Textile waste identification and classification method

Abstract

The invention provides a textile waste identification and classification method, which comprises the following steps: obtaining a cotton solid waste image, and obtaining a to-be-identified image after image processing; performing feature extraction on the to-be-identified image, wherein the extracted features include impurity distribution density, impurity coverage rate, strip length feature, structure density, edge ambiguity, white cotton proportion area, yellow debris proportion area, texture energy response and wrinkle edge distribution; and combining all the extracted features into a feature vector, inputting the feature vector into a random forest classifier for training and classification, and obtaining an identification result of the cotton solid waste. According to the method provided by the invention, specific feature extraction is carried out on some specific types of cotton solid wastes from the aspects of color, morphology, structural features and the like, visual features of new types of cotton solid wastes are comprehensively represented, and the random forest classifier is combined, so that the solid wastes can be more efficiently and accurately classified.

Description

Technical Field

[0001] The present invention relates to the technical field of textile waste classification, and in particular to a method for identifying and classifying textile waste. Background Art

[0002] There are many types of waste generated by textiles, and sorting and classification is very important. Among them, common solid waste can be roughly divided into several categories according to the actual material characteristics: slag seeds with more debris, including broken seed cotton and dust slag cotton; formed silk yarns, including sizing waste yarn, natural color waste yarn and blended waste; and lumpy block types, including bellows and flying flower natural color; a total of seven categories. The traditional solution based on manual sorting is inefficient, time-consuming and labor-intensive.

[0003] With the rapid development of image recognition technology, the use of image recognition methods for waste identification and classification has become a reliable method to improve classification efficiency. For example, in the Chinese invention patent with application number 202211455556.0, a method and device for automatic classification of textile waste based on image recognition is disclosed. This method extracts reflectivity features, color moment features, edge granularity features, overall curling features, deep learning abstract features, etc. from the target image to be identified, and then classifies them step by step based on the extracted color features and texture features, and finally completes the classification of the basic categories of the target image to be identified.

[0004] However, the features extracted from the above classification methods and other existing classification methods fail to accurately display the material characteristics of different wastes, and the accuracy and efficiency of classification are still far from practical application. In addition, in the judgment method adopted, as long as there is a judgment bias in the parent node, it will affect the overall analysis results. Therefore, there is an urgent need for a method that can accurately and efficiently identify and classify textile waste to improve the efficiency of sorting and recycling of textile waste. Summary of the invention

[0005] The present invention aims to solve at least one of the above-mentioned technical problems existing in the prior art.

[0006] To this end, the present invention provides a method for identifying and classifying textile waste.

[0007] The present invention provides a method for identifying and classifying textile waste, comprising:

[0008] Acquire cotton solid waste images, and obtain the image to be identified after image processing;

[0009] Extracting features of the image to be identified, wherein the extracted features include impurity distribution density, impurity coverage, thin strip length characteristics, structure density, edge fuzziness, white cotton area, yellow debris area, texture energy response, and wrinkle edge distribution;

[0010] All the extracted features are combined into a feature vector, and the feature vector is input into a random forest classifier for training and classification to obtain the recognition result of cotton solid waste;

[0011] Among them, the types of cotton solid waste include broken seed cotton, dusty cotton, sizing waste yarn, natural color waste yarn, blended waste, bellows waste and natural color flying waste.

[0012] The textile waste identification and classification method according to the above technical solution of the present invention may also have the following additional technical features:

[0013] In the above technical solution, the image to be identified is a grayscale image;

[0014] The method for extracting impurity coverage characteristics and impurity distribution density characteristics from the image to be identified includes:

[0015] Adaptive threshold binarization algorithm is used to extract impurity areas other than cotton from the image to be identified, and connected domain analysis is used to identify independent impurities.

[0016] Calculate the impurity distribution density and impurity coverage:

[0017]

[0018] Where W represents the width of the image to be identified; H represents the height of the image to be identified; N particle Indicates the total amount of impurities; D particle Indicates the density of impurity distribution; A i represents the area of ​​the i-th impurity; C particle represents the impurity coverage;

[0019] Wherein, the connected domain analysis includes:

[0020] Set the connectivity between pixels;

[0021] The non-background pixels in the binary image are traversed. According to the connectivity between pixels, those that meet the connectivity relationship are classified as a connected area, that is, a separate impurity area. A unique label value is assigned to each impurity area, and the total number of impurities and the area of ​​each impurity are further counted.

[0022] In the above technical solution, the image to be identified is a grayscale image;

[0023] The method for extracting the thin strip length feature from the image to be identified includes:

[0024] The stripe structure of the image to be identified is extracted using the Gabor filter, and the texture features of the image are extracted at the set frequency and direction; the Gabor filters in different directions are applied to the image to be identified, the response maps in each direction are obtained, and the absolute values ​​are taken to merge to obtain the enhanced response map;

[0025] The obtained enhanced response map is subjected to a binarization threshold segmentation, and then the binarized strip area is subjected to a closing operation to obtain a final strip structure area, and the length distribution of the strip structure is statistically calculated; wherein the strip structure is a white cotton area that is slender and curved;

[0026] Count the length distribution of the strip structures in the image to be identified, and calculate the average length L of the thin strip length feature avg and length variance L var :

[0027]

[0028] Among them, L j represents the length of the j-th strip structure; N strip Indicates the number of strip structures.

[0029] In the above technical solution, the image to be identified is a grayscale image;

[0030] The method for extracting structural density features from the image to be identified includes:

[0031] Different directional filters are used to extract the silk thread structure and block structure in the image to be identified; wherein, the silk thread structure is extracted by using the canny operator to obtain the edge map of the silk thread structure area, the image is binarized, and the area of ​​the silk thread structure area is counted; the block structure is extracted by using the Gaussian Laplace operator to obtain the LOG response map, the zero crossing point is found in the LOG response map, the block contour is obtained, the binarized contour edge is binarized and closed to obtain a closed block contour, the closed block contour is filled to obtain the block area, and the area of ​​the block structure area is counted;

[0032] Calculate the density C of the silk thread structure area in the image to be identified line and the density C of the blocky structure region block :

[0033]

[0034] Among them, A line Represents the area of ​​the wire structure region; A block Represents the area of ​​the block structure region; Atotal Represents the total area of ​​the image;

[0035] The structural density characteristics are calculated according to the set weight values ​​ω1, ω2, ω3:

[0036] D structure density =ω1·C line +ω2·C block +ω3·L avg

[0037] Among them, D structure density Represents the structural density characteristics.

[0038] In the above technical solution, the image to be identified is a grayscale image;

[0039] The method for extracting edge fuzziness features from the image to be identified includes:

[0040] The Sobel operator is used to calculate the gradient amplitude G(x,y) of the image to be identified, and the edge blur feature V is statistically calculated. edge :

[0041]

[0042] Among them, N width Indicates the number of pixels in the width of the image to be identified, a is the number of pixels in the width direction of the image to be identified;

[0043] The edge fuzziness uses the average absolute value of the edge gradient change as a measure of fuzziness, which can reflect the clarity of the waste boundary to distinguish between waste types with fuzzy edges and those with clear edges.

[0044] In the above technical solution, the image to be identified is an image in HSV space;

[0045] The method for extracting the white cotton area ratio feature from the image to be identified includes:

[0046] Set the color segmentation threshold to the portion of cotton solid waste that is close to white: h min ,h max =0,255;s min ,s max =0,10;v min ,v max =0,255;

[0047] Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I white , and count the white pixel area A of the segmented binary image white ;

[0048] Calculate the white cotton area ratio C white :

[0049]

[0050] Among them, A total Represents the total area of ​​the image.

[0051] In the above technical solution, the image to be identified is an image in HSV space;

[0052] The method for extracting the area ratio feature of the yellow debris region from the image to be identified includes:

[0053] Set the color segmentation threshold to the portion of cotton solid waste close to the yellow-brown debris: h min ,h max =0,80;s min ,s max =40,255;v min , v max =60,255;

[0054] Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I yellow , and count the yellow pixel area A of the segmented binary image yellow ;

[0055] Calculate the area ratio of yellow debris area C vellow :

[0056]

[0057] Among them, A total Represents the total area of ​​the image.

[0058] In the above technical solution, the image to be identified is a grayscale image;

[0059] The method for extracting texture energy response features from the image to be identified includes:

[0060] Reduce the grayscale of the image to be identified to 16 levels, define the initialization parameters, count the frequency of grayscale m and n appearing simultaneously in the specified direction and distance, form the grayscale co-occurrence matrix GLCM, divide each element in GLCM by the sum, and obtain the normalized matrix P(m,n);

[0061] Compute the texture energy response:

[0062]

[0063] Among them, Energy represents the texture energy response characteristics;

[0064] Among them, the texture energy response characteristics can reflect the uniformity of grayscale distribution on the solid waste texture and the measurement of texture coarseness and fineness. Different solid wastes have different uniformities in grayscale distribution.

[0065] In the above technical solution, the image to be identified is a grayscale image;

[0066] The method for extracting wrinkle edge distribution features from the image to be identified includes:

[0067] Use the Sobel operator to perform edge detection in the x direction on the image to be identified, set the binarization threshold, convert the edge image into a binary image, set the edge pixel size threshold, remove objects smaller than the set size threshold, extract the connected domain area of ​​the edge image, and perform connected domain analysis;

[0068] Calculate the area of ​​the minimum enclosing rectangle of each edge. If the ratio of the area of ​​the edge to the area of ​​the corresponding minimum enclosing rectangle is greater than the set value, the area is considered to be a valid wrinkle edge, and the label of the wrinkle edge is retained. Count the area A of all wrinkle edges whose area is greater than the set value. wrinkle And the number N wrinkle , and obtain the wrinkle edge distribution characteristics C wrinkle :

[0069]

[0070] Among them, the wrinkle edge distribution characteristics are used to reflect the overall texture distribution law in the entire solid waste image.

[0071] In the above technical solution, the feature vector is input into the random forest classifier for training and classification to obtain the recognition result of cotton solid waste, including:

[0072] Construct multiple decision trees, each of which randomly selects a certain number of feature subsets for training, and outputs the final classification result through a voting mechanism; among them, the final output result is obtained by comprehensive voting through the calculation of all trees;

[0073] Among them, the expression of the random forest classifier is:

[0074]

[0075] in, Represents the final classification result; argmax g∈G Indicates that the category with the highest number of votes is selected as the final prediction value; T represents the total number of decision trees; I represents the indicator function, which is used to determine whether the prediction is correct; f t(F) represents the prediction result of the t-th tree for the feature vector F; g represents the type of cotton solid waste corresponding to the current prediction result; G represents the types of all cotton solid wastes.

[0076] In summary, due to the adoption of the above technical features, the beneficial effects of the present invention are:

[0077] Compared with the existing cotton solid waste classification method, the present invention transforms the classification criteria into actual material properties, making the classification method more in line with actual application scenarios. It pays more attention to the physical form of the waste and the actual application needs, and improves the accuracy and applicability of the classification.

[0078] Specifically, the present invention extracts multiple features of the textile solid waste image to be identified according to the specific color and morphology of each classification category, including: impurity coverage, thin strip length features, structural density, edge fuzziness, white cotton area, yellow debris area, texture energy response, wrinkle edge distribution, etc., and then combines all features into a feature vector, inputs it into a random forest classifier, and outputs the final classification result. The method provided by the present invention for automatically classifying new categories of textile solid waste based on multi-dimensional feature extraction and using random forests performs targeted feature extraction on some specific categories of cotton solid waste from aspects such as color, morphology and structural features, comprehensively characterizes the visual characteristics of new categories of cotton solid waste, and can classify solid waste more efficiently and accurately in combination with a random forest classifier.

[0079] Additional aspects and advantages of the present invention will become apparent from the following description or may be learned by practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] The above and / or additional aspects and advantages of the present invention will become apparent and easily understood from the description of the embodiments in conjunction with the following drawings, in which:

[0081] Figure 1 is a flow chart of a method for identifying and classifying textile waste according to an embodiment of the present invention;

[0082] Figure 2 It is a framework diagram of a textile waste identification and classification method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0083] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0084] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Therefore, the protection scope of the present invention is not limited to the specific embodiments disclosed below.

[0085] Refer to the following Figure 1 and Figure 2 The following describes the textile waste identification and classification methods provided according to some embodiments of the present invention.

[0086] Some embodiments of the present application provide a method for identifying and classifying textile waste.

[0087] like Figure 1 and Figure 2 As shown, the first embodiment of the present invention proposes a method for identifying and classifying textile waste, including steps S1-S3.

[0088] S1. Obtain an image of cotton solid waste, and obtain an image to be identified after image processing.

[0089] Specifically, the purpose of step S1 is to preprocess the acquired cotton solid waste image, including but not limited to denoising, normalization, etc., to remove noise interference and obtain a higher quality input image. The image to be identified can be converted into a grayscale image or an image in HSV space according to the requirements of feature extraction.

[0090] S2. Extract features of the image to be identified, wherein the extracted features include impurity distribution density, impurity coverage, thin strip length characteristics, structure density, edge fuzziness, white cotton area, yellow debris area, texture energy response and wrinkle edge distribution.

[0091] S3. Combine all the extracted features into a feature vector, input the feature vector into a random forest classifier for training and classification, and obtain the recognition result of cotton solid waste.

[0092] In the present disclosure, the types of cotton solid waste include broken seed cotton, dusty cotton, sizing waste yarn, natural color waste yarn, blended waste, bellows waste and natural color flying waste. It should be noted that the above types of patterns are all well known to those skilled in the art, and have clear classification boundaries under manual operation.

[0093] Furthermore, the method for extracting each feature in step S2 includes the following steps S21-S28. It should be noted that the order of extracting each feature is not limited in the present disclosure and can be adjusted arbitrarily according to actual needs. In addition, the existing method for extracting corresponding features can also be applied in the present disclosure. The following extraction method is the preferred method provided by the present disclosure.

[0094] S21, the image to be identified is a grayscale image; the method for extracting impurity coverage characteristics and impurity distribution density characteristics from the image to be identified includes:

[0095] Adaptive threshold binarization algorithm is used to extract impurity areas other than cotton from the image to be identified, such as impurities in dregs and seeds; connected domain analysis is used to identify independent impurities;

[0096] Calculate the impurity distribution density and impurity coverage:

[0097]

[0098] Where W represents the width of the image to be identified; H represents the height of the image to be identified; N particle Indicates the total amount of impurities; D particle Indicates the density of impurity distribution; A i represents the area of ​​the i-th impurity; C particle Indicates impurity coverage.

[0099] In some embodiments, the connected domain analysis includes:

[0100] The connectivity relationship between pixels is set; in a specific embodiment, the connectivity relationship between pixels is set to 8 connectivity, that is, the adjacency in the up, down, left, right, and diagonal directions is considered, which is more in line with the irregular shape of impurities.

[0101] The non-background pixels in the binary image are traversed. According to the connectivity between pixels, those that meet the connectivity relationship are classified as a connected area, that is, a separate impurity area. A unique label value is assigned to each impurity area, and the total number of impurities and the area of ​​each impurity are further counted.

[0102] S22, the image to be identified is a grayscale image; the method for extracting the thin strip length feature from the image to be identified includes:

[0103] The stripe structure of the image to be identified is extracted using a Gabor filter, and the texture features of the image are extracted at a set frequency and direction; the Gabor filters of different directions are applied to the image to be identified, and the response maps of each direction are obtained, and the absolute values ​​are taken to merge to obtain an enhanced response map. In a specific embodiment, the parameters of the Gabor filter are defined as follows: the wavelength λ is set to 4, which can exclude smaller noise areas; the standard deviation is set to 2, the spatial aspect ratio γ is set to 0.5, the relative offset ψ is set to π / 2, and the direction θ selects multiple directions, including 0°, 45°, 90°, and 135°, to cover all possible stripe directions.

[0104] The obtained enhanced response map is subjected to binarization threshold segmentation, and then the binarized strip area is closed to obtain the final strip structure area, and the length distribution of the strip structure is statistically calculated; wherein the strip structure is, for example, a slender and curved white cotton area in broken seed cotton and dusty cotton.

[0105] Count the length distribution of the strip structures in the image to be identified, and calculate the average length L of the thin strip length features avg and length variance L var :

[0106]

[0107] Among them, L j represents the length of the j-th strip structure; N strip Indicates the number of strip structures.

[0108] S23, the image to be identified is a grayscale image; the method for extracting structural density features from the image to be identified includes:

[0109] Different directional filters are used to extract the thread structure and block structure in the image to be recognized.

[0110] The canny operator is used to extract the silk thread structure, and the edge map of the silk thread structure area is obtained. The image is binarized and the area of ​​the silk thread structure area is counted. Specifically, the canny operator is used to calculate the gradient amplitude of the image, and the pixels with the largest gradient amplitude are retained to refine the edges. The higher threshold is set to 150 and the lower threshold is set to 50. The edges greater than the high threshold are set as strong edges, and the edges between the thresholds are set as weak edges. The strong edges and weak edges are connected to obtain the edge map of the silk thread structure area, and the image is binarized to count the area of ​​the silk thread structure area.

[0111] The block structure is extracted using the Laplacian of Gaussian (LOG) operator to obtain a LOG response graph, and the zero crossing point is found in the LOG response graph to obtain a block contour. The binarized contour edge is binarized and closed to obtain a closed block contour. The closed block contour is filled to obtain a block area, and the area of ​​the block structure area is counted.

[0112] Calculate the density C of the silk thread structure area in the image to be identified line and the density C of the blocky structure region block :

[0113]

[0114] Among them, A line Represents the area of ​​the wire structure region; A block Represents the area of ​​the block structure region; Atotal Represents the total area of ​​the image;

[0115] The structural density characteristics are calculated according to the set weight values ​​ω1, ω2, ω3:

[0116] D structure density =ω1·C line +ω2·C block +ω3·L avg

[0117] Among them, D structure density Represents the structural density characteristics.

[0118] S24, the image to be identified is a grayscale image; the method for extracting edge fuzziness features from the image to be identified includes:

[0119] The Sobel operator is used to calculate the gradient amplitude G(x,y) of the image to be identified, and the edge blur feature V is statistically calculated. edge :

[0120]

[0121] Among them, N width Indicates the number of pixels in the width of the image to be identified, a is the number of pixels in the width direction of the image to be identified;

[0122] Among them, the edge blur uses the average absolute value of the edge gradient change as a measure of blur, which can reflect the clarity of the waste boundary to distinguish between waste types with blurred edges and clear edges. Specifically, the edges of blocky solid wastes such as bellflower are mostly blurred, in contrast to the clear edges of thread-like wastes.

[0123] S25, the image to be identified is an image in HSV space; the method for extracting the area ratio feature of the white cotton region from the image to be identified includes:

[0124] Set the color segmentation threshold to the portion of cotton solid waste that is close to white: h min ,h max =0,255;s min ,s max =0,10;v min ,v max =0,255;

[0125] Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I white , and count the white pixel area A of the segmented binary image white ;

[0126] Calculate the white cotton area ratio C white :

[0127]

[0128] Among them, A total Represents the total area of ​​the image.

[0129] S26, the image to be identified is an image in HSV space; the method for extracting the area ratio feature of the yellow debris region from the image to be identified includes:

[0130] Set the color segmentation threshold to the portion of cotton solid waste close to the yellow-brown debris: h min ,h max =0,80;s min ,s max =40,255;v min , v max =60,255;

[0131] Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I yellow , and count the yellow pixel area A of the segmented binary image yellow ;

[0132] Calculate the area ratio of yellow debris area C vellow :

[0133]

[0134] Among them, A total Represents the total area of ​​the image.

[0135] S27, the image to be identified is a grayscale image; the method for extracting texture energy response features from the image to be identified includes:

[0136] The grayscale of the image to be identified is reduced to 16 levels, and initialization parameters are defined. The initialization parameters include distance and angle. In a specific embodiment, during initialization, the distance d=1 and the angle θ=45°; the frequency of simultaneous occurrence of grayscale levels m and n in the specified direction and distance is counted to form a grayscale co-occurrence matrix GLCM, and each element in the GLCM is divided by the sum to obtain a normalized matrix P(m,n);

[0137] Compute the texture energy response:

[0138]

[0139] Among them, Energy represents the texture energy response characteristics;

[0140] Among them, the texture energy response characteristics can reflect the uniformity of grayscale distribution and the measurement of texture coarseness on the solid waste texture. Different solid wastes have different uniformities of grayscale distribution. Specifically, the uniformity of grayscale distribution of different solid wastes is different. For example, the silk thread type is more uniform, and the slag seed type is not so uniform. At the same time, the coarseness of their edge textures is also different.

[0141] S28, the image to be identified is a grayscale image; the method for extracting wrinkle edge distribution features from the image to be identified includes:

[0142] Use the Sobel operator to perform edge detection in the x direction on the image to be identified, set a binarization threshold, such as 50, to convert the edge image into a binary image, set an edge pixel size threshold, such as 150, to remove objects smaller than the set size threshold and reduce invalid noise interference; extract the connected domain area of ​​the edge image and perform connected domain analysis. The analysis method can be the same as that in S21.

[0143] Calculate the area of ​​the minimum enclosing rectangle of each edge. If the ratio of the area of ​​the edge to the area of ​​the corresponding minimum enclosing rectangle is greater than the set value, the area is considered to be a valid wrinkle edge, and the label of the wrinkle edge is retained. Count the area A of all wrinkle edges whose area is greater than the set value. wrinkle And the number N wrinkle , and obtain the wrinkle edge distribution characteristics C wrinkle :

[0144]

[0145] Among them, the wrinkle edge distribution feature is used to reflect the overall texture distribution law in the entire solid waste image. Specifically, the wrinkle edge distribution feature is used to extract the relatively slender and complete areas in the edge, which reflect the overall texture distribution law in the entire solid waste image. For example, the edges of silk threads are longer, and the edges of dregs are shorter and denser, reflecting the shape of cotton in the image.

[0146] After the feature extraction of S21-S28, the feature vector can be obtained after combination: F =

[0147] [D particle ,C particle ,L avg ,L var ,C line ,C block ,V edge ,C white ,C yellow ,Energy,C wrinkle ].

[0148] In step S3, the feature vector is input into a random forest classifier for training and classification to obtain the recognition result of cotton solid waste, including:

[0149] Construct multiple decision trees, each of which randomly selects a certain number of feature subsets for training, and outputs the final classification result through a voting mechanism; among them, the final output result is obtained by comprehensive voting through the calculation of all trees;

[0150] Among them, the expression of the random forest classifier is:

[0151]

[0152] in, Represents the final classification result; argmax g∈G Indicates that the category with the highest number of votes is selected as the final prediction value; T represents the total number of decision trees; I represents the indicator function, which is used to determine whether the prediction is correct; f t (F) represents the prediction result of the t-th tree for the feature vector F; g represents the type of cotton solid waste corresponding to the current prediction result; G represents the types of all cotton solid wastes.

[0153] For example, in a specific embodiment, a feature vector of a solid waste image is input to a random forest classifier, and one decision tree selects four features: impurity coverage, white cotton area ratio, yellow debris area ratio, and structural density. The structural density can be used for learning to distinguish between slag seeds, silk threads, and blocks. The impurity coverage can be further refined to learn to distinguish between slag seeds and other categories. In addition, the tree's branch white cotton area ratio features and yellow debris area ratio can further distinguish between broken seed cotton and dusty slag cotton, as well as other categories. This tree can effectively classify broken seed cotton and dusty slag cotton. Selecting other feature subsets can effectively classify other categories. Finally, all trees are calculated and voted on to obtain the final output result.

[0154] In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0155] 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 identifying and classifying textile waste, characterized in that: include: Acquire cotton solid waste images, and obtain the image to be identified after image processing; Extracting features of the image to be identified, wherein the extracted features include impurity distribution density, impurity coverage, thin strip length characteristics, structure density, edge fuzziness, white cotton area, yellow debris area, texture energy response, and wrinkle edge distribution; All the extracted features are combined into a feature vector, and the feature vector is input into a random forest classifier for training and classification to obtain the recognition result of cotton solid waste; Among them, the types of cotton solid waste include broken seed cotton, dusty cotton, sizing waste yarn, natural color waste yarn, blended waste, bellows waste and natural color flying waste.

2. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is a grayscale image; The method for extracting impurity coverage characteristics and impurity distribution density characteristics from the image to be identified includes: Adaptive threshold binarization algorithm is used to extract impurity areas other than cotton from the image to be identified, and connected domain analysis is used to identify independent impurities. Calculate the impurity distribution density and impurity coverage: Where W represents the width of the image to be identified; H represents the height of the image to be identified; N particle Indicates the total amount of impurities; D particle Indicates the density of impurity distribution; A i represents the area of ​​the i-th impurity; C particle represents the impurity coverage; Wherein, the connected domain analysis includes: Set the connectivity between pixels; The non-background pixels in the binary image are traversed. According to the connectivity between pixels, those that meet the connectivity relationship are classified as a connected area, that is, a separate impurity area. A unique label value is assigned to each impurity area, and the total number of impurities and the area of ​​each impurity are further counted.

3. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is a grayscale image; The method for extracting the thin strip length feature from the image to be identified includes: The stripe structure of the image to be identified is extracted using the Gabor filter, and the texture features of the image are extracted at the set frequency and direction; the Gabor filters in different directions are applied to the image to be identified, the response maps in each direction are obtained, and the absolute values ​​are taken to merge to obtain the enhanced response map; The obtained enhanced response map is subjected to a binarization threshold segmentation, and then the binarized strip area is subjected to a closing operation to obtain a final strip structure area, and the length distribution of the strip structure is statistically calculated; wherein the strip structure is a white cotton area that is slender and curved; Count the length distribution of the strip structures in the image to be identified, and calculate the average length L of the thin strip length features avg and length variance L var : Among them, L j represents the length of the j-th strip structure; N strip Indicates the number of strip structures.

4. The textile waste identification and classification method according to claim 2, characterized in that: The image to be identified is a grayscale image; The method for extracting structural density features from the image to be identified includes: Different directional filters are used to extract the silk thread structure and block structure in the image to be identified; wherein, the silk thread structure is extracted by using the canny operator to obtain the edge map of the silk thread structure area, the image is binarized, and the area of ​​the silk thread structure area is counted; the block structure is extracted by using the Gaussian Laplace operator to obtain the LOG response map, the zero crossing point is found in the LOG response map, the block contour is obtained, the binarized contour edge is binarized and closed to obtain a closed block contour, the closed block contour is filled to obtain the block area, and the area of ​​the block structure area is counted; Calculate the density C of the silk thread structure area in the image to be identified line and the density C of the blocky structure region block : Among them, A line Represents the area of ​​the wire structure region; A block Represents the area of ​​the block structure region; A total Represents the total area of ​​the image; The structural density characteristics are calculated according to the set weight values ​​ω1, ω2, ω3: D structure density =ω1·C line +ω2·C block +ω3·L avg Among them, D structure density Represents the structural density characteristics.

5. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is a grayscale image; The method for extracting edge fuzziness features from the image to be identified includes: The Sobel operator is used to calculate the gradient amplitude G(x, y) of the image to be identified, and the edge blur feature V is statistically calculated. edge : Among them, N width Indicates the number of pixels in the width of the image to be identified, a is the number of pixels in the width direction of the image to be identified; The edge fuzziness uses the average absolute value of the edge gradient change as a measure of fuzziness, which can reflect the clarity of the waste boundary to distinguish between waste types with fuzzy edges and those with clear edges.

6. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is an image in HSV space; The method for extracting the white cotton area ratio feature from the image to be identified includes: Set the color segmentation threshold to the portion of cotton solid waste that is close to white: h min ,h max =0,255;s min ,s max =0,10;v min , v max =0,255; Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I white , and count the white pixel area A of the segmented binary image white ; Calculate the white cotton area ratio C white : Among them, A total Represents the total area of ​​the image.

7. The method for identifying and classifying textile waste according to claim 1, characterized in that: The image to be identified is an image in HSV space; The method for extracting the area ratio feature of the yellow debris region from the image to be identified includes: Set the color segmentation threshold to the portion of cotton solid waste close to the yellow-brown debris: h min ,h max =0,80;s min ,s max =40,255;v min , v max =60,255; Use the set threshold to generate an image mask and perform color segmentation to obtain the segmented image I yellow , and count the yellow pixel area A of the segmented binary image yellow ; Calculate the area ratio of yellow debris area C yellow : Among them, A total Indicates the total area of ​​the image.

8. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is a grayscale image; The method for extracting texture energy response features from the image to be identified includes: Reduce the grayscale of the image to be identified to 16 levels, define the initialization parameters, count the frequency of grayscale m and n appearing simultaneously in the specified direction and distance, form the gray level co-occurrence matrix GLCM, divide each element in GLCM by the sum, and obtain the normalized matrix P(m, n); Compute the texture energy response: Among them, Energy represents the texture energy response characteristics; Among them, the texture energy response characteristics can reflect the uniformity of grayscale distribution on the solid waste texture and the measurement of texture coarseness and fineness. Different solid wastes have different uniformities in grayscale distribution.

9. The textile waste identification and classification method according to claim 1, characterized in that: The image to be identified is a grayscale image; The method for extracting wrinkle edge distribution features from the image to be identified includes: Use the Sobel operator to perform edge detection in the x direction on the image to be identified, set the binarization threshold, convert the edge image into a binary image, set the edge pixel size threshold, remove objects smaller than the set size threshold, extract the connected domain area of ​​the edge image, and perform connected domain analysis; Calculate the area of ​​the minimum enclosing rectangle of each edge. If the ratio of the area of ​​the edge to the area of ​​the corresponding minimum enclosing rectangle is greater than the set value, the area is considered to be a valid wrinkle edge, and the label of the wrinkle edge is retained. Count the area A of all wrinkle edges whose area is greater than the set value. wrinkle And the number N wrinkle , and obtain the wrinkle edge distribution characteristics C wrinkle : Among them, the wrinkle edge distribution characteristics are used to reflect the overall texture distribution law in the entire solid waste image.

10. The textile waste identification and classification method according to claim 1, characterized in that: The feature vector is input into the random forest classifier for training and classification to obtain the recognition result of cotton solid waste, including: Construct multiple decision trees, each of which randomly selects a certain number of feature subsets for training, and outputs the final classification result through a voting mechanism; among them, the final output result is obtained by comprehensive voting through the calculation of all trees; Among them, the expression of the random forest classifier is: in, Represents the final classification result; argmax g∈G Indicates that the category with the highest number of votes is selected as the final prediction value; T represents the total number of decision trees; I represents the indicator function, which is used to determine whether the prediction is correct; f t (F) represents the prediction result of the t-th tree for the feature vector F; g represents the type of cotton solid waste corresponding to the current prediction result; G represents the types of all cotton solid wastes.

Images