A method for detecting abnormality in fabric texture

Abstract

The present invention relates to a fabric texture anomaly detection method, which comprises the following process steps: 1) collecting fabric image information; 2) decomposing fabric tissue texture feature information; 3) extracting texture feature vectors; 4) judging and classifying texture anomalies; 5) issuing corresponding control instructions to a lower computer according to texture anomaly information. The fabric texture anomaly detection method of the present invention extracts texture feature information based on image scale spatial distribution characteristics, thereby realizing fabric texture anomaly classification and recognition; it comprehensively considers the factors of texture anomalies, improves the detection rate, and makes the detection result accurate.

Description

[Technical field]

[0001] The invention relates to a fabric detection method, in particular to a fabric texture anomaly detection method, and belongs to the technical field of textile detection. [Background technology]

[0002] Textile fabrics are divided into two categories: woven fabrics and knitted fabrics. Woven fabrics are formed by interweaving warp and weft yarns in two directions to form various structures (plain, twill, satin); while the structure of knitted fabrics is more complex and diverse, and texture abnormalities are inevitable during the weaving process, affecting the quality of the fabric.

[0003] Fabric texture anomaly detection is the key and difficulty of textile quality inspection. Fabric texture anomalies include missed stitches, spot stitches, sparse density, cross-roads, broken yarns, etc. However, the existing fabric defect detection methods lack detection methods for fabric texture anomalies. The reason is that there is a lack of understanding and comprehensive consideration of textile technologies such as weaving mechanism, fabric structure, and weaving control, resulting in a low detection rate. The detection results are difficult to feed back to the production site to improve fabric quality.

[0004] Therefore, in order to solve the above technical problems, it is necessary to provide an innovative fabric texture anomaly detection method to overcome the above defects in the prior art. [Summary of the invention]

[0005] The object of the present invention is to provide a method for detecting abnormality in fabric texture which has simple process, comprehensive consideration of abnormal characteristics and high detection result accuracy.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a method for detecting abnormality in fabric texture, which comprises the following process steps:

[0007] 1), collect fabric image information;

[0008] 2), decomposing fabric texture feature information;

[0009] 3), extracting texture feature vector;

[0010] 4), determine and classify texture anomalies;

[0011] 5) Send corresponding control instructions to the lower computer based on the texture abnormality information.

[0012] The fabric texture anomaly detection method of the present invention further comprises: the image information acquisition of step 1) comprises the following steps:

[0013] 1-1), a number of high-speed industrial cameras are arranged at equal intervals on the width of the fabric;

[0014] 1-2), rear light source lighting method, highlighting the abnormal characteristics of fabric texture;

[0015] 1-3), high bandwidth image acquisition card acquires image information.

[0016] The fabric texture anomaly detection method of the present invention is further as follows: in step 1-2), the fabric texture anomaly features include holes, missed stitches, flower stitches, hanging stitches, sparse and dense stitches, horizontal stitches, broken yarns, and triangular eyes.

[0017] The fabric texture anomaly detection method of the present invention further comprises: the step 2) is specifically:

[0018] 2-1), the fabric image is expanded through the window empirical mode to obtain the time domain and frequency domain texture feature signals with periodic fluctuations. The fabric image is decomposed and expanded as follows:

[0019] 2-2), solve the local maximum and minimum of the signal, and its modal components are expressed as follows:

[0020] h1(t)=x(t)-m1(t);

[0021] 2-3), use cubic spline curve to enclose the extreme points;

[0022] 2-4), find the average value m1(t) of the extreme point encirclement line;

[0023] 2-5), based on the original signal, the average signal is screened. The specific method is as follows:

[0024] r1(t)=x(t)-c1(t)

[0025] 2-6), the screening signal is compared with the standard deviation and controlled as follows:

[0026] When the SD value is between 0.2 and 0.3, the screening condition is met;

[0027] 2-7), the new signal is obtained by subtracting the original signal from the screened signal, which is:

[0028] The number of extreme values ​​of r is less than 2.

[0029] The fabric texture anomaly detection method of the present invention further comprises: the step 3) is specifically:

[0030] 3-1), the fabric texture features are divided into loop and string relationships;

[0031] 3-2), obtain the image grayscale distribution uniformity eigenvector and complexity eigenvector through the grayscale co-occurrence matrix; a pair of M×N images with grayscale S, set the direction parameters a and b, which can take values ​​0 or 1; the grayscale value of any point (x, y) in the image is i, then the grayscale value of the point (x+a, y+b) is j, and (i, j) is the grayscale co-occurrence point;

[0032] The uniformity of image grayscale distribution is described by equation:

[0033] The complexity of image grayscale distribution is described by equation:

[0034] 3-3), the texture change trajectory feature vector is expressed by directionality, and the formula is:

[0035] where n p is the number of peaks in the probability histogram, P is the peak point, φ p is the center point of the wave front, w p is the distance between the troughs on both sides of the peak, H D (φ) is the marginal probability;

[0036] 3-3), multiple groups of feature vectors are fused, and the fused feature CWI is expressed as: CWI=A+E+F.

[0037] The fabric texture anomaly detection method of the present invention further comprises: the step 4) is specifically:

[0038] 4-1), build an evolutionary neural network model;

[0039] 4-2), perform global optimization in the evolutionary neural network to obtain the initialization population of the optimal Harris Hawk model. The position vector of the Harris Hawk model is expressed as follows:

[0040] Where X prey (t) is the optimal position, X(t) represents the current position, and X rand (t) is a random position, r i is a random number between [0,1], UB and LB are the upper and lower limits of the variable, and X m (t) is the individual average position, expressed as:

[0041] 4-3), optimize the weight value (w) and bias value (b) in the optimal Harris Hawk model to obtain the optimal regularized extreme learning machine, expressed as: h i (x) = g(w i , b i , x) = g(w i x+b i), where g(x) is the Sigmoid activation function;

[0042] 4-4), texture anomaly judgment and classification detection are performed based on an improved regularized extreme learning machine.

[0043] The fabric texture anomaly detection method of the present invention may also be: the specific method of building the evolutionary neural network model is:

[0044] Randomly generate an initial population of size M×N, whose individuals are represented by: X i (0) = [x i，1 (0), x i，2 (0), x i，3 (0), ..., x i，N (0)], (i = 1, 2, 3, ..., M), where X i，j (0) indicates the 0th generation population, with an initial population size of 50;

[0045] The DE algorithm is used to achieve individual variation, expressed as: V i (g) = X best (g)+F·[X p1 (g)-X p2 (g)]+F·[X p3 (g)-X p4 (g)], where X best (g) represents the best individual, X pi (g) represents a random individual, and the scaling factor F value is 1;

[0046] Crossover mutation is expressed as: Where cr∈[0,1], the crossover probability cr value is selected as 0.8;

[0047] The optimal representation is:

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The fabric texture anomaly detection method of the present invention extracts texture feature information based on the image scale space distribution characteristics, thereby realizing the classification and identification of fabric texture anomalies.

[0050] 2. The fabric texture anomaly detection method of the present invention comprehensively considers factors of texture anomaly, thereby improving the detection rate and making the detection result accurate.

[0051] 3. The fabric texture anomaly detection method of the present invention can send corresponding control instructions to production site equipment based on texture anomaly information, thereby improving the production quality of the fabric.

Brief Description of the Drawings

[0052] Figure 1 It is an overall flow chart of the fabric texture anomaly detection method of the present invention.

[0053] Figure 2 It is a flow chart of the decomposition of fabric texture feature information in step 2) of the present invention.

[0054] Figure 3 This is a flow chart of fabric texture feature vector extraction in step 3) of the present invention.

[0055] Figure 4 It is a flow chart of texture anomaly judgment and classification detection in step 4) of the present invention. [Specific implementation method]

[0056] Please refer to the instruction manual Figure 1 To Attachment Figure 4 As shown, the present invention relates to a method for detecting abnormality in fabric texture, which comprises the following process steps:

[0057] 1) Collecting fabric image information; which specifically includes the following steps:

[0058] 1-1), a number of high-speed industrial cameras are arranged at equal intervals on the width of the fabric;

[0059] 1-2), rear light source lighting method, highlighting the abnormal characteristics of fabric texture;

[0060] 1-3), high bandwidth image acquisition card acquires image information.

[0061] 2) Decompose the fabric texture feature information; please refer to the appendix of the manual. Figure 2 As shown, it specifically includes the following steps:

[0062] 2-1), the fabric image is expanded through the window empirical mode to obtain the time domain and frequency domain texture feature signals with periodic fluctuations. The fabric image is decomposed and expanded as follows: The abnormal fabric texture features include holes, missed stitches, flower stitches, hanging stitches, sparse and dense lines, horizontal lines, broken yarns, and triangular eyes.

[0063] 2-2), solve the local maximum and minimum of the signal, and its modal components are expressed as follows:

[0064] h1(t)=x(t)-m1(t).

[0065] 2-3), use cubic spline curve to enclose the extreme points.

[0066] 2-4), find the average value m1(t) of the extreme point enclosing line.

[0067] 2-5), based on the original signal, the average signal is screened. The specific method is as follows:

[0068] r1(t)=x(t)-c1(t).

[0069] 2-6), the screening signal is compared with the standard deviation and controlled as follows:

[0070] When the SD value is between 0.2 and 0.3, the screening condition is met.

[0071] 2-7), the new signal is obtained by subtracting the original signal from the screened signal, which is:

[0072] The number of extreme values ​​of r is less than 2.

[0073] 3) Extract texture feature vector; please refer to the appendix of the manual Figure 3 As shown, it specifically includes the following steps:

[0074] 3-1), the fabric texture features are divided into loops (texture primitives) and string relationships (spatial dependencies);

[0075] 3-2), obtain the image grayscale distribution uniformity eigenvector and complexity eigenvector through the grayscale co-occurrence matrix; a pair of M×N images with grayscale S, set the direction parameters a and b, which can take values ​​0 or 1; the grayscale value of any point (x, y) in the image is i, then the grayscale value of the point (x+a, y+b) is j, and (i, j) is the grayscale co-occurrence point;

[0076] The uniformity of image grayscale distribution is described by equation:

[0077] The complexity of image grayscale distribution is described by equation:

[0078] 3-3), the texture change trajectory feature vector is expressed by directionality, and the formula is:

[0079] where n p is the number of peaks in the probability histogram, P is the peak point, φ p is the center point of the wave front, w p is the distance between the troughs on both sides of the peak, H D (φ) is the marginal probability;

[0080] 3-3), multiple groups of feature vectors are fused, and the fused feature CWI is expressed as: CWI=A+E+F.

[0081] 4) Determine and classify texture anomalies; please refer to the appendix of the manual. Figure 4 As shown, it specifically includes the following steps:

[0082] 4-1), build an evolutionary neural network model, the specific method is:

[0083] Randomly generate an initial population of size M×N, whose individuals are represented by: X i (0) = [x i，1 (0), x i，2 (0), x i，3 (0),......x i，N (0)], (i = 1, 2, 3, ..., M), where X i，j (0) indicates the 0th generation population, with an initial population size of 50;

[0084] The DE algorithm is used to achieve individual variation, expressed as: V i (g) = X best (g)+F·[X p1 (g)-X p2 (g)]+F·[X p3 (g)-X p4 (g)], where X best (g) represents the best individual, X pi (g) represents a random individual, and the scaling factor F value is 1;

[0085] Crossover mutation is expressed as: Where cr∈[0,1], the crossover probability cr value is selected as 0.8;

[0086] The optimal representation is:

[0087] 4-2), perform global optimization in the evolutionary neural network to obtain the initialization population of the optimal Harris Hawk model. The position vector of the Harris Hawk model is expressed as follows:

[0088] Where X pery (t) is the optimal position, X(t) represents the current position, and X rand (t) is a random position, r i is a random number between [0,1], UB and LB are the upper and lower limits of the variable, and X m (t) is the individual average position, expressed as:

[0089] 4-3), optimize the weight value (w) and bias value (b) in the optimal Harris Hawk model to obtain the optimal regularized extreme learning machine, expressed as: h i (x) = g(w i , b i , x) = g(w i x+b i), where g(x) is the Sigmoid activation function;

[0090] 4-4), texture anomaly judgment and classification detection are performed based on an improved regularized extreme learning machine.

[0091] 5) Send corresponding control instructions to the lower computer based on the texture abnormality information.

[0092] The fabric texture anomaly detection method of the present invention comprehensively considers the factors of texture anomaly, improves the detection rate, and makes the detection result accurate; at the same time, it can send corresponding control instructions to production site equipment according to texture anomaly information, thereby improving the production quality of the fabric.

[0093] The above specific implementation methods are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for detecting abnormality in fabric texture, characterized in that: The process steps include: 1), collect fabric image information; 2), decomposing fabric texture feature information; Specifically: 2-1), the fabric image is expanded through the window empirical mode to obtain the time domain and frequency domain texture feature signals with periodic fluctuations. The fabric image is decomposed and expanded as follows: 2-2), solve the local maximum and minimum of the signal, and its modal components are expressed as follows: h1(t)=x(t)-m1(t); 2-3), use cubic spline curve to enclose the extreme points; 2-4), find the average value m1(t) of the extreme point encirclement line; 2-5), based on the original signal, the average signal is screened. The specific method is as follows: r1(t)=x(t)-c1(t) 2-6), the screening signal is compared with the standard deviation and controlled as follows: When the SD value is between 0.2 and 0.3, the screening condition is met; 2-7), the new signal is obtained by subtracting the original signal from the screened signal, which is: The number of extreme values ​​of r is less than 2; 3), extracting texture feature vector; Specifically: 3-1), the fabric texture features are divided into loop and string relationships; 3-2), obtain the image grayscale distribution uniformity eigenvector and complexity eigenvector through the grayscale co-occurrence matrix; a pair of M×N images with grayscale S, set the direction parameters a and b, which can take values ​​0 or 1; the grayscale value of any point (x, y) in the image is i, then the grayscale value of the point (x+a, y+b) is j, and (i, j) is the grayscale co-occurrence point; The uniformity of image grayscale distribution is described by equation: The complexity of image grayscale distribution is described by equation: 3-3), the texture change trajectory feature vector is expressed by directionality, and the formula is: Where n p is the number of peaks in the probability histogram, P is the peak point, φ p is the center point of the wave front, w p is the distance between the troughs on both sides of the peak, H D (φ) is the marginal probability; 3-3), multiple groups of feature vectors are fused, and the fused feature CWI is expressed as: CWI = A + E + F; 4), determine and classify texture anomalies; Specifically: 4-1), build an evolutionary neural network model; 4-2), perform global optimization in the evolutionary neural network to obtain the initialization population of the optimal Harris Hawk model. The position vector of the Harris Hawk model is expressed as follows: Where X prey (t) is the optimal position, X(t) represents the current position, and X rand (t) is a random position, r i is a random number between [0,1], UB and LB are the upper and lower limits of the variable, and X m (t) is the individual average position, expressed as: 4-3), optimize the weight value (w) and bias value (b) in the optimal Harris Hawk model to obtain the optimal regularized extreme learning machine, expressed as: h i (x) = g(w i , b i , x) = g(w i x+b i ), where g(x) is the Sigmoid activation function; 4-4), texture anomaly judgment and classification detection based on improved regularized extreme learning machine; The specific method of building the evolutionary neural network model in step 4-1) is as follows: Randomly generate an initial population of size M×N, whose individuals are represented by: X i (0) = [x i，1 (0), x i，2 (0), x i，3 (0), ..., x i，N (0)], (i = 1, 2, 3, ..., M), where X i，j (0) indicates the 0th generation population, with an initial population size of 50; The DE algorithm is used to achieve individual variation, which is expressed as: V i (g) = X best (g)+F·[X p1 (g)-X p2 (g)]+F·[X p3 (g)-X p4 (g)], where X best (g) represents the best individual, X pi (g) represents a random individual, and the scaling factor F value is 1; Crossover mutation is expressed as: Where cr∈[0,1], the crossover probability cr value is selected as 0.8; The optimal representation is: 5) Send corresponding control instructions to the lower computer based on the texture abnormality information.

2. The fabric texture anomaly detection method according to claim 1, characterized in that: The image information collection of step 1) includes the following steps: 1-1), a number of high-speed industrial cameras are arranged at equal intervals on the width of the fabric; 1-2), rear light source lighting method, highlighting the abnormal characteristics of fabric texture; 1-3), high bandwidth image acquisition card acquires image information.

3. The fabric texture anomaly detection method according to claim 1, characterized in that: In step 1-2), the abnormal fabric texture features include holes, missed stitches, flower stitches, hanging stitches, sparse and dense stitches, horizontal stitches, broken yarns, and triangular eyes.

Images