Intelligent control method and system for automatic textile equipment

Abstract

The invention discloses an intelligent control method of automatic spinning equipment, which comprises the following steps: step 1, obtaining a spinning pattern to be spun, and performing feature point extraction on the spinning pattern to obtain a feature point set; 2, discrete cosine transform is carried out on the textile pattern, the textile pattern is transformed from a spatial domain to a frequency domain, and a frequency domain coefficient matrix of the textile pattern is obtained; 3, generating a mark matrix corresponding to the frequency domain coefficient matrix, wherein each element in the mark matrix is in one-to-one correspondence with each element in the frequency domain coefficient matrix; according to the technical scheme, the actual textile pattern is monitored through the real-time image sensing technology, feature point comparison is conducted on the actual textile pattern and the original design pattern, and the dynamic offset between the actual position of the stitch (or the knitting needle) and the expected position is accurately calculated.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of image processing, in particular to an intelligent control method and system of an automatic textile equipment. BACKGROUND

[0002] Intelligent textile technology can automatically generate a textile pattern code through programming and drive a textile equipment to control the movement of cloth to hook the required pattern on the cloth with different colored threads.

[0003] This process usually relies on a transport mechanism to move the cloth under the stitches. However, in practice, there is a key problem with this textile method: although the transport mechanism is driven by a motor and the speed control is accurate, the movement of the cloth mainly relies on friction. This results in uneven speed of the cloth during movement. Therefore, the actual landing position of the stitches deviates from the expected position, ultimately reducing the accuracy of the textile pattern. SUMMARY

[0004] The summary part of the present application is used to introduce the concept in a brief form, which will be described in detail in the specific embodiment part. The summary part of the present application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0005] As a first aspect of the present application, in order to solve the technical problems mentioned in the background part, some embodiments of the present application provide an intelligent control method of an automatic textile equipment, comprising the following steps:

[0006] Step 1: obtaining a textile pattern to be woven, extracting feature points from the textile pattern to obtain a feature point set;

[0007] Step 2: performing discrete cosine transform on the textile pattern to transform the textile pattern from spatial domain to frequency domain and obtaining a frequency domain coefficient matrix of the textile pattern;

[0008] Step 3: generating a flag matrix corresponding to the frequency domain coefficient matrix, each element in the flag matrix and the frequency domain coefficient matrix corresponding one by one;

[0009] Step 4: labeling the frequency domain coefficient matrix with the flag matrix to obtain a labeled frequency domain matrix;

[0010] Step 5: real-time detecting a monitoring pattern collected in the weaving process, randomly extracting a plurality of feature points from the monitoring pattern, obtaining the corresponding positions of the feature points in the flag matrix, and judging the offset between the current needle position and the actual position according to the offset of the corresponding positions;

[0011] Step 6: adjusting the weaving scheme according to the offset between the current needle position and the actual position.

[0012] Further, S01: extracting feature points in the textile pattern based on the SIFT algorithm; specifically:

[0013] For the textile pattern I(x, y), the scale space L(x, y, sigma) is defined as: L(x, y, sigma) = G(x, y, sigma) * I(x, y);

[0014] Wherein, G(x, y, sigma) is a Gaussian function;

[0015]

[0016] S02: for each feature point, determine one or more main directions based on the gradient direction histogram within the neighborhood of the key point;

[0017] The gradient amplitude m and direction theta of the feature point are respectively:

[0018]

[0019] Wherein, x and y are the horizontal and vertical coordinates of the pixel point, sigma is the standard deviation of the Gaussian function, m(x, y) represents the gradient amplitude at position (x, y), and L refers to the image I after Gaussian blur at different scales.

[0020] Further, the feature points are extracted based on the neural network model in step 1, and the neural network model learns the corresponding relationship between the fusion features and the labels, and automatically labels each region.

[0021] Further, step 1 includes the following steps:

[0022] Step 11: divide the textile pattern into multiple regions divided by multiple uniform rectangles.

[0023] Step 12: the iteration unit randomly selects a region i, and selects a segmentation label according to the state information of the region i, and the state information includes the pixel information of the region, the pixel information of the adjacent region.

[0024] Step 13: calculate the reward value r of the segmentation label given to region i.

[0025] Further, the state information includes S wi , S ei , S li ;

[0026] For region i, the pixel information of region i is S wi , the pixel information of the remaining regions adjacent to region i is S ei ; and the label distribution of the remaining regions adjacent to region i is S li ;

[0027] S ei = {E 1i , E 2i , E 3i …E ei …}, wherein E ei is the pixel information of the e-th region adjacent to region i;

[0028] S li = {L 1i , L 2i , L 3i …L li …}, L li is a distribution vector, wherein the l-th element represents the label of the l-th region adjacent to region i.

[0029] In the technical solution provided in the present application, when updating the label of a region, the pixel information of the region and the labels of adjacent regions are used for updating, so that relatively less annotation data is required, the accuracy of the region in annotation is increased, and the relationship between information can be better found in the iteration process.

[0030] When training the model, some annotation data is required, and these annotation data are not all correct data to a certain extent, but also include a lot of incorrect data. These incorrect data are actually difficult to distinguish, and the use in the training of the labeling module will cause the labeling module to learn some incorrect information. Therefore, the present application provides the following technical solution:

[0031] Further, the cycle unit selects regions different from the labels of surrounding regions from all regions after all regions are assigned labels, and inputs the regions to the iteration unit for reiteration.

[0032] In the technical solution provided in the present application, regions possibly having labeling errors are selected based on the distribution of labels between regions, and then the regions are guided for further reinforcement learning, so that the influence of incorrect data on prediction accuracy is reduced.

[0033] Further, the model is affected by the reward value in the iteration process, and the design of the reward value will affect the convergence rate of the model and increase the training cost of the model. Therefore, the present application provides the following technical solution:

[0034] Further, the reward value is r,

[0035] is a Gaussian function, and min(d i ) represents the distance from region i to the true segmentation boundary G iThe shortest geodesic distance in all points, the geodesic distance refers to the length of the shortest path from one point to another, G i This refers to the correct segmentation result of region i, the minimum geodesic distance of region i to the true value segmentation boundary G i The label of region i, and α and β are a pair of proportional parameters.

[0036] Further, the training process of the neural network model is as follows:

[0037] S1: initialize the parameters of the neural network model RLSegNet. Such as setting the number of iterations, step size, α, β.

[0038] S2: for each training image, segment it into regions, and initialize the label of each region as empty or a random label.

[0039] S3: start iteration:

[0040] For each region i, calculate its fusion feature S wi .

[0041] Calculate the average fusion feature S ei of the adjacent regions of region i.

[0042] Calculate the label distribution S li of the adjacent regions of region i.

[0043] Input S wi , S ei , S li as state information into the RLSegNet network, and the network outputs the predicted label of region i.

[0044] S4: for each region i, calculate the minimum geodesic distance min(d i ) according to its predicted label and the true value segmentation boundary G i .

[0045] Calculate the reward value r t according to the Gaussian function, where α = 1, β = 1, and the cumulative reward value R;

[0046] S5: update the parameters of the RLSegNet network using the backpropagation algorithm according to the cumulative reward value R and the result of the reward calculation unit.

[0047] Further, the algorithm for obtaining the frequency domain coefficient matrix of the textile pattern is as follows:

[0048]

[0049] where u and v are the horizontal and vertical coordinates in the frequency coefficient matrix, C(u, v) represents the frequency domain information at the image (x, y), M and N are the row and column numbers of the pixel points in the textile pattern, and a(u) and a(v) are normalization factors.

[0050] Further, step 5 comprises the following steps:

[0051] Step 51: using a predefined mapping function Φ(x k ,y k )

[0052]

[0053] W*H: original textile pattern resolution, M*N: DCT frequency domain matrix dimension, (x k ,y k ) represents the feature point space coordinate component.

[0054] Step 52: DCT transformation on the real-time collected monitoring pattern:

[0055] C′ monitor = DCT(I monitor ), C′ monitor represents the DCT coefficient matrix of the monitoring pattern;

[0056] Obtain the coefficient at a specific frequency domain position:

[0057] c′ monitor = C′ monitor [u k ,v k ];

[0058] [u k ,v k ] represents the frequency domain coordinates, and c′ monitor represents the specific frequency domain coefficient value;

[0059] Step 53: XOR positioning using the pre-stored flag matrix:

[0060]

[0061] K represents the flag matrix, represents the XOR operator, and c original represents the converted frequency domain coefficient;

[0062] Step 54: calculate the expected position through local inverse DCT:

[0063]

[0064] The expected location value of the k-th feature point is represented by α(u) and α(v), which represent the DCT normalization coefficients. original This represents the calibrated frequency domain coefficient values.

[0065] Step 55: Calculate the offset based on the current expected position and the actual position.

[0066]

[0067] Δp k Δx represents the 3D offset of the k-th feature point. k Indicates the X-axis position offset, Δy k Indicates the Y-axis position offset, Δθ k Indicates the rotation angle offset;

[0068] Step 6 includes the following steps:

[0069] Step 61: Establish a control parameter correction model;

[0070]

[0071] v new The linear speed of the fabric conveyor mechanism after correction;

[0072] ω new The needle mechanism corrects the angular velocity;

[0073] K p : Proportional / derivative gain matrix of PID controller;

[0074] Step 62: Put v new ω new The commands are converted into motor control instructions and sent to the actuator via real-time Ethernet, with parameter updates completed before the start of the next pin cycle.

[0075] The beneficial effects of this application are as follows:

[0076] Significantly improves the accuracy of textile patterns: Real-time image sensing technology monitors the actual textile pattern and compares its feature points with the original design pattern to accurately calculate the dynamic offset between the actual and expected positions of the needles (or knitting needles). Based on this offset, the textile scheme (such as fabric feed speed or needle trajectory) is adjusted in real time, effectively compensating for positioning errors caused by uneven fabric movement (such as friction slippage), thereby greatly improving the geometric accuracy and detail reproduction of the final textile pattern.

[0077] Enhance system anti-interference ability and stability: This method does not rely on the ideal uniform motion assumption of the transportation mechanism itself. By actively sensing and compensating for the inevitable speed fluctuations and position drifts during fabric movement, the stability and reliability of intelligent textile equipment in actual working conditions (with disturbances such as friction and tension changes) are significantly improved.

[0078] Implement closed-loop intelligent control: Image sensing, feature matching, offset calculation, and actuator adjustment are closely integrated to form a real-time closed-loop feedback control system. The system can "perceive-judge-adjust", making the textile process self-adaptive, reducing the over-reliance on mechanical transmission precision.

[0079] Optimize production quality and efficiency: High-precision pattern weaving reduces the rate of defective and waste products caused by misalignment. At the same time, real-time adjustment avoids the need for downtime correction due to excessive cumulative errors, helping to improve production efficiency and yield, and reduce production costs.

[0080] Provide reliable offset detection mechanism: Use Discrete Cosine Transform (DCT) and unique logo matrix encryption method to process original pattern information, providing a structured, more anti-interference data basis for subsequent real-time monitoring of pattern feature point extraction and matching, which helps to accurately and efficiently identify feature points and calculate offset in complex textile texture background. BRIEF DESCRIPTION OF DRAWINGS

[0081] The accompanying drawings, which form a part of this application, are intended to provide further understanding of the application and are incorporated herein in their entirety. The schematic embodiment drawings of the present application and their descriptions are used to explain the present application and do not constitute an undue limitation on the present application.

[0082] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the elements and elements are not necessarily drawn to scale.

[0083] In the drawings:

[0084] Figure 1 Flowchart of the intelligent control method for automated textile equipment. DETAILED DESCRIPTION

[0085] Embodiments of the present application will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present application are shown in the drawings, it should be understood that the present application can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present application. It should be understood that the drawings and embodiments of the present application are only for illustrative purposes, and are not intended to limit the scope of protection of the present application.

[0086] It should be further noted that only the parts related to the application are shown in the drawings for ease of description. The embodiments in the present application and the features in the embodiments can be combined with each other without conflict.

[0087] The present application will be described in detail below with reference to the drawings and in conjunction with the embodiments.

[0088] The intelligent control method of the automatic textile equipment comprises the following steps:

[0089] Step 1: Obtain a textile pattern to be woven, extract feature points from the textile pattern, and obtain a feature point set.

[0090] In step 1, the feature points in the textile pattern are mainly obtained, which need to have secrecy, invisibility, and anti-transformation ability. That is, after the image is operated such as clarity change and rotation, the feature points will not change. Generally, when extracting image feature points, the SIFT algorithm is used.

[0091] The specific scheme is as follows: the present application uses the following scheme to extract feature points:

[0092] S01: Extract feature points in the textile pattern based on the SIFT algorithm; specifically:

[0093] For the textile pattern I(x, y), the scale space L(x, y, σ) is defined as: L(x, y, σ) = G(x, y, σ) x I(x, y);

[0094] Where G(x, y, σ) is a Gaussian function;

[0095]

[0096] S02: For each feature point, determine one or more main directions based on the gradient direction histogram in the neighborhood of the key point;

[0097] The gradient amplitude m and direction θ of the feature point are respectively:

[0098]

[0099]

[0100] Where x and y are the horizontal and vertical coordinates of the pixel point, respectively, σ is the standard deviation of the Gaussian function, m(x, y) represents the gradient amplitude at position (x, y), and L refers to the image I blurred by the Gaussian function at different scales.

[0101] The above scheme can obtain all feature points in the textile pattern, and then obtain a feature point set. When extracting the feature points, the above scheme needs to apply the SIFT algorithm to the entire image, resulting in low encryption efficiency. Therefore, the neural network model is provided to extract the feature points.

[0102] Specifically, step 1 includes the following steps:

[0103] Step 11: dividing the textile pattern into multiple regions of multiple uniform division rectangles.

[0104] Step 12: the iteration unit randomly selects a region i, and selects a segmentation label according to the state information of the region i, and the state information includes the pixel information of the region and the pixel information of the adjacent region.

[0105] Step 13: calculating the reward value r of the segmentation label given to the region i;

[0106] The reward value is actually an adaptive function, and the greater the reward value r, the higher the operation of the iteration unit this time, and vice versa.

[0107] The loop unit guides the iteration unit to continuously loop until the maximum cumulative reward value R is obtained.

[0108] In the present scheme, the kernel of the iteration unit is the hidden layer of the neural network model to be trained, which is the RLSegNet network. The loop unit is the number of times of controlling the iteration unit during training, and the reward calculation unit is used to judge the effectiveness of the iteration operation this time.

[0109] Specifically, the label in the present scheme is actually the position of the feature point in the region. For example, in the present scheme, the region specification is 10*10, and the label is an identification matrix of whether the pixel point is a feature point. Among them, 0 represents not a feature point, and 1 represents a feature point.

[0110] During training, a large number of samples are needed, and whether the sample is a feature point is determined by the feature extraction algorithm.

[0111] Further, in the iteration unit:

[0112] The state information includes S wi , S ei , and S li .

[0113] For region i, the pixel information of region i is S wi , the pixel information of the remaining regions adjacent to region i is S ei , and the label distribution of the remaining regions adjacent to region i is S li .

[0114] Sei = {E 1i , E 2i , E 3i …E ei …}, wherein E ei is the pixel information of the e-th region adjacent to region i;

[0115] S li = {L 1i , L 2i , L 3i …L li …}, L li is a distribution vector, wherein the l-th element represents the label of the l-th region adjacent to region i;

[0116] The adjacent region generally refers to a region adjacent to the region. In the present scheme, it mainly refers to a region within a ring, that is, a region directly adjacent to the region. In practice, it can also be a region within a two-ring, that is, an indirectly adjacent region.

[0117] Further, the cycle unit filters out the region different from the label of the surrounding region from all the regions after all the regions are assigned with the label, and inputs it to the iteration unit for reiteration.

[0118] Further, the model is affected by the reward value in the iteration process, and the design of the reward value affects the convergence rate of the model, increases the training cost of the model, and therefore the present application provides the following technical scheme:

[0119] Further, the reward value is r t ,

[0120] is a Gaussian function, and min(di) represents the shortest geodesic distance from region i to all points on the true value segmentation boundary G i , wherein the geodesic distance refers to the length of the shortest path from one point to another point, and G i refers to the correct segmentation result of region i, the minimum geodesic distance from region i to the true value segmentation boundary Gi is the label of region i, and a and β are a pair of proportional parameters, and r t represents the index at the t-th iteration, t represents the index of the iteration number, a t represents the action at the t-th iteration, so a t = G i represents that this segmentation result is correct, and vice versa. In the present scheme, a = 1 and β = 1.

[0121] Specifically, the training process is as follows:

[0122] S1: Initialize the parameters of the neural network model RLSegNet. Such as setting the number of iterations, step size, alpha, beta.

[0123] S2: For each training image, divide it into regions and initialize the label of each region as empty or a random label.

[0124] S3: Start iteration:

[0125] For each region i, calculate its fusion feature S wi .

[0126] Calculate the average fusion feature S ei of the adjacent regions of region i.

[0127] Calculate the label distribution S li of the adjacent regions of region i.

[0128] Input S wi , S ei , S li as state information into the RLSegNet network, and the network outputs the predicted label of region i.

[0129] S4: For each region i, calculate the minimum geodesic distance min(d i ) according to its predicted label and ground truth segmentation boundary G i .

[0130] Calculate the reward value r t according to the Gaussian function, where alpha = 1, beta = 1, and the cumulative reward value R.

[0131] Loop unit operation:

[0132] Check if all regions have been assigned labels, filter out regions with labels different from surrounding regions, and re-input these regions into the iteration unit for re-iteration. Repeat the iteration process until the maximum number of iterations is reached or the cumulative reward value R no longer increases significantly.

[0133] S5: According to the cumulative reward value R and the result of the reward calculation unit, update the parameters of the RLSegNet network using the backpropagation algorithm. Until the RLSegNet network converges or reaches the preset training round number: save the trained RLSegNet model parameters.

[0134] In order to increase the convergence speed of the model:

[0135] In this scheme, alpha and beta are dynamically changing, alpha and beta are related to the number of iterations, and the current number of iterations is t, and the total number of iterations T;

[0136]

[0137] where U and k are pre-set control parameters, U = 3, k = 1. In this way, in this scheme, at the beginning of iteration, the growth rate of the reward function is low, that is, both the reward and the punishment are small, so the iteration unit has more selection space when iterating, and can avoid the model from falling into a local optimal solution, while in the second half of the iteration, the growth rate of the reward function will increase, at this time, the reward or punishment received by the wrong selection will be greater, so when the iteration parameter is selected, the more correct operation will be selected as much as possible to increase the model convergence rate.

[0138] At the time of prediction:

[0139] (1) : Load the trained RLSegNet model parameters.

[0140] (2) : Divide the image to be predicted into regions and calculate the fusion features S of each region wi .

[0141] (3) : For each region i, initialize its label as empty or random label, calculate the average fusion features S ei and label distribution S li of the adjacent regions of region i, input S wi , S ei , S li as state information into the RLSegNet network, and the network outputs the predicted label of region i. Repeat the iteration process until all regions are assigned labels or the preset number of iterations is reached. Post-processing is performed on the prediction results, such as smoothing the boundary and removing isolated points, to improve the accuracy of the segmentation results.

[0142] Output the segmentation result: integrate the prediction results into a complete image segmentation map and output.

[0143] Step 2: Perform discrete cosine transform on the textile pattern to transform the textile pattern from spatial domain to frequency domain and obtain the frequency domain coefficient matrix of the textile pattern.

[0144] Specifically:

[0145]

[0146] where u and v are the horizontal and vertical coordinates in the frequency coefficient matrix respectively, C(u, v) represents the frequency domain information at the image (x, y), M and N are the number of rows and columns of the pixel points in the textile pattern, and a(u) and a(v) are normalization factors.

[0147] Step 2 is to transform the textile pattern from spatial domain to frequency domain, so the frequency domain information will be more prominent.

[0148] Step 3: generate a flag matrix corresponding to the frequency domain coefficient matrix, and each element in the flag matrix corresponds to each element in the frequency domain coefficient matrix.

[0149] Step 4: label the frequency domain coefficient matrix with the flag matrix to obtain a labeled frequency domain matrix;

[0150] After obtaining the frequency domain coefficient matrix of the textile pattern, all information of the textile pattern can be represented by using the frequency domain coefficient matrix.

[0151] In the frequency domain, the transform coefficients are subjected to a transform operation.

[0152]

[0153] where C(u, v) is the transformed frequency domain coefficient, K(u, v) is the flag matrix, represents the XOR operation, and C'(u, v) is the encrypted frequency domain coefficient.

[0154] Step 5: real-time detection of the monitoring pattern collected during the textile process, randomly extracting a plurality of feature points from the monitoring pattern, obtaining the corresponding positions of the feature points in the flag matrix, and judging the offset between the current needle position and the actual position according to the offset of the corresponding positions;

[0155] Specifically, the surface image of the cloth under the stitches is captured in real time by a high-frame-rate industrial camera, denoted as a monitoring pattern, and the image acquisition frequency is ≥ 2 times the stitch movement frequency of the textile equipment (satisfying the Shannon sampling theorem).

[0156] The following scheme is used to compare and detect the collected monitoring image and the textile image in step 5;

[0157] Step 51: use a predefined mapping function Φ(x k ,y k )

[0158]

[0159] W*H: original textile pattern resolution, M*N: DCT frequency domain matrix dimension, (x k ,y k ) represents the spatial coordinate components of the feature points.

[0160] Step 52: perform DCT transform on the real-time collected monitoring pattern:

[0161] C′ monitor =DCT(I monitor ), C′ monitor represents the DCT coefficient matrix of the monitoring pattern;

[0162] Obtaining the coefficient of the specific frequency domain position:

[0163] c′ monitor = C′ monitor [u k ,v k ];

[0164] [u k ,v k ] represents the frequency domain coordinates, c′ monitor represents the specific frequency domain coefficient value;

[0165] Step 53: XOR positioning using the pre-stored flag matrix:

[0166]

[0167] K represents the flag matrix, represents the XOR operator, c original represents the converted frequency domain coefficient;

[0168] Step 54: Calculate the expected position by local inverse DCT;

[0169]

[0170] represents the expected position value of the kth feature point, α(u), α(v) represents the DCT normalization coefficient, c original represents the calibrated frequency domain coefficient value.

[0171] Step 55: Calculate the offset based on the current expected position and the actual position.

[0172]

[0173] , Δp k represents the three-dimensional offset of the kth feature point, Δx k represents the X-axis position offset, Δy k represents the Y-axis position offset, Δθ k represents the rotation angle offset.

[0174] Step 6: Adjust the textile scheme according to the current needle position and the actual position offset.

[0175] Step 6 includes the following steps:

[0176] Step 61: Establish a control parameter correction model;

[0177]

[0178] v new : corrected line speed of the fabric conveying mechanism;

[0179] ω new : corrected angular velocity of needle mechanism

[0180] K p : proportional / differential gain matrix of PID controller (calibrated by device dynamics)

[0181] Step 62: convert v new , ω new to motor control commands, update parameters before the next needle cycle starts.

[0182] The above description is merely some preferred embodiments of the present application and a description of the principles of the technology used. Those skilled in the art should understand that the scope of the application involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combinations of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or equivalent features without departing from the above inventive concept. For example, the above features are replaced with each other to form a technical solution with similar functions disclosed in the embodiments of the present application (but not limited to).

Claims

1. An intelligent control method for automated textile equipment, characterized in that: Includes the following steps: Step 1: Obtain the textile pattern to be woven, extract feature points from the textile pattern, and obtain a set of feature points; Step 2: Perform discrete cosine transform on the textile pattern to transform it from the spatial domain to the frequency domain and obtain the frequency domain coefficient matrix of the textile pattern. Step 3: Generate a flag matrix that corresponds to the frequency domain coefficient matrix, with each element in the flag matrix and the frequency domain coefficient matrix having a one-to-one correspondence; Step 4: Mark the frequency domain coefficient matrix with a label matrix to obtain the labeled frequency domain matrix; Step 5: Real-time detection of the monitoring patterns collected during the textile process, random selection of several feature points from the monitoring patterns, acquisition of the corresponding positions of the feature points in the marker matrix, and determination of the offset between the current needle position and the actual position based on the offset of the corresponding position. Step 6: Adjust the weaving scheme according to the offset between the current needle position and the actual position.

2. The intelligent control method for automated textile equipment according to claim 1, characterized in that: S01: Extracting feature points from textile patterns based on the SIFT algorithm; specifically: For a textile pattern I(x, y), the scale space L(x, y, σ) is defined as: L(x, y, σ) = G(x, y, σ) × I(x, y); Where G(x, y, σ) is a Gaussian function; S02: For each feature point, one or more principal directions are determined based on the gradient orientation histogram in the neighborhood of the key point. The gradient magnitude m and direction θ of the feature point are as follows: Where x and y are the x and y coordinates of the pixel, respectively, σ is the standard deviation of the Gaussian function, m(x, y) represents the gradient magnitude at position (x, y), and L refers to the image after Gaussian blurring of image I at different scales.

3. The intelligent control method for automated textile equipment according to claim 2, characterized in that: In step 1, feature points are extracted based on a neural network model. The neural network model learns and integrates the correspondence between features and labels, and automatically labels each region.

4. The intelligent control method for automated textile equipment according to claim 3, characterized in that: Step 1 includes the following steps: Step 11: Divide the textile pattern into multiple regions that are evenly divided into multiple rectangles. Step 12: The iterative unit randomly selects a region i and selects a segmentation label based on the state information of region i. The state information includes the pixel information of the region and the pixel information of adjacent regions. Step 13: Calculate the reward value r for the segmentation label assigned to region i.

5. The intelligent control method for automated textile equipment according to claim 3, characterized in that: Status information includes S wi S ei S li ; For region i, its pixel information is S wi The pixel information of the remaining regions adjacent to region i is S. ei The label distribution of the remaining adjacent regions of region i is S. li ; S ei ={E 1i E 2i E 3i …E ei …}, where E ei It is the pixel information of the e-th region among the remaining regions adjacent to region i; S li ={L 1i L 2i L 3i …L li …}, L li It is a distribution vector, where the l-th element represents the label of the l-th region among the remaining regions adjacent to region i.

6. The intelligent control method for automated textile equipment according to claim 2, characterized in that: After all regions are labeled, the loop unit filters out regions whose labels differ from those of the surrounding regions and inputs them into the iteration unit for re-iteration.

7. The intelligent control method for automated textile equipment according to claim 6, characterized in that: The reward value is r. φ μ,σ 2 Let be a Gaussian function, min(d) i ) represents the region i to the true value partition boundary G. i The shortest geodesic distance among all points, where the geodesic distance is the length of the shortest path from one point to another, G. i This refers to the correct segmentation result for region i, and the distance from region i to the true value segmentation boundary G. i The minimum geodesic distance is the label of region i, and α and β are a pair of scale parameters.

8. The intelligent control method for automated textile equipment according to claim 7, characterized in that: The frequency domain coefficient matrix of the textile pattern is obtained by executing the following algorithm: Where u and v are the horizontal and vertical coordinates in the frequency coefficient matrix, respectively, C(u, v) represents the frequency domain information at (x, y) in the image, M and N are the number of rows and columns of pixels in the textile pattern, respectively, and a(u) and a(v) are normalization factors.

9. The intelligent control method for automated textile equipment according to claim 7, characterized in that: Step 5 includes the following steps: Step 51: Use the predefined mapping function Φ(x) k ,y k ) W*H: Original textile pattern resolution, M*N: DCT frequency domain matrix dimension, (x k ,y k ) represents the spatial coordinate components of the feature point. Step 52: Perform DCT transformation on the real-time acquired monitoring pattern: C′ monitor =DCT(I monitor ), C′ monitor Represents the DCT coefficient matrix of the monitoring pattern; Obtain the coefficients at a specific frequency domain location: c′ monitor =C′ monitor [u k ,v k ]; [u k ,v k ] represents the frequency domain coordinates, c′ monitor Represents a specific frequency domain coefficient value; Step 53: Perform XOR positioning using the pre-stored flag matrix: c original =c′ monitor ⊕K[u k ,v k ]; K represents the flag matrix, ⊕ represents the XOR operator, and c original Represents the frequency domain coefficients after conversion; Step 54: Calculate the expected location using local inverse DCT; The expected location value of the k-th feature point is represented by α(u) and α(v), which represent the DCT normalization coefficients. original This represents the calibrated frequency domain coefficient values. Step 55: Calculate the offset based on the current expected position and the actual position. Δp k Δx represents the 3D offset of the k-th feature point. k Indicates the X-axis position offset, Δy k Indicates the Y-axis position offset, Δθ k Indicates the rotation angle offset; Step 6 includes the following steps: Step 61: Establish a control parameter correction model; v new The linear speed of the fabric conveying mechanism after correction; ω new The needle mechanism corrects the angular velocity; K p : Proportional / derivative gain matrix of PID controller; Step 62: Put v new ω new The commands are converted into motor control instructions and sent to the actuator via real-time Ethernet, with parameter updates completed before the start of the next pin cycle.

10. An intelligent control system for automated textile equipment, characterized in that, The intelligent control method of the automated textile equipment according to any one of claims 1 to 8 is used to fabricate patterns.

Images