A method for intelligent inspection and control of quality in the production process of socks production line

Abstract

The present invention relates to the technical field of sock production detection, specifically to a method for intelligent inspection and control of the quality of a sock production process; comprising: sampling to obtain a first sock sample, determining a corresponding first mold and a second mold according to the model data of the first sock sample; photographing the first sock sample according to the mold to obtain a first stitching image and a second stitching image; constructing a stitching abnormality recognition model to recognize the first stitching image and the second stitching image; obtaining a stitching structure abnormality coefficient, a stitching position abnormality coefficient and a stitching area abnormality coefficient, and calculating to obtain a stitching quality abnormality coefficient, and judging stitching quality according to the stitching quality abnormality coefficient; obtaining a stitching quality abnormality coefficient of the first sock sample, and obtaining a stitching abnormality time series according to stitching time sorting; constructing a pipeline early warning recognition model to analyze the stitching abnormality time series to obtain a pipeline early warning coefficient. The present invention controls the production line through the pipeline early warning coefficient.

Description

Technical Field

[0001] The invention relates to the technical field of sock production detection, and in particular to a method for intelligent quality inspection and control of a sock production line production process. Background Art

[0002] Socks are one of the common daily textile products that can protect the feet, absorb moisture and perspiration, and keep warm. During the sock production process, socks are designed differently for left and right feet, so that the socks have a better fit and increase comfort. They also provide better support for people who often engage in high-intensity exercise, and therefore are gradually becoming popular in the market.

[0003] The main production process of socks includes knitting, sewing, dyeing, shaping and packaging. In the knitting stage, the socks are knitted into a cylindrical shape through knitting equipment; in the sewing stage, the opening of the toe is further sewn up, which is also called toe stitching. The toe stitching area is one of the parts of socks that are most prone to quality problems, and its stitching quality directly affects the wearing comfort and durability.

[0004] As sock knitting and sewing equipment age, their needle length and needle pressure may change, and errors may occur in sewing force and precision; thus, the sewing effect of socks is reduced during the sock sewing stage, affecting product quality. Therefore, it is necessary to conduct quality inspection and identification of the sewing steps in the sock production process.

[0005] In the existing technology, there are few technologies for detecting and identifying the stitching quality of sock toes. With the development of image recognition technology, it is a feasible way to identify defects by acquiring images of the stitching area, and image recognition technology can effectively improve the efficiency and accuracy of identification.

[0006] Therefore, an intelligent inspection and control method for the quality of socks production line was proposed. Summary of the invention

[0007] The purpose of the present invention is to provide a method for intelligent inspection and control of the quality of a sock production line; the method comprises the following steps: sampling a first sock sample, determining a corresponding first mold and a second mold according to the model data of the first sock sample; photographing the first sock sample according to the mold to obtain a first stitching image and a second stitching image; constructing a stitching abnormality recognition model to recognize the first stitching image and the second stitching image; obtaining a stitching structure abnormality coefficient, a stitching position abnormality coefficient and a stitching area abnormality coefficient, and calculating and obtaining a stitching quality abnormality coefficient, and judging stitching quality according to the stitching quality abnormality coefficient; obtaining a stitching quality abnormality coefficient of the first sock sample, and obtaining a stitching abnormality time series according to stitching time; constructing a pipeline early warning recognition model to analyze the stitching abnormality time series to obtain a pipeline early warning coefficient. The present invention controls the production line through the pipeline early warning coefficient.

[0008] To achieve the above object, the present invention provides the following technical solutions: A method for intelligent inspection and control of quality in a socks production line production process, comprising: S10. Sampling the socks products that have been sewn at the toe on the socks production line to obtain the first sock sample; S20. Identify the model data of the first sock sample, and determine the corresponding first mold and second mold according to the model data; wear the first sock sample on the first mold, and photograph the seam to obtain a first stitching image; turn the first sock sample over so that the inner surface of the first sock sample faces outward, and then wear the first sock sample on the second mold, and photograph the seam to obtain a second stitching image; S30. Constructing a suture abnormality recognition model to recognize the first sutured image and the second sutured image; the suture abnormality recognition model includes a structure abnormality recognition unit, a position abnormality recognition unit and a region abnormality recognition unit, which respectively recognize the first sutured image and the second sutured image to obtain a suture structure abnormality coefficient, a suture position abnormality coefficient and a suture region abnormality coefficient; S40. Calculate the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient, and judge the suture quality according to the suture quality abnormality coefficient; S50. Obtain the stitching quality abnormality coefficient of the first sock sample, and sort the samples according to the stitching time sequence of the first sock sample to obtain a stitching abnormality time series; S60. Construct an assembly line early warning recognition model, analyze the suture abnormality time series, and obtain an assembly line early warning coefficient; and control the production line according to the assembly line early warning coefficient.

[0009] The process of determining the first mold and the second mold according to the model data of the first sock sample is as follows: The model data of the first sock sample includes type data and size data, the type data includes left foot type and right foot type; the size data includes length data and width data; Obtaining size data of a first sock sample, and obtaining a left foot mold and a right foot mold of corresponding sizes; The type data of the first sock sample is obtained, and a mold having the same type as the first sock sample is used as a first mold; and a mold having a different type from the first sock sample is used as a second mold.

[0010] During the shooting process of the first stitched image and the second stitched image, the camera device is located directly above the mold to shoot the stitching position; and the relative position between the camera device and the mold is fixed, so that in the first stitched image and the second stitched image, the positions of the first mold and the second mold are symmetrical.

[0011] The suture abnormality recognition model includes a structural abnormality recognition unit, a position abnormality recognition unit and a regional abnormality recognition unit; The structural abnormality recognition unit recognizes the first stitched image to obtain a first structural abnormality coefficient; recognizes the second stitched image to obtain a second structural abnormality coefficient; and weights the first structural abnormality coefficient and the second structural abnormality coefficient to obtain a stitched structural abnormality coefficient; The position abnormality recognition unit recognizes the first stitched image to obtain a first position abnormality coefficient; recognizes the second stitched image to obtain a second position abnormality coefficient; and weights the first position abnormality coefficient and the second position abnormality coefficient to obtain a stitching position abnormality coefficient; The regional anomaly identification unit identifies the first stitched image to obtain a first regional anomaly coefficient; identifies the second stitched image to obtain a second regional anomaly coefficient; and obtains a stitched regional anomaly coefficient by weighting the first regional anomaly coefficient and the second regional anomaly coefficient.

[0012] The recognition process of the structural abnormality recognition unit on the first stitched image is as follows: The stitching area of ​​the first stitching image is identified according to the texture features and the color features to obtain an initial stitching area; and then the yarn structure at the edge of the area where the initial stitching area is located is calibrated to obtain a target stitching area; The yarn distribution structure in the target stitching area is obtained, and the yarn distribution structure is compared and identified with the corresponding standard yarn distribution structure to obtain first structure difference data; and a first structure abnormality coefficient is obtained according to the first structure difference data.

[0013] The process of the position abnormality recognition unit recognizing the first stitched image is as follows: Identify a target stitching area of ​​the first stitching image; construct a coordinate system on the first stitching image, and obtain position distribution data of the target stitching area according to the coordinate system; The position distribution data of the target suture area is compared and identified with the corresponding standard position distribution data to obtain first position difference data; and a first position abnormality coefficient is obtained according to the first position difference data.

[0014] The process of the regional abnormality identification unit identifying the first stitched image is as follows: Acquire a target stitching area of ​​the first stitching image, acquire a sock area around the target stitching area according to a distance threshold, and obtain a stitching adjacent area; The stitching adjacent area is identified according to the yarn structure data and color distribution data of the standard stitching adjacent area to obtain the first area difference data, and the first area abnormality coefficient is obtained according to the first area difference coefficient.

[0015] The formula for calculating the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient is: ; in, It represents the abnormal coefficient of suture quality; represents the structural weight; represents the coefficient of abnormality of suture structure; represents the position weight; represents the coefficient of abnormal suture position; represents the regional weight; Represents the abnormal coefficient of the suture area.

[0016] The process of judging suture quality according to suture quality abnormality coefficient is as follows: ; in, Indicates the suture quality. When the suture quality is equal to 1, it means the suture quality is qualified; when the suture quality is equal to 0, it means the suture quality is unqualified; It represents the abnormal coefficient of suture quality; Indicates the abnormal suture quality threshold.

[0017] The pipeline identification and warning model is constructed based on the LSTM model, and its training process is as follows: Data collection and processing are performed on the sock production line to obtain a historical production data set; the historical production data set includes a time series of historical stitching quality abnormality coefficients and a time period when abnormalities occurred in processing equipment; According to the time period when the processing equipment is abnormal and the time point of the historical stitching quality abnormal coefficient, the historical production data set is divided to obtain a normal production data set and an abnormal production data set; after data preprocessing of the normal production data set and the abnormal production data set, an early warning model training data set is obtained; The initialized pipeline recognition warning model is trained and optimized according to the warning model training data set.

[0018] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention obtains type data and size data of socks, determines a first mold and a second mold according to the size data and type data of the socks; fixes the relative position of a camera device and the mold, wears a first sock sample on the first mold, and photographs a seam to obtain a first stitching image; turns the first sock sample over so that the inner surface of the first sock sample faces outward, then wears the first sock sample on the second mold, and photographs a seam to obtain a second stitching image; accurately obtains a first stitching image and a second stitching image having a corresponding relationship, and provides a data basis for identifying abnormalities at the seam.

[0019] 2. The present invention identifies the structural anomaly of the stitched image through a structural anomaly identification unit. First, the stitching area of ​​the stitched image is identified according to texture features and color features to obtain an initial stitching area. Then, the yarn structure at the edge of the initial stitching area is identified and calibrated to obtain a target stitching area. The structural anomaly coefficient is accurately identified through the difference between the yarn distribution structure of the target stitching area and the standard yarn distribution structure.

[0020] 3. The present invention identifies the position anomaly of the stitched image through a position anomaly identification unit; identifies the target stitching area of ​​the stitched image; constructs a coordinate system on the stitched image, and obtains the position distribution data of the target stitching area according to the coordinate system; accurately identifies the position anomaly coefficient through the difference data between the position distribution data of the target stitching area and the corresponding standard position distribution data.

[0021] 4. The present invention identifies regional anomalies of a stitched image through a regional anomaly identification unit; obtains a target stitching area in the stitched image, and obtains a stitching adjacent area based on an edge of the target stitching area and a preset distance threshold; identifies the stitching adjacent area based on yarn structure data and color distribution data of a standard stitching adjacent area to obtain regional difference data; and accurately identifies a regional anomaly coefficient based on the regional difference data.

[0022] 5. The present invention performs weighted measurement based on the identified abnormal coefficient of the suture structure, the abnormal coefficient of the suture position and the abnormal coefficient of the suture area to obtain the abnormal coefficient of the suture quality; the suture quality coefficient is distinguished by the abnormal threshold of the suture quality to obtain qualified suture products and unqualified suture products, and the suture quality of the socks products is accurately identified; and the suture abnormality time series is analyzed by the assembly line identification and early warning model to obtain the assembly line early warning coefficient, and the production line can be accurately controlled by the assembly line early warning coefficient. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 It is a flow chart of a method for intelligent quality inspection and control of a socks production line production process of the present invention; Figure 2 It is a structural schematic diagram of the suture abnormality recognition model of the present invention; Figure 3 is a schematic diagram of the stitching image and coordinate system construction of the first sock sample Y of the present invention; Figure 4 is a schematic diagram of a stitching position in a second stitching image of a first sock sample Y of the present invention; Figure 5 It is a schematic diagram of the fabric yarn structure of the stitching adjacent area of ​​the present invention.

[0024] In the figure: 10, first stitched image; 20, second stitched image; a, first origin; 11, first x-axis; 12, first y-axis; b, second origin; 21, second x-axis; 22, second y-axis. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] Embodiment 1 The present invention proposes a method for intelligent inspection and control of the quality of a socks production line. The process of the method is as follows: Figure 1 As shown, including: S10. Sampling the socks products that have been stitched on the socks production line to obtain the first socks sample.

[0027] S20. Identify the model data of the first sock sample, and determine the corresponding first mold and second mold according to the model data; wear the first sock sample on the first mold, and photograph the seams to obtain a first stitching image; turn the first sock sample over so that the inner surface of the first sock sample faces outward, and then wear the first sock sample on the second mold, and photograph the seams to obtain a second stitching image.

[0028] The process of determining the first mold and the second mold according to the model data of the first sock sample is as follows: The model data of the first sock sample includes type data and size data, the type data includes left foot type and right foot type; the size data includes length data and width data; the size data of the first sock sample is obtained, and a left foot mold and a right foot mold of corresponding sizes are obtained; the type data of the first sock sample is obtained, and a mold consistent with the type of the first sock sample is used as a first mold; and a mold inconsistent with the type of the first sock sample is used as a second mold.

[0029] During the shooting process of the first stitched image and the second stitched image, the camera device is located directly above the mold to shoot the stitching position; and the relative position between the camera device and the mold is fixed, so that in the first stitched image and the second stitched image, the positions of the first mold and the second mold are symmetrical.

[0030] The present invention obtains type data and size data of socks, determines a left foot mold and a right foot mold of corresponding sizes according to the size data of the socks, and then determines a first mold and a second mold according to the type data of the socks; fixes the relative position of a camera device and the mold, wears a first sock sample on the first mold, and photographs a seam to obtain a first seam image; turns the first sock sample over so that the inner surface of the first sock sample faces outward, then wears the first sock sample on the second mold, and photographs the seam to obtain a second seam image; accurately obtains a first seam image and a second seam image with a corresponding relationship, and provides a data basis for identifying abnormalities at the seam.

[0031] S30. Construct a suture abnormality recognition model to recognize the first sutured image and the second sutured image; the suture abnormality recognition model includes a structure abnormality recognition unit, a position abnormality recognition unit and a region abnormality recognition unit, which respectively recognize the first sutured image and the second sutured image to obtain a suture structure abnormality coefficient, a suture position abnormality coefficient and a suture region abnormality coefficient.

[0032] The structure of the suture abnormality recognition model is as follows Figure 2 As shown, it includes a structural anomaly recognition unit, a position anomaly recognition unit and a regional anomaly recognition unit; The structural abnormality recognition unit recognizes the first stitched image to obtain a first structural abnormality coefficient; recognizes the second stitched image to obtain a second structural abnormality coefficient; and weights the first structural abnormality coefficient and the second structural abnormality coefficient to obtain a stitched structural abnormality coefficient; The position abnormality recognition unit recognizes the first stitched image to obtain a first position abnormality coefficient; recognizes the second stitched image to obtain a second position abnormality coefficient; and weights the first position abnormality coefficient and the second position abnormality coefficient to obtain a stitching position abnormality coefficient; The regional anomaly identification unit identifies the first stitched image to obtain a first regional anomaly coefficient; identifies the second stitched image to obtain a second regional anomaly coefficient; and obtains a stitched regional anomaly coefficient by weighting the first regional anomaly coefficient and the second regional anomaly coefficient.

[0033] The present invention constructs a suture abnormality recognition model to recognize a first sutured image and a second sutured image; the model comprises the following steps: recognizing the first sutured image and the second sutured image by a structure abnormality recognition unit to obtain a suture structure abnormality coefficient; recognizing the first sutured image and the second sutured image by a position abnormality recognition unit to obtain a suture position abnormality coefficient; recognizing the first sutured image and the second sutured image by a region abnormality recognition unit to obtain a suture region abnormality coefficient; and accurately recognizing the suture abnormality by the suture structure, the suture position and the suture region.

[0034] The recognition process of the structural abnormality recognition unit on the first stitched image is as follows: The stitching area of ​​the first stitching image is identified according to the texture features and the color features to obtain an initial stitching area; and then the yarn structure at the edge of the area where the initial stitching area is located is calibrated to obtain a target stitching area; The yarn distribution structure in the target stitching area is obtained, and the yarn distribution structure is compared and identified with the corresponding standard yarn distribution structure to obtain first structure difference data; and a first structure abnormality coefficient is obtained according to the first structure difference data.

[0035] The present invention identifies the structural abnormality of a stitched image through a structural abnormality identification unit. First, the stitching area of ​​the stitched image is identified according to texture features and color features to obtain an initial stitching area. Then, the yarn structure at the edge of the initial stitching area is identified and calibrated to obtain a target stitching area. The structural abnormality coefficient is accurately identified through the difference between the yarn distribution structure of the target stitching area and the standard yarn distribution structure.

[0036] The process of the position abnormality recognition unit recognizing the first stitched image is as follows: Identify a target stitching area of ​​the first stitching image; construct a coordinate system on the first stitching image, and obtain position distribution data of the target stitching area according to the coordinate system; The position distribution data of the target suture area is compared and identified with the corresponding standard position distribution data to obtain first position difference data; and a first position abnormality coefficient is obtained according to the first position difference data.

[0037] The present invention identifies the position anomaly of a stitched image through a position anomaly identification unit; identifies and obtains a target stitching area of ​​the stitched image; constructs a coordinate system on the stitched image, and obtains position distribution data of the target stitching area according to the coordinate system; and accurately identifies and obtains a position anomaly coefficient through difference data between the position distribution data of the target stitching area and corresponding standard position distribution data.

[0038] The process of the regional abnormality identification unit identifying the first stitched image is as follows: Acquire a target stitching area of ​​the first stitching image, acquire a sock area around the target stitching area according to a distance threshold, and obtain a stitching adjacent area; The stitching adjacent area is identified according to the yarn structure data and color distribution data of the standard stitching adjacent area to obtain the first area difference data, and the first area abnormality coefficient is obtained according to the first area difference coefficient.

[0039] The present invention identifies regional anomalies of a stitched image through a regional anomaly identification unit; obtains a target stitching area in the stitched image, and obtains a stitching adjacent area according to an edge of the target stitching area and a preset distance threshold; identifies the stitching adjacent area according to yarn structure data and color distribution data of a standard stitching adjacent area to obtain regional difference data; and accurately identifies a regional anomaly coefficient according to the regional difference data.

[0040] Among them, the standard yarn distribution structure, standard position distribution data and standard stitching adjacent area data are obtained by identifying and obtaining qualified sock samples.

[0041] S40. Calculate the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient, and judge the suture quality according to the suture quality abnormality coefficient.

[0042] The formula for calculating the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient is: ; in, It represents the abnormal coefficient of suture quality; represents the structural weight; represents the coefficient of abnormality of suture structure; represents the position weight; represents the coefficient of abnormal suture position; represents the regional weight; Represents the abnormal coefficient of the suture area.

[0043] The process of judging suture quality according to suture quality abnormality coefficient is as follows: ; in, Indicates the suture quality. When the suture quality is equal to 1, it means the suture quality is qualified; when the suture quality is equal to 0, it means the suture quality is unqualified; It represents the abnormal coefficient of suture quality; Indicates the abnormal suture quality threshold.

[0044] The present invention performs weighted measurement based on the identified abnormal coefficients of the stitching structure, the abnormal coefficients of the stitching position and the abnormal coefficients of the stitching area to obtain the abnormal coefficients of the stitching quality; the stitching quality coefficients are distinguished by the abnormal threshold of the stitching quality to obtain qualified stitching products and unqualified stitching products, and the stitching quality of the socks products is accurately identified.

[0045] S50. Obtain the stitching quality abnormality coefficient of the first sock sample, and sort the stitching quality coefficients of the first sock sample according to the stitching time sequence to obtain a stitching abnormality time series.

[0046] S60. Construct an assembly line early warning recognition model, analyze the suture abnormality time series, and obtain an assembly line early warning coefficient; and control the production line according to the assembly line early warning coefficient.

[0047] The pipeline identification and warning model is constructed based on the LSTM model, and its training process is as follows: Data collection and processing are performed on the sock production line to obtain a historical production data set; the historical production data set includes a time series of historical stitching quality abnormality coefficients and a time period when abnormalities occurred in processing equipment; According to the time period when the processing equipment is abnormal and the time point of the historical stitching quality abnormal coefficient, the historical production data set is divided to obtain a normal production data set and an abnormal production data set; after data preprocessing of the normal production data set and the abnormal production data set, an early warning model training data set is obtained; The initialized pipeline recognition warning model is trained and optimized according to the warning model training data set.

[0048] The present invention collects and processes data on a sock production line to obtain a historical production data set; divides the historical data set according to the time period when abnormalities occur in processing equipment and the time point of historical stitching quality abnormality coefficients to obtain an early warning model training data set; and trains and optimizes an initialized pipeline recognition early warning model according to the early warning model training data set.

[0049] The present invention obtains a first sock sample by sampling, and determines the corresponding first mold and second mold according to the model data of the first sock sample; photographs the first sock sample according to the mold to obtain a first stitching image and a second stitching image; constructs a stitching abnormality recognition model to recognize the first stitching image and the second stitching image; obtains a stitching structure abnormality coefficient, a stitching position abnormality coefficient and a stitching area abnormality coefficient, and calculates and obtains a stitching quality abnormality coefficient, and judges the stitching quality according to the stitching quality abnormality coefficient; obtains the stitching quality abnormality coefficient of the first sock sample, and obtains a stitching abnormality time series according to the stitching time sequence; constructs an assembly line early warning recognition model to analyze the stitching abnormality time series to obtain an assembly line early warning coefficient. The present invention controls the production line through the assembly line early warning coefficient.

[0050] Embodiment 2 Sock weaving factory G is mainly engaged in the production and weaving of socks. Its weaving process includes knitting and sewing. The socks are first knitted into a cylindrical shape, and then the opening in the toe area is sewn up. However, in the actual production process, with the aging of machinery and equipment, deviations are easily caused during the knitting and sewing of socks, affecting the quality of sock sewing.

[0051] Therefore, the sock weaving factory G adopts the intelligent inspection and control method for the quality of the sock assembly line production process described in the present invention to detect and identify the sewing process.

[0052] The process of the intelligent inspection and control method for the quality of the socks production line production process is as follows: Figure 1 As shown, including: S10. Sampling the socks products that have been sewn at the toe on the socks production line to obtain the first sock sample; S20. Identify the model data of the first sock sample, and determine the corresponding first mold and second mold according to the model data; wear the first sock sample on the first mold, and photograph the seam to obtain a first stitching image; turn the first sock sample over so that the inner surface of the first sock sample faces outward, and then wear the first sock sample on the second mold, and photograph the seam to obtain a second stitching image; The process of determining the first mold and the second mold according to the model data of the first sock sample is as follows: The model data of the first sock sample includes type data and size data, the type data includes left foot type and right foot type; the size data includes length data and width data; Obtaining size data of a first sock sample, and obtaining a left foot mold and a right foot mold of corresponding sizes; The type data of the first sock sample is obtained, and a mold having the same type as the first sock sample is used as a first mold; and a mold having a different type from the first sock sample is used as a second mold.

[0053] Sock knitting factory G samples and obtains a first sock sample, and collects type data and size data of the first sock sample, and selects a left foot mold and a right foot mold of corresponding sizes according to the size data of the first sock sample; if the type of the first sock sample is a left foot type, the left foot mold is used as the first mold, and the right foot mold is used as the second mold; if the type of the first sock sample is a right foot type, the right foot mold is used as the first mold, and the left foot mold is used as the second mold; Taking the first sock sample Y as an example, the first sock sample Y is a right foot type, so a right foot mold of a corresponding size is selected as the first mold, and a left foot mold of a corresponding size is selected as the second mold; Wear the first sock sample Y on the first mold, and photograph the seam directly above the mold by using a camera to obtain a first stitched image; then flip the first sock sample Y so that the inner surface of the first sock sample Y faces outward, and then wear the first sock sample Y on the second mold, and photograph the second mold in the same manner as the first stitched image, so that the positions of the first mold and the second mold are symmetrical in the first stitched image and the second stitched image; S30. Constructing a suture abnormality recognition model to recognize the first sutured image and the second sutured image; the suture abnormality recognition model includes a structure abnormality recognition unit, a position abnormality recognition unit and a region abnormality recognition unit, which respectively recognize the first sutured image and the second sutured image to obtain a suture structure abnormality coefficient, a suture position abnormality coefficient and a suture region abnormality coefficient; The suture abnormality recognition model includes a structural abnormality recognition unit, a position abnormality recognition unit and a regional abnormality recognition unit; The structural abnormality identification unit identifies the first stitched image to obtain a first structural abnormality coefficient; identifies the second stitched image to obtain a second structural abnormality coefficient; and obtains the stitched structural abnormality coefficient by weighting the first structural abnormality coefficient and the second structural abnormality coefficient; the weights of the first structural abnormality coefficient and the second structural abnormality coefficient are both 0.5.

[0054] The position anomaly recognition unit recognizes the first stitched image to obtain a first position anomaly coefficient; recognizes the second stitched image to obtain a second position anomaly coefficient; and obtains a stitching position anomaly coefficient by weighting the first position anomaly coefficient and the second position anomaly coefficient; the weights of the first position anomaly coefficient and the second position anomaly coefficient are both 0.5.

[0055] The regional anomaly identification unit identifies the first stitched image to obtain a first regional anomaly coefficient; identifies the second stitched image to obtain a second regional anomaly coefficient; and obtains a stitched regional anomaly coefficient by weighting the first regional anomaly coefficient and the second regional anomaly coefficient; the weights of the first regional anomaly coefficient and the second regional anomaly coefficient are both 0.5.

[0056] The above weight settings reflect the importance of the inner and outer surface stitching areas, and are obtained through customer surveys and sales data analysis.

[0057] The recognition process of the structural abnormality recognition unit on the first stitched image is as follows: The stitching area of ​​the first stitching image is identified according to the texture features and the color features to obtain an initial stitching area; and then the yarn structure at the edge of the area where the initial stitching area is located is calibrated to obtain a target stitching area; The yarn distribution structure in the target stitching area is obtained, and the yarn distribution structure is compared and identified with the corresponding standard yarn distribution structure to obtain first structure difference data; and a first structure abnormality coefficient is obtained according to the first structure difference data.

[0058] The structural abnormality of the suture specifically includes missing, broken, falling off or irregular sutures at the suture site, and whether the spacing between the sutures is uniform, etc., and is mainly used to identify abnormal distribution of sutures at the suture site.

[0059] In order to verify the recognition effect of the structural abnormality recognition unit, experiments are conducted on it, including: a first structural recognition unit, a second structural recognition unit and a third structural recognition unit; Among them, the first structure recognition unit is the structural abnormality recognition unit described in the present invention, that is, the stitching area of ​​the first stitching image is firstly recognized by texture features and color features to obtain an initial stitching area; then the yarn structure at the edge of the area where the initial stitching area is located is calibrated to obtain the first stitching area; the second structure recognition unit recognizes the stitching area of ​​the first stitching image only by texture features and color features to obtain the second stitching area; the third structure recognition unit recognizes the stitching area only by the yarn structure to obtain the third stitching area; The present invention uses the F1 index as the recognition effect label of the structural anomaly recognition unit; the calculation formula of the F1 index is: ; in, represents the F1 index; It represents the accuracy, which is the ratio of the overlap area between the identified stitching area and the actual stitching area to the area of ​​the identified stitching area; It represents the recall rate, which is the ratio of the area of ​​the overlap between the identified stitching area and the actual stitching area to the area of ​​the actual stitching area.

[0060] The experimental data obtained through multiple verifications are shown in Table 1.

[0061] Table 1 Structural anomaly recognition unit verification data table

[0062] It can be seen from the experimental data in Table 1 that the F1 score of the first structure recognition unit is the highest, so the recognition effect of the structural anomaly recognition unit described in the present invention is the best.

[0063] The process of the position abnormality recognition unit recognizing the first stitched image is as follows: Identify a target stitching area of ​​the first stitching image; construct a coordinate system on the first stitching image, and obtain position distribution data of the target stitching area according to the coordinate system; The position distribution data of the target suture area is compared and identified with the corresponding standard position distribution data to obtain first position difference data; and a first position abnormality coefficient is obtained according to the first position difference data.

[0064] The position abnormality recognition unit is mainly used to recognize the position abnormality of the suture area; Taking the first sock sample Y as an example, its coordinate system is constructed as follows Figure 3 As shown; Figure 3 10 is the first stitched image, i.e., the image of the outer surface Y of the first sock sample; 20 is the second stitched image, i.e., the image of the inner surface Y of the first sock sample; a represents the first origin; 11 represents the first x-axis; 12 represents the first y-axis; b represents the second origin; 21 represents the second x-axis; and 22 represents the second y-axis.

[0065] The first origin and the second origin are points in the first mold and the second mold that have a corresponding relationship, respectively, and the first coordinate axis and the second coordinate axis are in a corresponding relationship.

[0066] Figure 4 The seam scene of the second stitched image of the first sock sample Y is shown.

[0067] The process of the regional abnormality identification unit identifying the first stitched image is as follows: The target stitching area of ​​the first stitching image is obtained, and the sock area around the target stitching area is obtained according to a distance threshold; and a stitching adjacent area is obtained; the distance threshold is determined according to the area and distribution of the stitching area.

[0068] The stitching adjacent area is identified according to the yarn structure data and color distribution data of the standard stitching adjacent area to obtain the first area difference data, and the first area abnormality coefficient is obtained according to the first area difference coefficient.

[0069] The regional anomaly recognition unit analyzes the stress conditions at the edge of the stitching area. If the stitching line is too loose or too tight, it may cause the fabric to wrinkle or deform. If the tension is too high, the fabric will tear. If the tension is too low, the stitching may become loose. Different situations will cause the yarn structure of the fabric to change at the edge of the stitching area. The stitching image of the first sock sample is identified to obtain the yarn structure of the fabric in the stitching adjacent area. Figure 5 shown.

[0070] According to the stitching anomaly recognition model, the first sock sample obtained by sampling the sock weaving factory G is identified, and the stitching structure anomaly coefficient, stitching position anomaly coefficient and stitching area anomaly coefficient are obtained. The obtained data are shown in Table 2: Table 2: Abnormal identification data table of the first sock sample

[0071] S40. Calculate the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient, and judge the suture quality according to the suture quality abnormality coefficient; The formula for calculating the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient is: ; in, It represents the abnormal coefficient of suture quality; represents the structural weight; represents the coefficient of abnormality of suture structure; represents the position weight; represents the coefficient of abnormal suture position; represents the regional weight; Represents the abnormal coefficient of the suture area.

[0072] The process of judging suture quality according to suture quality abnormality coefficient is as follows: ; in, Indicates the suture quality. When the suture quality is equal to 1, it means the suture quality is qualified; when the suture quality is equal to 0, it means the suture quality is unqualified; It represents the abnormal coefficient of suture quality; Indicates the abnormal suture quality threshold.

[0073] The structural weight, position weight, regional weight and stitching quality abnormality threshold are obtained based on the historical data of sock production.

[0074] The present invention performs weighted measurement based on the identified abnormal coefficients of the stitching structure, the abnormal coefficients of the stitching position and the abnormal coefficients of the stitching area to obtain the abnormal coefficients of the stitching quality; the stitching quality coefficients are distinguished by the abnormal threshold of the stitching quality to obtain qualified stitching products and unqualified stitching products, and the stitching quality of the socks products is accurately identified.

[0075] In order to verify the judgment effect of the suture quality abnormality coefficient obtained by the suture abnormality recognition model of the present invention on the suture quality, a verification experiment was carried out on the suture abnormality recognition model.

[0076] In the verification of the effect of recognition and quality judgment of the suture abnormality recognition model, the present invention constructs four models, including model one, model two, model three and model four; Among them, model 1 is the suture abnormality identification model of the present invention, which measures and obtains the suture quality abnormality coefficient by identifying the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient; Model 2 is based on Model 1, without considering the abnormal coefficient of suture structure. It only identifies the abnormal coefficient of suture position and the abnormal coefficient of suture area to calculate the abnormal coefficient of suture quality. Model 3 is based on Model 1, but does not consider the abnormal coefficient of suture position. It only identifies the abnormal coefficient of suture structure and the abnormal coefficient of suture area to calculate the abnormal coefficient of suture quality. Model 4 is based on model 1. It does not consider the abnormal coefficient of the suture area. It only identifies the abnormal coefficient of the suture structure and the abnormal coefficient of the suture position to calculate the abnormal coefficient of the suture quality.

[0077] The stitching quality of the sock samples was judged by the stitching quality anomaly coefficients obtained by different models, and the F1 index of each model was calculated based on the judgment results. The F1 index is calculated based on the precision and recall of each model. The precision of the judgment is the ratio of the number of samples predicted to be abnormal samples and actually abnormal samples to the number of samples predicted to be abnormal samples. The recall of the judgment is the ratio of the number of samples predicted to be abnormal samples and actually abnormal samples to the number of actual abnormal samples.

[0078] The obtained data are shown in Table 3.

[0079] Table 3. Suture abnormality recognition model validation data table

[0080] According to the data in Table 3, model 1 has the best recognition effect on the stitching quality of sock samples.

[0081] S50. Obtain the stitching quality abnormality coefficient of the first sock sample, and sort the samples according to the stitching time sequence of the first sock sample to obtain a stitching abnormality time series; S60. Construct an assembly line early warning recognition model, analyze the suture abnormality time series, and obtain an assembly line early warning coefficient; and control the production line according to the assembly line early warning coefficient.

[0082] The pipeline identification and warning model is constructed based on the LSTM model, and its training process is as follows: Data collection and processing are performed on the sock production line to obtain a historical production data set; the historical production data set includes a time series of historical stitching quality abnormality coefficients and a time period when abnormalities occurred in processing equipment; According to the time period when the processing equipment is abnormal and the time point of the historical stitching quality abnormal coefficient, the historical production data set is divided to obtain a normal production data set and an abnormal production data set; after data preprocessing the normal production data set and the abnormal production data set, an early warning model training data set is obtained; the data preprocessing process includes removing abnormal data; The initialized pipeline recognition warning model is trained and optimized according to the warning model training data set.

[0083] The process of controlling the production line through the early warning coefficient of the production line is: The early warning coefficient of the assembly line is judged. When it is judged that there is an abnormality, an early warning message is sent to the production line; when it is judged that there is no abnormality, sampling and testing continue.

[0084] By means of the intelligent inspection and control method for the quality of the socks production line described in the present invention, the socks knitting factory G accurately identifies the quality problems of socks stitching in its socks production line and performs production control according to the identification results.

[0085] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for intelligent inspection and control of the quality of a socks production line, characterized in that: include: S10. Sampling the socks products that have been sewn at the toe on the socks production line to obtain the first sock sample; S20. Identify the model data of the first sock sample, and determine the corresponding first mold and second mold according to the model data; Putting the first sock sample on the first mold, and photographing the seam to obtain a first seam image; Turning over the first sock sample so that the inner surface of the first sock sample faces outward, then putting the first sock sample on the second mold, and photographing the seam to obtain a second seam image; S30. Constructing a suture abnormality recognition model to recognize the first sutured image and the second sutured image; the suture abnormality recognition model includes a structure abnormality recognition unit, a position abnormality recognition unit and a region abnormality recognition unit, which respectively recognize the first sutured image and the second sutured image to obtain a suture structure abnormality coefficient, a suture position abnormality coefficient and a suture region abnormality coefficient; S40. Calculate the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient, and judge the suture quality according to the suture quality abnormality coefficient; S50. Obtaining the stitching quality abnormality coefficient of the first sock sample, and sorting the first sock sample according to the stitching time sequence to obtain a stitching abnormality time series; S60. Constructing an assembly line warning recognition model, analyzing the abnormal stitching time series, and obtaining an assembly line warning coefficient; Control the production line according to the assembly line early warning coefficient.

2. According to claim 1, the method for intelligent inspection and control of the quality of the socks production line is characterized by: The process of determining the first mold and the second mold according to the model data of the first sock sample is as follows: The model data of the first sock sample includes type data and size data, the type data includes left foot type and right foot type; the size data includes length data and width data; Obtaining size data of a first sock sample, and obtaining a left foot mold and a right foot mold of corresponding sizes; Acquire type data of a first sock sample, and use a mold that is consistent with the type of the first sock sample as a first mold; A mold that does not match the type of sock sample from the first is used as the second mold.

3. According to claim 1, the method for intelligent inspection and control of the quality of the socks production line is characterized by: During the shooting process of the first stitched image and the second stitched image, the camera device is located directly above the mold to shoot the stitching position; and the relative position between the camera device and the mold is fixed, so that in the first stitched image and the second stitched image, the positions of the first mold and the second mold are symmetrical.

4. The method for intelligent inspection and control of quality of socks production line according to claim 1 is characterized by: The suture abnormality recognition model includes a structural abnormality recognition unit, a position abnormality recognition unit and a regional abnormality recognition unit; The structural abnormality recognition unit recognizes the first stitched image to obtain a first structural abnormality coefficient; recognizes the second stitched image to obtain a second structural abnormality coefficient; Obtaining a suture structure abnormality coefficient by weighting the first structure abnormality coefficient and the second structure abnormality coefficient; The position abnormality recognition unit recognizes the first stitched image to obtain a first position abnormality coefficient; recognizes the second stitched image to obtain a second position abnormality coefficient; and weights the first position abnormality coefficient and the second position abnormality coefficient to obtain a stitching position abnormality coefficient; The regional anomaly identification unit identifies the first stitched image to obtain a first regional anomaly coefficient; identifies the second stitched image to obtain a second regional anomaly coefficient; and obtains a stitched regional anomaly coefficient by weighting the first regional anomaly coefficient and the second regional anomaly coefficient.

5. According to claim 4, the method for intelligent inspection and control of the quality of the socks production line is characterized by: The recognition process of the structural abnormality recognition unit on the first stitched image is as follows: The stitching area of ​​the first stitching image is identified according to the texture features and the color features to obtain an initial stitching area; and then the yarn structure at the edge of the area where the initial stitching area is located is calibrated to obtain a target stitching area; The yarn distribution structure in the target stitching area is obtained, and the yarn distribution structure is compared and identified with the corresponding standard yarn distribution structure to obtain first structure difference data; and a first structure abnormality coefficient is obtained according to the first structure difference data.

6. The method for intelligent inspection and control of the quality of the socks production line according to claim 4 is characterized by: The process of the position abnormality recognition unit recognizing the first stitched image is as follows: Identify a target stitching area of ​​the first stitching image; construct a coordinate system on the first stitching image, and obtain position distribution data of the target stitching area according to the coordinate system; The position distribution data of the target suture area is compared and identified with the corresponding standard position distribution data to obtain first position difference data; and a first position abnormality coefficient is obtained according to the first position difference data.

7. The method for intelligent inspection and control of the quality of the socks production line according to claim 4 is characterized by: The process of the regional abnormality identification unit identifying the first stitched image is as follows: Acquire a target stitching area of ​​the first stitching image, acquire a sock area around the target stitching area according to a distance threshold, and obtain a stitching adjacent area; The stitching adjacent area is identified according to the yarn structure data and color distribution data of the standard stitching adjacent area to obtain the first area difference data, and the first area abnormality coefficient is obtained according to the first area difference coefficient.

8. The method for intelligent inspection and control of quality of socks production line according to claim 1 is characterized by: The formula for calculating the suture quality abnormality coefficient according to the suture structure abnormality coefficient, the suture position abnormality coefficient and the suture area abnormality coefficient is: ； in, It represents the abnormal coefficient of suture quality; represents the structural weight; represents the coefficient of abnormality of suture structure; represents the position weight; represents the coefficient of abnormal suture position; represents the regional weight; Represents the abnormal coefficient of the suture area.

9. The method for intelligent inspection and control of quality of socks production line according to claim 1 is characterized by: The process of judging suture quality according to suture quality abnormality coefficient is as follows: ； in, Indicates the suture quality. When the suture quality is equal to 1, it means the suture quality is qualified; when the suture quality is equal to 0, it means the suture quality is unqualified; It represents the abnormal coefficient of suture quality; Indicates the abnormal suture quality threshold.

10. The method for intelligent inspection and control of quality of sock production line according to claim 1 is characterized by: The pipeline identification and warning model is constructed based on the LSTM model, and its training process is as follows: Data collection and processing are performed on the sock production line to obtain a historical production data set; the historical production data set includes a time series of historical stitching quality abnormality coefficients and a time period when abnormalities occurred in processing equipment; According to the time period when the processing equipment is abnormal and the time point of the historical stitching quality abnormal coefficient, the historical production data set is divided into a normal production data set and an abnormal production data set; After data preprocessing of the normal production data set and the abnormal production data set, the early warning model training data set is obtained; The initialized pipeline recognition warning model is trained and optimized according to the warning model training data set.

Images