise, mark the last one as defective and continue to find the label with the second largest difference, repeat the above qualified determination process.

4. The label defect identification, cutting, and collection method as described in claim 3, characterized in that, Using the gray-level co-occurrence matrix, the uniformity and texture coarseness of the target label image in the corresponding channel and direction are calculated, including: The target label image is converted into a single-channel image; The co-occurrence matrix is ​​calculated by selecting a direction perpendicular to the woven label direction and the distance between pixels, based on the woven label direction of the fabric. Pixel values ​​are divided into 25 levels, and the co-occurrence frequency of each pixel with orientation and adjacent pixels at the single-channel level is counted. Construct a 25x25 co-occurrence matrix, where each element CLCM(i,j) represents the number of times pixel levels i and j occur in pairs, given the chosen direction and distance; The co-occurrence matrix is ​​normalized to obtain the co-occurrence probability matrix, where each element CLCM(i,j) represents the probability that i and j appear in pairs. The image energy is calculated using the co-occurrence probability matrix and the following formula, reflecting the uniformity and texture coarseness of the target label image in the specified channel and direction. ； Repeat the above steps to calculate the red, green, and blue channels, and obtain three sets of cached data.

5. The label defect identification, cutting, and collection method as described in claim 1, characterized in that, S7 includes: The fabric is moved forward to the arranging device, which is controlled by an X-axis motor, a Y-axis motor and a Z-axis motor. The head of the arranging device is equipped with an air pump that matches the number of labels in each row. Based on the location information of the fabric and the location of the waste, control the air pump to move to the target point; Move downwards and activate the air pump valves at all qualified label locations to attract and move the qualified labels to the listing area where the packaging boxes or baskets to be stored are placed. Turn on the air pump, and after the label falls, move it downwards to press the label firmly.

6. The label defect identification, cutting, and collection method as described in claim 5, characterized in that, Controlling the air pump to move to the target point involves adjusting the position of the air pump on the sliding guide rail, including: Add the same number of air pumps to the sliding guide rail according to the number of labels in each row; The spacing between adjacent air pumps is set according to the interval between the labels; Once the cut label is moved to the listing area, the air pump moves down, sucks up the cut label, and then lifts it up. The Y-axis motor of the air pump drives the label on the adsorption to move to the top of the box; The air pump is lowered to press the adsorbed label into the box; Return to the initial position to complete the collection and listing of a whole row of tags.

Images