CN105931246A - Fabric flaw detection method based on wavelet transformation and genetic algorithm - Google Patents

Abstract

The invention relates to a fabric flaw detection method based on wavelet transform and genetic algorithm, comprising the following steps: preprocessing the collected image with flaws; using wavelet decomposition on the preprocessed image to obtain multi-scale sub-images; The sub-images are fused to obtain the optimal defect edge information; the threshold is calculated by genetic algorithm for the sub-images, and the fused image is thresholded using the threshold; the morphological processing is performed on the thresholded image. The processed cloth blemishes of the invention have accurate segmentation effect, fast segmentation speed, and well retain the original shape of blemishes.

Description

A Fabric Flaw Detection Method Based on Wavelet Transform and Genetic Algorithm

technical field

The invention relates to the technical field of fabric flaw detection, in particular to a fabric flaw detection method based on wavelet transform and genetic algorithm.

Background technique

Fabric defect detection is an indispensable link in the fabric production process, and the manual detection method has been widely used in textile enterprises for a long time. This traditional cloth inspection method is inefficient, the labor intensity of workers is high, and the detection accuracy cannot be guaranteed. It can be seen that the traditional manual inspection method has been difficult to meet the modern management requirements of enterprises. Therefore, the development of a fabric automatic detection equipment has important economic significance for the quality control of textile enterprises and the saving of labor costs.

At present, there are relatively few automatic fabric defect detection systems facing the market, and only a few foreign companies have produced fabric defect detection systems for textile enterprises, such as EVS in Israel and Barco in Switzerland. In the fabric defect detection system, the line array camera is generally used as the image sensor to collect the fabric image. However, due to the fast movement of the cloth and the large size of the cloth during the detection process, the image quality is easily affected by the industrial environment, and the correct extraction of the defective area is called the key point and difficulty in the cloth detection. The detection algorithm of fabric defects is the core part of the automatic detection system, so designing an algorithm with high precision and fast processing speed is the key to realize the online detection of fabric defects.

Contents of the invention

The technical problem to be solved by the present invention is to provide a fabric defect detection method based on wavelet transform and genetic algorithm, so that the cloth defect segmentation effect is accurate and the segmentation speed is fast.

The technical solution adopted by the present invention to solve the technical problems is: provide a method for detecting fabric defects based on wavelet transform and genetic algorithm, comprising the following steps:

(1) Preprocessing the collected images with defects;

(2) Use wavelet decomposition on the preprocessed image to obtain multi-scale sub-images;

(3) Fusion the sub-images to obtain the optimal defect edge information;

(4) Utilize genetic algorithm to calculate threshold value to sub-image, and adopt described threshold value to carry out threshold value segmentation to the image after fusion;

(5) Perform morphological processing on the thresholded image.

The step (1) specifically includes: performing grayscale processing on the collected image with defects first, and then using histogram averaging and median filtering to improve the quality of the original image as a whole.

Described step (2) is specifically: carry out high-pass and low-pass filter and carry out down-sampling to image respectively by row, obtain two self-images of general size of input image, carry out high-pass and low-pass filter to its self-image afterwards by column and Perform down-sampling to obtain 4 sub-images of a quarter size of the input image are A _j+1 , D _j+1 ^H , D _j+1 ^V , D _j+1 ^D ; where A _j+1 is the row The result of low-pass filtering in the direction of row and column, representing the overview signal of the next scale; D _j+1 ^H is the result of low-pass filtering in the row direction and high-pass filtering in the column direction, representing the detail signal in the vertical direction in the horizontal direction D _j+1 ^V is the result of high-pass filtering in the row direction and low-pass filtering in the column direction, representing the overview of the detail signal in the horizontal direction in the vertical direction; D _j+1 ^D is the result of the two directions of row and column The result of high-pass filtering, representing the detail signal in the diagonal direction.

Described step (3) specifically comprises following substep:

(31) Starting from the largest scale, let the scale j be the largest scale J, link the non-zero pixels in the edge image E _j (x, y) with similar modulus and phase angle, and set the chain length threshold, delete less than The short chain of the chain length obtains the single-pixel wide edge image E _j (x, y) at the scale j;

(32) For each edge pixel in the edge image E _j (x, y), search for a 3×3 neighborhood centered on the pixel corresponding to the point at scale j-1, and calculate the phase angle of the model value in the neighborhood All similar points are added to the edge image E _j (x, y) as edge points, and the non-zero pixels with similar modulus and phase angles in the edge image E _j (x, y) are linked, and the chain Long threshold, delete short chains smaller than the chain length, so as to obtain the defect edge image E _j-1 (x, y) under the scale j-1;

(33) If the scale j>1, then return to step (32), if j=1, then the obtained edge image E ₁ (x, y) is the defect edge image after multi-scale fusion;

Wherein, E _j (x, y) is an edge image at each scale j, where j=1, 2, 3, . . . , J.

Described step (4) specifically comprises the following substeps:

(41) Perform histogram calculation on the sub-image to obtain a low-resolution version of the histogram;

(42) generating an initial population based on the histogram of the low-resolution version;

(43) Define the fitness function;

(44) applying the learning strategy to improve the fitness value of the character string;

(45) Compare the optimal character string and the current character string, if the optimal character string is better than the current character string, replace the current character string with the optimal character string; otherwise, perform selection, crossover and mutation operations to generate the next population, And return to step (42);

(46) Project the obtained optimal threshold to the original space to obtain the optimal threshold segmentation effect of the original image.

Beneficial effect

Due to the adoption of the above-mentioned technical scheme, the present invention has the following advantages and positive effects compared with the prior art: the present invention combines wavelet transform and genetic algorithm, and first uses discrete wavelet transform to decompose the preprocessed image, extract The components in different dimensions of the image are obtained to obtain a lower-resolution image, which achieves the purpose of dimensionality reduction. Then, the genetic algorithm is used to process the approximate image histogram after wavelet decomposition to obtain a threshold value, and then threshold value segmentation is performed on the original image to separate the defect part from the fabric background. After wavelet decomposition, the genetic algorithm is used to make the calculation faster than the traditional genetic algorithm. After being processed by the algorithm of the present invention, the segmentation effect of the cloth blemish is accurate, the segmentation speed is fast, and the original shape of the blemish is well preserved.

Description of drawings

Fig. 1 is a schematic diagram of automatic detection equipment for fabric defects;

Fig. 2 is a flow chart of the present invention;

Figure 3 is the original picture of the collected fabric;

Fig. 4 is the sub-image obtained through wavelet decomposition;

Fig. 5 is the image after fusion;

Fig. 6 is a diagram of detection results obtained by thresholding.

detailed description

Below in conjunction with specific embodiment, further illustrate the present invention. It should be understood that these examples are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the teachings of the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms also fall within the scope defined by the appended claims of the present application.

Embodiments of the present invention relate to a fabric defect detection method based on wavelet transform and genetic algorithm, as shown in Figure 1, comprising the following steps:

(1) Preprocessing the collected images with defects;

(2) Use wavelet decomposition on the preprocessed image to obtain multi-scale sub-images;

(3) Fusion the sub-images to obtain the optimal defect edge information;

(4) Utilize genetic algorithm to calculate threshold value to sub-image, and adopt described threshold value to carry out threshold value segmentation to the image after fusion;

(5) Perform morphological processing on the thresholded image.

details as follows:

preprocessing

The collected images are processed in grayscale first, and then histogram mean and median filter are used to improve the quality of the original image as a whole, which is convenient for the next step of processing.

wavelet decomposition

In the fabric, most of the defects have a certain direction, such as broken warp, missing weft, double warp, etc. Some regional defects such as oil stains, holes, etc., are produced in both directions of warp and weft. Irregular texture. After the wavelet decomposition of the image containing defects, the local maximum value of the wavelet coefficient will appear in the detail sub-image in the horizontal direction and the detail sub-image in the vertical direction, and it will appear as a gray singular point in the gray scale. Because the wavelet decomposition has directionality, the horizontal and vertical detail sub-images obtained from the wavelet decomposition can be used to represent the texture information in the horizontal and vertical directions of the fabric. Wavelet transform also has sparsity, and these maximum values are more prominent than those before decomposition, which is more conducive to the detection of defects.

The wavelet decomposition of the image is to perform one-dimensional discrete wavelet transform on the image in the horizontal and vertical directions respectively. First perform high-pass and low-pass filtering on the image by row and down-sampling to obtain two self-images of the same size as the input images, and then perform high-pass and low-pass filtering on the self-image by column and down-sampling to obtain 4 input images The sub-images of the quarter size of the image are A _j+1 , D _j+1 ^H , D _j+1 ^V , D _j+1 ^D . Among them, A _j+1 is the result of low-pass filtering in the row and column directions, representing the overview signal of the next scale; D _j+1 ^H is the result of low-pass filtering in the row direction and high-pass filtering in the column direction, representing the vertical The overview of the detail signal in the horizontal direction; D _j+1 ^V is the result of high-pass filtering in the row direction and low-pass filtering in the column direction, representing the overview of the detail signal in the horizontal direction in the vertical direction; D _j+1 ^D It is the result of high-pass filtering in the row and column directions, and represents the detail signal in the diagonal direction.

sub-image fusion

After wavelet decomposition, the edge information of fabric defects in multiple scales is obtained, and the edge sub-images in each scale are fused with certain rules to obtain the optimal edge information of defects. What used in the present invention is the edge focusing method, and the steps are as follows:

(1) Starting from the largest scale, let j=J, link the non-zero pixels in E _j (x, y) with similar modulus and phase angle, and set the chain length threshold to delete short chains smaller than the chain length , to obtain a single-pixel-wide edge image E _j (x, y) at scale j;

(2) After obtaining the edge image at scale j, for each edge pixel in E _j (x, y), search for a 3×3 neighborhood centered on the corresponding pixel at scale j-1, Add the points with similar modulus and phase angles in the neighborhood as edge points to E _j (x, y), and link the non-zero pixels with similar modulus and phase angles in E _j (x, y), And set the chain length threshold to delete short chains smaller than the chain length, so that the defect edge image E _j-1 (x, y) under the scale j-1;

(3) Let j=j-1, if j>1, go to step (2), if j=1, then the obtained E ₁ (x, y) is the defect edge image after multi-scale fusion.

Wherein, E _j (x, y) is an edge image at each scale j, where j=1, 2, 3, . . . , J.

Threshold Segmentation Using Genetic Algorithm

Simple threshold segmentation is easily affected by noise and irregularities in the target area, and the ideal degree of segmentation is determined by the threshold, so the research on the selection of threshold is very important for image segmentation. However, using the standard genetic algorithm to segment the flawed image has the disadvantages of slow convergence and long calculation time. The segmentation algorithm in this algorithm belongs to a multi-level threshold method. Firstly, the wavelet transform is used to decompose the image from different scales into approximate signal and detail signal, and then the second-order wavelet transform is performed on the obtained approximate signal to decompose it into the next level of approximate signal and detail signal. The genetic algorithm process combined with wavelet transform is as follows:

(1) The dimensions of the image after wavelet transformation have been reduced, and the histogram calculation is performed on the image to obtain a low-resolution version of the histogram;

(2) Generate an initial population based on this low-resolution version of the histogram;

(3) Define the fitness function;

(4) Apply the learning strategy to improve the fitness value of the string;

(5) Compare the optimal string with the current string, and if the optimal string is better than the current string, replace the current string with the optimal string. Otherwise, perform selection, crossover and mutation operations to generate the next population, go to step 2;

(6) Project the obtained optimal threshold to the original space to obtain the optimal threshold segmentation effect of the original image.

Morphological processing

Morphological processing of erosion and dilation is performed on the thresholded image. There will be gaps between the flawed regions, and the noise in the image can be removed through morphological methods such as corrosion and expansion, and the flawed regions will become connected.

This method can be applied in the automatic detection system of fabric defect as shown in Figure 2, and this system comprises belt detection cloth module 1, illuminating device module 2, cloth conveying device module 3, photographing device module 4, rotary encoder module 5, image Acquisition device module 6, host computer processing and reality device module 7. The lighting device module 2 provides good light conditions for the collection of fabric images, improves the quality of the fabric images, and is beneficial to the next image processing. The cloth conveying device module 3 is a cloth conveying device, so that the linear camera and the cloth form a constant relative motion. The cloth conveying device module 3 includes a cloth withdrawing roller, a cloth feeding roller, a motor and its drive, and a motor controller. The photographing device module 4 is a CCD line scan camera, which is used for photographing images in industry. The main function of the rotary encoder module 5 is to provide the main timing pulse for the camera. The image acquisition device module 6 is an image acquisition card, so that the data obtained by each camera is converted into a digital image by the image acquisition card. The host computer processing and display device module 7 is a host computer processing and display device, which processes the collected pictures and outputs analysis results.

The algorithm principle of the patent digital image processing of the present invention is as follows:

The invention first preprocesses the collected cloth defect image, and then uses discrete wavelet transform to decompose the image to extract components in different dimensions. Then, the genetic algorithm is used to process the approximate image histogram after wavelet decomposition to obtain a threshold, which is used to perform threshold segmentation on the image, and finally the image separated from the blemish and the background image is obtained.

This embodiment is mainly divided into the following steps:

(1) Preprocessing the collected images with defects: grayscale the collected images first, and then use histogram averaging and median filtering to improve the quality of the original image (see Figure 3) as a whole , to facilitate the next step of processing;

(2) Use wavelet decomposition to obtain multi-scale sub-images: the wavelet decomposition of the image is to perform one-dimensional discrete wavelet transform on the image in the horizontal and vertical directions respectively. First perform high-pass and low-pass filtering on the image by row and down-sampling to obtain two self-images of the same size as the input images, and then perform high-pass and low-pass filtering on the self-image by column and down-sampling to obtain 4 input images A subimage the size of a quarter of the image. The sub-image obtained by wavelet decomposition is shown in Figure 4.

(3) Fusion the sub-images obtained in step (2): After wavelet decomposition, the edge information of fabric defects at multiple scales is obtained, and the edge sub-images at each scale are fused with certain rules to obtain the optimal Defect edge information. The image obtained after fusion is shown in Figure 5.

(4) According to the sub-image obtained in step (2), the threshold value is calculated by genetic algorithm, and the threshold value is used to perform threshold segmentation on the fusion image obtained in step (3).

(5) Morphological processing: There will be gaps between the defective regions, and the noise in the image can be removed through morphological methods such as corrosion and expansion, and the defective regions will become connected.

It is not difficult to find that the present invention combines wavelet transform and genetic algorithm, first decomposes the preprocessed image with discrete wavelet transform, extracts the components on different dimensions of the image, obtains a lower resolution image, and achieves dimensionality reduction the goal of. Then, the genetic algorithm is used to process the approximate image histogram after wavelet decomposition to obtain a threshold value, and then threshold value segmentation is performed on the original image to separate the defect part from the fabric background. After wavelet decomposition, the genetic algorithm is used to make the calculation faster than the traditional genetic algorithm. After being processed by the algorithm of the present invention, the segmentation effect of the cloth blemish is accurate, the segmentation speed is fast, and the original shape of the blemish is well preserved.

Claims (5)

1. A fabric flaw detection method based on wavelet transform and genetic algorithm, is characterized in that, comprises the following steps: (1) Preprocessing the collected images with defects; (2) Use wavelet decomposition on the preprocessed image to obtain multi-scale sub-images; (3) Fusion the sub-images to obtain the optimal defect edge information; (4) Utilize genetic algorithm to calculate threshold value to sub-image, and adopt described threshold value to carry out threshold value segmentation to the image after fusion; (5) Perform morphological processing on the thresholded image. 2. the fabric flaw detection method based on wavelet transform and genetic algorithm according to claim 1, is characterized in that, described step (1) is specifically: the image with flaw that gathers is first carried out grayscale processing, Then use histogram averaging and median filtering to improve the quality of the original image as a whole. 3. the fabric flaw detection method based on wavelet transform and genetic algorithm according to claim 1, is characterized in that, described step (2) is specifically: carry out high-pass and low-pass filtering and carry out down-sampling to image respectively by line, Obtain two self-images of the general size of the input image, and then perform high-pass and low-pass filtering and down-sampling on the self-image by column, and obtain four sub-images of the quarter size of the input image, which are A _j+1 , D _j+1 ^H , D _j+1 ^V , D _j+1 ^D ; among them, A _j+1 is the result of low-pass filtering in the row and column directions, representing the profile signal of the next scale; D _j+1 ^H is the result of low-pass filtering in the row direction and high-pass filtering in the column direction, representing the overview of the detail signal in the vertical direction in the horizontal direction; D _j+1 ^V is the result of high-pass filtering in the row direction and low-pass filtering in the column direction, representing the horizontal The overview of the detail signal in the direction in the vertical direction; D _j+1 ^D is the result of high-pass filtering in the row and column directions, representing the detail signal in the diagonal direction. 4. the fabric defect detection method based on wavelet transform and genetic algorithm according to claim 1, is characterized in that, described step (3) specifically comprises the following substeps: (31) Starting from the largest scale, let the scale j be the largest scale J, link the non-zero pixels in the edge image E _j (x, y) with similar modulus and phase angle, and set the chain length threshold, delete less than The short chain of the chain length obtains the single-pixel wide edge image E _j (x, y) at the scale j; (32) For each edge pixel in the edge image E _j (x, y), search for a 3×3 neighborhood centered on the pixel corresponding to the point at scale j-1, and calculate the phase angle of the model value in the neighborhood All similar points are added to the edge image E _j (x, y) as edge points, and the non-zero pixels with similar modulus and phase angles in the edge image E _j (x, y) are linked, and the chain Long threshold, delete short chains smaller than the chain length, so as to obtain the defect edge image E _j-1 (x, y) under the scale j-1; (33) If the scale j>1, then return to step (32), if j=1, then the obtained edge image E ₁ (x, y) is the defect edge image after multi-scale fusion; Wherein, E _j (x, y) is an edge image at each scale j, where j=1, 2, 3, . . . , J. 5. the fabric defect detection method based on wavelet transform and genetic algorithm according to claim 1, is characterized in that, Described step (4) comprises following substep: (41) Perform histogram calculation on the sub-image to obtain a low-resolution version of the histogram; (42) generating an initial population based on the histogram of the low-resolution version; (43) Define the fitness function; (44) applying the learning strategy to improve the fitness value of the character string; (45) Compare the optimal character string and the current character string, if the optimal character string is better than the current character string, replace the current character string with the optimal character string; otherwise, perform selection, crossover and mutation operations to generate the next population, And return to step (42); (46) Project the obtained optimal threshold to the original space to obtain the optimal threshold segmentation effect of the original image.