Image recognition-based textile waste automatic classification method and device

Abstract

The application provides a kind of textile waste automatic classification method and device based on image recognition, it is related to image analysis and processing field.The method extracts reflectivity features, color moment features, edge granularity features, overall curling degree features, deep learning abstract features and the like by respectively extracting the target image to be identified, and sequentially based on the color features and texture features extracted gradually classifies, and finally completes the classification of the 8 basic categories of the target image to be identified.The textile waste automatic classification method and device based on image recognition provided by the application have high feasibility, different image features and abstract features are extracted for cotton textile waste with different colors, shapes and textures, which can represent various textile waste while improving the recognition efficiency and classification accuracy of the entire waste image, and the scheme has high applicability, can effectively identify and classify multiple types of textile waste, improve the sorting and recycling efficiency of textile waste, and the scheme is efficient, practical, objective and accurate.

Description

Technical Field

[0001] This invention relates to the field of image analysis and processing, and in particular to an automatic classification method and apparatus for textile waste based on image recognition. Background Technology

[0002] Cotton is an important textile raw material, but due to natural growth, variety differences, harvesting, and transportation, cotton raw materials vary in shape and quality. In cotton screening, cotton that meets the requirements of textile processing is usually selected, while the remaining cotton containing impurities or of poor quality is considered textile waste. If this textile waste can be further classified and screened, some categories can still be used for textile processing or other reuses. Traditional manual sorting is labor-intensive and lacks accuracy. Therefore, how to quickly, conveniently, and accurately classify this type of textile waste is an urgent problem to be solved. Summary of the Invention

[0003] To address the aforementioned problems, this invention proposes an automatic classification method and apparatus for textile waste based on image recognition. By using image processing technology to extract multiple features and automatically classify textile waste, the invention achieves automatic identification of cotton textile waste types.

[0004] The technical solution adopted in this invention is as follows:

[0005] An automatic classification method for textile waste based on image recognition is proposed. This method extracts reflectivity features, color moment features, edge granularity features, overall curvature features, and deep learning abstract features from the target image to be identified. Then, it performs step-by-step classification based on these extracted color and texture features, and finally completes the classification of the target image into eight basic categories.

[0006] On the other hand, the present invention also provides an automatic textile waste sorting device based on image recognition. The device is an automatic sorting device composed of module units corresponding to the steps of the aforementioned automatic textile waste sorting method, for automatically sorting the categories to which multiple types of textile waste to be identified belong.

[0007] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0008] The image recognition-based automatic classification method and device for textile waste provided by this invention are highly feasible. It extracts different image features and abstract features for cotton textile waste with different colors, shapes and textures. While being able to characterize various types of textile waste, it improves the recognition efficiency and classification accuracy of the entire waste image. The solution has high applicability and can effectively identify and classify multiple types of textile waste, improving the sorting and reuse efficiency of textile waste. The solution is efficient, practical, objective and accurate. Attached Figure Description

[0009] Figure 1 This is a flowchart of an automatic classification method for textile waste based on image processing provided in an embodiment of the present invention.

[0010] Figure 2 These are standard sample images of eight basic types of textile waste provided in this embodiment of the invention: (a) - "gray non-reflective"; (b) - "white non-reflective"; (c) - "reflective long strips of yarn"; (d) - "reflective clump-shaped yarn"; (e) - "reflective blocky flocs"; (f) - "reflective clump-shaped flocs"; (g) - "reflective non-clump / blocky flocs with black dust"; and (h) - "reflective non-clump / blocky flocs without black dust".

[0011] Figure 3 This is a schematic diagram of three types of feature extraction provided in the embodiments of the present invention. Detailed Implementation

[0012] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Based on the embodiments in this application, other similar embodiments obtained by those skilled in the art without creative effort should all fall within the scope of protection of this application.

[0013] In the following embodiments, textile waste is selected as the target identification fabric to provide a detailed description of the specific solution. In other embodiments, the textile waste mentioned may also refer to textile fabrics that need to be automatically classified in similar scenarios. It is merely the analysis object targeted by the technical solution of the present invention, and the solution shall be based on its applicability to the solution of the present invention and the solution to the corresponding technical problem.

[0014] Example 1

[0015] like Figure 1 As shown in the figure, this embodiment is an automatic classification method for textile waste based on image recognition, which includes the following steps:

[0016] S01, Extract the target image to be identified. In this embodiment, the target image to be identified is a set of images of various types of textile waste to be classified, which are obtained by a specially designed textile waste box image acquisition box.

[0017] In this embodiment, textile waste refers to solid waste of textile raw materials, which includes at least several categories such as yarn, clumps, threads, and lint. For effective identification and classification, this embodiment divides the target textile waste into eight basic categories: "gray non-reflective," "white non-reflective," "reflective long threads," "reflective clumps," "reflective clumps," "reflective clumps," "reflective non-clump / block lint with black dust," and "reflective non-clump / block lint without black dust." Figure 2 As shown in (a)-(h), these are standard sample images of textile waste corresponding to the eight basic categories.

[0018] After setting the classification criteria, images of various textile waste materials to be analyzed and classified are collected as a set of images to be classified. In this embodiment, a textile waste classification model will be constructed, and image features will be extracted based on differences in fabric appearance and material for identification and classification.

[0019] S02, extract the reflectivity features of textile waste images, and divide the image set to be classified into two categories: reflective and non-reflective based on the reflectivity features;

[0020] When classifying and recognizing target images, different image features are extracted sequentially and used for classification. In this embodiment, the reflectivity feature of the textile waste image is first extracted. The reflectivity feature can be used to reflect the reflectivity of the material to which the target textile waste belongs. Through this reflectivity feature, all images to be classified can be divided into two main categories: reflective and non-reflective. The specific steps include the following:

[0021] S201, convert all the RGB images of textile waste to be classified to the HSV color space, calculate the histogram of the vector on the V (brightness) channel, record the number of peaks p (vertical axis) of the histogram and its corresponding brightness vm (horizontal axis), and record the minimum brightness (horizontal axis) of the histogram as v0 and the maximum brightness (horizontal axis) as v1, and obtain the image reflectance Ri = 2p / (v1-v0);

[0022] S202, Select standard samples of reflective and non-reflective types and calculate their reflectance Rr and Rnr respectively, where Rr is the standard reflectance value of the reflective type and Rnr is the standard reflectance value of the non-reflective type.

[0023] S203, using the k-means clustering algorithm, calculate the reflectance Ri of the target image to be identified in the brightness space, and calculate the Euclidean distance d between Ri and the reflectance of the two types of standard samples respectively:

[0024] d(Ri,Rr)=(Ri-Rr) 2 d(Ri,Rnr)=(Ri-Rnr) 2 (1).

[0025] If d(Ri, Rr) < d(Ri, Rnr), the corresponding target image belongs to the reflective category; if d(Ri, Rr) > d(Ri, Rnr), the corresponding target image belongs to the non-reflective category; based on this, the first classification is completed.

[0026] S03, extract the color moments of the non-reflective target images to be recognized, and further classify the reflective images to be recognized into two basic categories, "gray non-reflective" and "white non-reflective", according to the color moment features. The specific steps are as follows:

[0027] S301, as Figure 3 shown, convert the RGB image of non-reflective textile waste to be recognized into the YUV color space, and calculate the first three-order color moments of each non-reflective target image to be recognized.

[0028] Color moments are features formed by three variables, the first-order moment (mean), the second-order moment (variance), and the third-order moment (skewness) of statistical colors. The color spaces include RGB (red, green, blue) and HSV (hue, saturation, brightness) spaces, etc. In this embodiment, the color moment features of the HSV space are adopted, that is, the first-order moments, second-order moments, and third-order moments of the three variables H, S, and V in the image are calculated respectively to obtain nine values as its color moment feature values. The calculation formulas of the first three-order color moments are shown as follows (2)-(4):

[0029]

[0030]

[0031]

[0032] S302, form a 9-dimensional histogram vector F from the first three-order color moments of the three components Y, U, and V spaces of the image to be recognized color = [μ Y , σ Y , s Y , μ U , σ U , s U , μ V , σ V , s V , select two standard samples of gray non-reflective and white non-reflective and calculate the corresponding standard color moments F gray and F white , where F gray is the standard color moment of the gray non-reflective category, and F white is the standard color moment of the white non-reflective category.

[0033] S303, the Euclidean distance d(Fi,F) is calculated using the k-means clustering algorithm according to formula (1). gray ) and d(Fi,F white ), where Fi is the color moment of the i-th image to be identified.

[0034] If d(Fi,F) gray ) <d(Fi,F white If d(Fi,F), then the image of the non-reflective textile waste with serial number i belongs to the gray non-reflective category; gray )>d(Fi,F white If the image of the reflective textile waste with serial number i is a white, non-reflective material, then the image belongs to the white, non-reflective category.

[0035] S04, use the deep learning MobilenetV3 network to further classify reflective images, specifically into two non-basic categories: thread-like and block / flocculent, including the following steps:

[0036] S401, firstly, the reflective image I0 is normalized by converting the pixel value data range of 0-255 in the regular RGB image into floating-point numbers of 0-1 for easier input into the network calculation. Then, based on the mean μ and standard deviation σ of the existing RGB three channels, the preprocessed image I1 is obtained by standardizing according to Formula 1:

[0037]

[0038] S402, through the forward process of the deep learning network, the normalized reflective image is converted into multi-channel low-resolution deep features F, which can be regarded as abstract features and used for further classification.

[0039] S403 inputs the multi-channel, low-resolution deep features F into the final MLP layer (Multilayer Perceptron) of the deep learning network, and aggregates the multi-channel deep features F into two outputs z. i Then, the Softmax function (as shown in Equation 6) is used to select z. i Maximum value. Based on the result, Softmax(z) i The value of ) determines whether the input target image belongs to the category of thread-like or block / flocculent textile waste:

[0040]

[0041] Where c represents the current channel, c n This represents the number of output channels. i With z cThese represent the output of the i-th class and the output of the current channel, respectively. The value of i is 1 or 2. When i = 1, z1 represents the output value calculated for the image of filament-type textile waste; when i = 2, z2 represents the output value calculated for the image of block / floss-type textile waste. Finally, Softmax(z... i This indicates that the category corresponding to the larger value between z1 and z2 is selected as the category to which the current image to be identified belongs.

[0042] Next, in this embodiment, the images of reflective filaments and reflective blocks / flocculents will be further classified using two different edge granularities, such as... Figure 3 As shown. The edge granularity includes lateral edge granularity and longitudinal edge granularity, which can be used to reflect the dispersion and granularity of the material to which the target image to be identified belongs. Typically, the lateral and longitudinal edge granularity of linear / clump / block / flossy fabrics are different.

[0043] S05, extract the lateral edge granularity of the reflective thread-like images, and based on this lateral edge granularity feature, further divide the thread-like images into two basic categories: reflective long strip threads and reflective clustered threads, specifically including:

[0044] S501 converts color images of textile waste such as silk threads into grayscale images, uses Gaussian filtering to remove image noise, and employs Canny edge detection to extract granular edges in the target image.

[0045] S502, using the extracted granular edges as rows as standards, calculate the distances between adjacent edge particle points to obtain the row vector of adjacent particle distances in the i-th row [d1,d2,d3,....]. The horizontal granularity R of the i-th row is obtained by weighting and averaging the vector elements. i = (d1+d2+d3+...+dN) / N, then the overall horizontal granularity of the image is the column vector R = [R1,R2,R3,...]. T ;

[0046] S503, select one standard sample image each of long strip filaments and tufted filaments, and denote the lateral granularity vector of the long strip filament standard sample image as Rsa=[Rs1,Rs2,Rs3,... T The lateral granularity vector of the standard sample image of the tufted filaments is Rsb = [Rs1*, Rs2*, Rs3*, ...]. T The horizontal granularity vector of the target image to be classified is denoted as Rp = [R1, R2, R3, ... T Calculate the Euclidean distances d(Rp,Rsa) and d(Rp,Rsb) between vectors Rp and Rsa, and Rsb, respectively.

[0047] If d(Rp, Rsa) < d(Rp, Rsb), the target image to be classified belongs to the reflective long strip silk thread category; if d(Rp, Rsa) > d(Rp, Rsb), the image to be classified belongs to the reflective mass silk thread category.

[0048] S06, extract the longitudinal edge granularity of the reflective block / floc-like image, and further divide the reflective block / floc state image into two basic categories, namely the reflective block / mass floc and other floc categories, based on this longitudinal edge granularity feature, specifically including:

[0049] S601, convert the color image of the block / floc-like textile waste into a grayscale image, use Gaussian filtering to remove image noise, and extract the granular edges in the target image using canny edge detection;

[0050] S602, taking the columns as the standard for the extracted granular edges, calculate the distances between adjacent edge granular points, and obtain the adjacent granular distance column vector [d1, d2, d3,...] of the j-th column T , and take the weighted mean of the vector elements to obtain the longitudinal granularity C of the j-th column j = (d1 + d2 + d3 +... + dN) / N, then the overall longitudinal granularity of the image is the row vector R = [R1, R2, R3,...];

[0051] S603, select one standard sample image each for the block / mass floc category and other floc categories, denote the longitudinal granularity vector of the block / mass floc category standard sample image as Rsa* = [Rs1*, Rs2*, Rs3,...], the longitudinal granularity vector of the other floc category standard sample image as Rsb* = [Rs1**, Rs2**, Rs3**,...], and in addition, denote the longitudinal granularity vector of the target image to be classified as Rp* = [R1*, R2*, R3,...] T , and calculate the Euclidean distances d(Rp*, Rsa*) and d(Rp*, Rsb*) between the Rp* vector and the Rsa* and Rsb* vectors respectively.

[0052] If d(Rp*, Rsa*) < d(Rp*, Rsb*), it is considered that the target image to be classified belongs to the block / mass floc category; if d(Rp*, Rsa*) > d(Rp*, Rsb*), it is considered that the target image to be classified belongs to the other floc category.

[0053] S07, extract the overall curl degree image feature of the reflective block / mass floc image, and further divide the reflective block / mass floc image into two basic categories, namely the reflective block floc category and the reflective mass floc category, based on this overall curl degree image feature;

[0054] S701. Set the image of block / clump-like flocculent textile waste as I, and use the 3×3 Sobel operators Gx and Gy to calculate the horizontal and vertical gradients of the image respectively:

[0055]

[0056] S702. Calculate the overall convolution degree Dcur of a single image I using formula (8). The overall convolution degree Dcur can be used to reflect the degree of curling of the material of the target image to be recognized. For different fabrics, the more easily curled and clumped fabrics have smaller horizontal and vertical differences. Its essence is the weighted average of the ratio of horizontal and vertical gradients of each pixel point:

[0057]

[0058] where i and j represent the pixel coordinates in the image, and N is the total number of pixels in the image.

[0059] S703. Select one standard sample image each of block-like and clump-like textile waste. Denote the overall convolution degree of the standard sample image of block-like textile waste as Dcur1, and denote the overall convolution degree of the standard sample image of clump-like textile waste as Dcur2.

[0060] S704. Use the k-means clustering algorithm to calculate the corresponding overall convolution degree Dcur_i for all block / clump-like flocculent textile waste images to be classified, and calculate the Euclidean distances between Dcur_i, Dcur1, and Dcur2:

[0061] d(Dcur_i,Dcur1)=(Dcur_i - Dcur1) 2 ,d(Dcur_i,Dcur2)=(Dcur_i - Dcur2) 2 .

[0062] If d(Dcur_i,Dcur1) < d(Dcur_i,Dcur2), then the target image to be classified belongs to the reflective block-like flocculent textile waste; if d(Dcur_i,Dcur1) < d(Dcur_i,Dcur2), then the target image to be classified belongs to the reflective clump-like flocculent textile waste, thus completing the classification of the test image into two basic categories of block-like and clump-like.

[0063] S08. Extract the color moment features of other flocculent images, and further classify other flocculent images into two basic categories: reflective non-clump / block-like flocculent with black chips and reflective non-clump / block-like flocculent without black chips based on the color moment features;

[0064] S801. Calculate the color moment features of other flocculent textile waste images using the color moment calculation formulas (2)-(4) in step S301;

[0065] S802. Select standard sample images of textile waste with and without black chips respectively, and calculate the color moment features Fb and Fw.

[0066] S803. Let the color moment of the k-th image to be measured be Fk. Similarly, use the k-means clustering algorithm to calculate the Euclidean distances d(Fk, Fb) and d(Fk, Fw) between Fk, Fb, and Fw.

[0067] If d(Fk, Fb) < d(Fk, Fw), then the k-th target image to be recognized belongs to the textile waste of the floc type with black chips; if d(Fk, Fb) > d(Fk, Fw), then the k-th target image to be recognized belongs to the textile waste of the floc type without black chips.

[0068] At this time, this embodiment finally completes the classification of all textile waste images to be classified into eight basic categories.

[0069] Embodiment 2

[0070] This embodiment is an automatic classification device for textile waste based on image recognition. The device is an automatic classification device composed of module units corresponding to the steps of the automatic classification method for textile waste in any of the foregoing embodiments, and is used to automatically classify the categories of various types of textile waste to be recognized.

[0071] In summary, the automatic classification method and device for textile waste based on image recognition provided by the present invention have high feasibility. Different image features and abstract features are extracted for cotton textile waste with different colors, shapes, and textures, which can improve the recognition efficiency and classification accuracy of the entire waste image while characterizing various types of textile waste. The solution has high applicability, can effectively identify and classify various types of textile waste, and improve the sorting and recycling efficiency of textile waste. The solution is efficient, practical, objective, and accurate.

[0072] All features disclosed in this specification, or all steps in the disclosed methods or processes, except for mutually exclusive features and / or steps, can be combined in any way.

[0073] The present invention is not limited to the foregoing specific embodiments. The present invention extends to any new feature or any new combination disclosed in this specification, as well as any new combination of the steps of any new method or process disclosed.

Claims

1. An automatic classification method for textile waste based on image recognition, used for automatic identification and classification of images of multiple types of textile waste, characterized in that, It includes the following steps: Step S01, extract the target image to be recognized, and obtain the image set of various types of textile waste to be classified; Step S02, extract the reflectance features of the image set to be classified, and divide the image set to be classified into two major categories: reflective and non-reflective based on the image reflectance features; Step S03, extract the color moments of the non-reflective target images to be recognized, and further divide the recognized reflective images into two categories: gray non-reflective and white non-reflective according to the color moment features; Step S04, use a deep learning network to extract abstract features of the reflective target images to be recognized, and further divide the reflective target images to be recognized into two categories: silk thread category and block / fluffy category based on the abstract features; where the abstract feature is to convert the normalized reflective image into a multi-channel low-resolution deep feature through the forward process of the deep learning network; Step S05, extract the horizontal edge granularity of the reflective silk thread images, and further divide the reflective silk thread images into two categories: reflective long silk threads and reflective agglomerated silk threads based on the horizontal edge granularity features; Step S06, extract the vertical edge granularity of the reflective block / fluffy images, and further divide the reflective block / fluffy images into two categories: reflective block / agglomerated fluff and other fluff categories based on the vertical edge granularity features; Step S07, extract the overall curl image feature of the reflective block / agglomerated fluff images, and further divide the reflective block / agglomerated fluff images into two categories: reflective block fluff category and reflective agglomerated fluff category based on the overall curl image feature; Step S08, extract the color moment features of the other fluff images, and further divide the other fluff images into two categories: reflective non-block / agglomerated black屑 fluff category and reflective non-block / agglomerated non-black屑 fluff category based on the color moment features.

2. The automatic classification method for textile waste based on image recognition as described in claim 1, characterized in that, The textile waste is cotton textile waste.

3. The automatic classification method for textile waste based on image recognition as described in claim 1, characterized in that, The specific content of the said step S02 includes: Step S201, convert all RGB images to be classified to the HSV color space, calculate the histogram of the vector on the V brightness channel, record the peak number p of the histogram and its corresponding brightness vm, record the minimum brightness of the histogram as v0, and the maximum brightness as v1, and obtain the image reflectance Ri = 2p / (v1 - v0); Step S202, select the standard samples of the reflective and non-reflective categories and calculate their reflectances Rr and Rnr respectively, where Rr is the reflectance standard value of the reflective category and Rnr is the reflectance standard value of the non-reflective category; Step S203, calculate the reflectance Ri of the target image to be recognized, and calculate the Euclidean distances d between Ri and the reflectances of the two types of standard samples respectively: d(Ri,Rr)=(Ri-Rr) 2 ,d(Ri,Rnr)=(Ri-Rnr) 2 (1); If d(Ri, Rr) < d(Ri, Rnr), then the corresponding target image belongs to the reflective category; if d(Ri, Rr) > d(Ri, Rnr), then the corresponding target image belongs to the non-reflective category.

4. The automatic classification method for textile waste based on image recognition as described in claim 3, characterized in that, The said step S03 includes: S301, convert the non-reflective RGB images to be recognized to the YUV color space, and calculate the first three color moments of the three components Y, U, V spaces of each non-reflective target image to be recognized; The calculation formulas of the first three color moments are as shown in the following (2)-(4): （2） （3） （4）； S302, the first three color moments of the three components Y, U, V space are combined into a 9-dimensional histogram vector F. color =[μ Y ,σ Y ,s Y ,μ U ,σ U ,s U ,μ V ,σ V ,s V Two standard samples, gray non-reflective and white non-reflective, were selected, and their corresponding standard color moments F were calculated. gray and F white , where F gray For standard color moments of gray non-reflective material, F white Standard color dimensions for white, non-reflective materials; S303, calculate the Euclidean distance d according to formula (1). Fi , F gray ) and d( Fi , F white ), where Fi is the color moment of the i-th image to be recognized; If d( Fi , F gray ) <d( Fi , F white If d( Fi , F gray )>d( Fi , F white If the i-th image of non-reflective textile waste belongs to the white non-reflective category, then the i-th image of non-reflective textile waste belongs to the white non-reflective category.

5. The automatic classification method for textile waste based on image recognition as described in claim 1, characterized in that, The specific steps of S04 include: Step S401: Perform normalization processing on the reflective target image to be recognized I0 to obtain the preprocessed normalized image I1. Step S402: Convert the normalized image I1 into a multi-channel low-resolution deep feature F through the forward processing of the deep learning network. Step S403: Input the multi-channel low-resolution deep features F into the last layer of the deep learning network, and aggregate the multi-channel deep features F into two types of output z. i Use formula (6) to select z i The maximum value is determined by the result Softmax(z). i The value of ) determines the category to which the input target image to be identified belongs: （6）， Where c represents the current channel, c n z is the number of output channels; i With z c These represent the output of the i-th class and the output of the current channel, respectively, where i takes the value of 1 or 2; when i=1, z1 represents the output value calculated for the image of filament-type textile waste, and when i=2, z2 represents the output value calculated for the image of block / floss-type textile waste. Softmax(z i This indicates that the category corresponding to the larger value between z1 and z2 is selected as the category to which the current image to be identified belongs.

6. The automatic classification method for textile waste based on image recognition as described in claim 5, characterized in that, The specific steps of S05 include: Step S501: Convert the silk-like color image into a grayscale image, remove image noise, and extract the granular edges of the target image. Step S502: Using the extracted granular edges as a row standard, calculate the distance to neighboring edge particle points to obtain the first... i The distances between adjacent particles in a row are given by the row vector [d1, d2, d3, ...]. The weighted average of the vector elements is taken to determine the horizontal granularity R of the i-th row. i =(d1+d2+d3+...+dN) / N, then the overall horizontal granularity of the target image is the column vector R. col =[R1,R2,R3,...] T ; Step S503: Select one standard sample image each of long strip filaments and tufted filaments, and denote the lateral granularity vector of the long strip filament standard sample image as Rsa=[Rs1,Rs2,Rs3,...]. T The lateral granularity vector of the standard sample image of the tufted filaments is Rsb=[Rs1 Rs2 Rs3 ,...] T The horizontal granularity vector of the target image to be classified is denoted as Rp=[R1,R2,R3,...]. T Calculate the Euclidean distances d(Rp,Rsa) and d(Rp,Rsb) between vectors Rp and Rsa and Rsb, respectively. If d(Rp, Rsa) < d(Rp, Rsb), the target image to be classified belongs to the reflective long-strip silk category; if d(Rp, Rsa) > d(Rp, Rsb), the image to be classified belongs to the reflective lump-like silk category.

7. The automatic classification method for textile waste based on image recognition as described in claim 1, characterized in that, The specific steps of S06 include: Step S601: Convert the block / fluffy-like color image into a grayscale image, remove image noise, and extract the granular edges of the target image. Step S602: Using columns as the standard, calculate the distances between adjacent edge particles of the extracted granular edges to obtain the column vector of adjacent particle distances in the j-th column [d1, d2, d3, ...]. T The weighted average of the vector elements is used to determine the vertical granularity C of the j-th column. j =(d1+d2+d3+...+dN) / N, then the overall vertical granularity of the image is the row vector R. row =[R1,R2,R3,...]; Step S603: Select one standard sample image each for block / clump-like flocs and other flocs, and denote the vertical granularity vector of the block / clump-like floc standard sample image as Rsa. =[Rs1 Rs2 Rs3 [,...], the vertical granularity vector of other flocculent standard sample images is Rsb =[Rs1 Rs2 Rs3 [,...], where the vertical granularity vector of the target image to be classified is Rp =[R1 R2 R3 ,...], calculate Rp respectively With Rsa Rsb Euclidean distance d(Rp) between vectors ,Rsa ) and d(Rp ,Rsb ); If d(Rp) ,Rsa ) <d(Rp ,Rsb If d(Rp) is a block / clump-like flocculent, then the target image to be classified is considered to belong to the block / clump-like flocculent class; if d(Rp) is a block / clump-like flocculent, then the target image to be classified is considered to belong to the block / clump-like flocculent class. ,Rsa )>d(Rp ,Rsb If the target image to be classified belongs to another category, then it is considered to belong to another category.

8. The automatic classification method for textile waste based on image recognition as described in claim 7, characterized in that, The specific steps of S07 include: Step S701: Let the block / lump / fluffy-like textile waste image be I, and use the Sobel operator to calculate the horizontal and vertical gradients Gx and Gy of the image I respectively. （7）； Calculate the overall convolution degree Dcur of a single image I using formula (8). （8）， Where i and j represent the pixel coordinates in the image, and N is the total number of pixels in the image. Step S702: Select one standard sample image each for the block and lump types of textile waste. Denote the overall convolution degree of the standard sample image of the block type of textile waste as Dcur1, and the overall convolution degree of the standard sample image of the lump type of textile waste as Dcur2. Step S703: Calculate the corresponding overall convolution degree Dcur_i for all reflective block / lump / fluffy-like images to be classified, and calculate the Euclidean distances between Dcur_i and Dcur1 and Dcur2. d(Dcur_i,Dcur1)=(Dcur_i-Dcur1) 2 ,d(Dcur_i,Dcur2)=(Dcur_i-Dcur2) 2 ; If d(Dcur_i, Dcur1) < d(Dcur_i, Dcur2), the target image to be classified belongs to the reflective block / fluffy category; if d(Dcur_i, Dcur1) > d(Dcur_i, Dcur2), the target image to be classified belongs to the reflective lump / fluffy category.

9. The automatic classification method for textile waste based on image recognition as described in claim 1, characterized in that, The specific steps of S08 include: Step S801: Calculate the color moment features of other fluffy-like images. Step S802: Select the standard sample images of textile waste with and without black chips and calculate the color moment features Fb and Fw respectively. Step S803: Let the color moment of the k-th image to be measured be Fk, and calculate the Euclidean distances d(Fk, Fb) and d(Fk, Fw) between Fk and Fb and Fw. If d(Fk, Fb) < d(Fk, Fw), the k-th target image to be recognized belongs to the reflective non-block / lump black-chip-containing fluffy category; if d(Fk, Fb) > d(Fk, Fw), the k-th target image to be recognized belongs to the reflective non-block / lump non-black-chip-containing fluffy category.

10. An automatic textile waste sorting device based on image recognition, characterized in that, The device is an automatic classification device composed of module units corresponding to the steps of any one of the textile waste automatic classification methods in claims 1-9, and is used for automatically classifying multiple types of textile waste to be recognized.

Images