Intelligent identification and sorting method for waste textiles

Abstract

The invention discloses an intelligent identification and sorting method for waste textiles, and belongs to the technical field of textile recovery. An intelligent identification and sorting method for waste textiles comprises the steps of scanning recycled waste textiles by using a hyperspectral camera, obtaining hyperspectral image data of the waste textiles, preprocessing the obtained hyperspectral image data to improve data quality, constructing a material identification model, inputting dimension-reduced hyperspectral features into the model, and identifying the waste textiles according to the dimension-reduced hyperspectral features. Through training and learning spectral feature differences of different materials, the materials of the waste textiles are accurately identified; recording the surface condition of the waste spinning through a monitoring device, and analyzing the influence of the surface condition on spectrum identification; according to the material recognition result and the surface recognition result, sorting equipment is controlled to carry out sorting operation on polyester textiles in the waste textiles; a sorting result is fed back to a control system, the sorting accuracy rate is recorded, and the performance of the recognition model and the sorting equipment is evaluated and optimized.

Description

Technical Field

[0001] The present invention belongs to the technical field of textile recycling, and more specifically, relates to an intelligent identification and sorting method for waste textiles. Background Art

[0002] With the rapid development of the economy and the improvement of people's living standards, the amount of waste textiles generated has increased year by year. Polyester is a synthetic fiber widely used in textiles. It has the advantages of high strength, wear resistance, easy washing and quick drying, and has a high recycling value. The traditional method of processing waste textiles is mainly based on manual sorting, which is not only labor-intensive and inefficient, but also has low sorting accuracy, which is difficult to meet the needs of modern environmental protection and resource recycling.

[0003] Most of the existing methods for processing waste textiles use infrared spectroscopy recognition after pretreatment and then use automated robotic arms for sorting, which improves the efficiency and accuracy of sorting to a certain extent; however, the technical recognition accuracy may be affected by many factors, such as the moisture, color, stains and complexity of mixed fibers of textiles, which may cause errors in the recognition results and affect the subsequent sorting and reuse effects; taking an integrated intelligent garbage sorting and delivery device and delivery method (CN202111357510.0) disclosed in a Chinese invention patent document as an example, the invention obtains surface texture attribute information and surface morphology attribute information of the textile by surface visual recognition of the textile, so as to accurately screen recyclable textiles and improve the accuracy of screening; however, the invention only performs preliminary screening of textiles based on surface texture, and cannot avoid the influence of some undetected stains, wrinkles and damaged areas on subsequent infrared spectroscopy recognition due to missed detection, and at the same time, some trace stains, wrinkles and damage may cause waste of waste textiles due to failure to pass the screening.

[0004] Therefore, we need to develop an intelligent identification and sorting method for waste textiles with high recognition accuracy and more reasonable recycling. Summary of the invention

[0005] The purpose of the present invention is to provide an intelligent identification and sorting method for waste textiles, which has high identification accuracy and can more accurately identify recyclable waste textiles by combining the material identification results and surface identification results of the waste textiles, thereby avoiding the waste of waste textiles.

[0006] To achieve the above object, the present invention provides the following technical solutions: The present invention provides a method for intelligently identifying and sorting waste textiles, comprising the following steps: S1. Data collection: using a hyperspectral camera to scan the recycled waste textiles and obtain hyperspectral image data of the waste textiles; S2. Feature extraction and dimensionality reduction: preprocess the acquired hyperspectral image data, including noise removal and correction operations to improve data quality, and then use a method combining principal component analysis and autoencoder technology to extract features and reduce the dimensionality of the hyperspectral data; S3. Material recognition: construct a material recognition model, input the hyperspectral features after dimensionality reduction into the model, learn the spectral feature differences of different materials through training, and accurately identify the materials of waste textiles; S4. Surface identification, recording the surface condition of the waste textile through the monitoring device and analyzing its influence on spectral identification; S5. Sorting control, controlling the sorting equipment to perform sorting operations on polyester textiles in waste textiles according to the material recognition results and the surface recognition results; S6. Feedback and optimization of results: Feedback the sorting results to the control system, record the sorting accuracy, and evaluate and optimize the performance of the recognition model and sorting equipment.

[0007] As a further improvement of the present invention, step S1 collects several samples with polyester contents ranging from 0-100% in sequence, and collects data at 5 positions for each sample, so as to provide rich spectral features for subsequent identification and analysis.

[0008] As a further improvement of the present invention, the principal component analysis in step S2 is used to extract the main components in the data and reduce the data dimension, while the autoencoder further optimizes the feature expression and retains the key spectral feature information.

[0009] As a further improvement of the present invention, in the material identification described in step S3, the high-spectral characteristic peaks of the polyester material content in the waste textiles greater than 60% are mainly at 1130nm and 1660nm. This band is selected to accelerate data processing and reduce material identification time.

[0010] As a further improvement of the present invention, the surface condition of the waste textile in step S4 is represented by waste textile surface characteristic parameters, and the waste textile surface characteristic parameters are affected by stains, wrinkles and damage on the waste textile surface.

[0011] As a further improvement of the present invention, the calculation formula of the waste spinning surface characteristic parameter is: waste spinning surface characteristic parameter , where α is the influence coefficient of stains on reflectivity, indicating the proportion of reflectivity reduction caused by stains; S indicates the severity of stains, ranging from 0 to 1, and increases with the severity of stains; β is the influence coefficient of wrinkles on reflectivity, indicating the proportion of reflectivity reduction caused by wrinkles; W indicates the severity of wrinkles, ranging from 0 to 1, and increases with the severity of wrinkles; γ is the influence coefficient of damage on reflectivity, indicating the proportion of reflectivity reduction caused by damage; D indicates the severity of damage, ranging from 0 to 1, and increases with the severity of damage; θ is the influence coefficient of ambient light on reflectivity, indicating the proportion of reflectivity reduction caused by ambient light dimming; T indicates the severity of ambient light dimming, ranging from 0 to 1, and increases with the severity of ambient light dimming.

[0012] As a further improvement of the present invention, the material identification result and the surface identification result in step S5 are presented through the final reflectivity R. The final reflectivity has a characteristic peak, that is, the final reflectivity R=R0×V, wherein V represents the surface characteristic parameters of the waste textile, and R0 represents the directly measured reflectivity.

[0013] As a further improvement of the present invention, the recyclable waste textiles described in step S5 are separated by air blowing or a robot sorting device to obtain a primary sorted product, and the primary sorted product is cut and fed and then sorted again to obtain a secondary sorted product, and the secondary sorted product is recycled.

[0014] As a further improvement of the present invention, the sorting accuracy in step S6 is the ratio of the number of waste textiles whose polyester content percentage meets the sorting requirements to the total number of sorted waste textiles; by analyzing the changing trend of the sorting accuracy for five consecutive times, a basis is provided for the setting and optimization of subsequent stain parameters, wrinkle parameters, damage parameters or ambient light parameters.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: by constructing a material recognition model to identify the material of waste textiles and combining the monitoring device to identify the surface of waste textiles, the influence of the textile surface on the infrared spectrum of the textile material is reduced; by recording the size of the stain to change the characteristic wave peak value of the waste textiles, it is avoided that the scattered light caused by the stains makes the characteristic wave peak of the waste textiles difficult to identify; by recording the size of the wrinkles to change the characteristic wave peak value of the waste textiles, it is avoided that the spectral data at the wrinkles is greatly different from the spectral data at the flat part, which makes the characteristic wave peak of the waste textiles unclear; by recording the size of the damage to change the characteristic wave peak value of the waste textiles, it is avoided that the spectral data at the damage is discontinuous, which makes the spectral data at the damage difficult to identify. The characteristic peaks of fabrics are inconsistent, thereby avoiding the influence of surface parameters of waste textiles on material identification, improving identification accuracy and avoiding material waste; by combining principal component analysis with autoencoder technology to extract and reduce the dimensions of hyperspectral data, the feature expression can be further optimized, the key spectral feature information can be retained, and the accuracy of identification can be improved; the recyclable waste textiles are separated and sent to the recycling process through secondary sorting by air blowing or manipulator sorting devices. The automated processing method improves the sorting efficiency and the quality of the secondary sorted textiles; by analyzing the changing trend of the sorting accuracy, a basis is provided for the subsequent setting and optimization of stain parameters, wrinkle parameters, damage parameters or ambient light parameters, so that the sorting accuracy can remain stable. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 A schematic diagram of a process flow of an intelligent identification and sorting method for waste textiles according to the present invention; Figure 2 It is a schematic diagram of characteristic peaks of polyester reflectivity in a method for intelligent identification and sorting of waste textiles of the present invention. DETAILED DESCRIPTION

[0017] Specific embodiment 1: Please refer to Figure 1-Figure 2 A method for intelligent identification and sorting of waste textiles, comprising:

[0018] S1. Data collection: Use a hyperspectral camera to scan the recycled waste textiles and obtain hyperspectral image data of the waste textiles.

[0019] Specifically, in step S1, the hyperspectral camera can capture reflectance information at different wavelengths; by collecting a number of samples with polyester content ranging from 0-100%, each sample collects data at 5 positions, providing rich spectral features for subsequent identification and analysis.

[0020] S2. Feature extraction and dimensionality reduction, the acquired hyperspectral image data is preprocessed, including noise removal and correction operations to improve data quality, and then the method combining principal component analysis and autoencoder technology is used to extract features and reduce the dimensionality of the hyperspectral data.

[0021] Specifically, the principal component analysis in step S2 is used to extract the main components in the data and reduce the data dimension, while the autoencoder further optimizes the feature expression and retains the key spectral feature information.

[0022] Specifically, principal component analysis is performed by centering the spectral data, calculating the covariance matrix, calculating the eigenvalues ​​and eigenvectors of the covariance matrix, selecting the eigenvectors corresponding to the first k largest eigenvalues ​​as principal components, and projecting the original data onto the principal components to obtain the reduced-dimensional data.

[0023] Specifically, the autoencoder consists of an encoder and a decoder. The encoder compresses the input data into low-dimensional features, and the decoder restores the low-dimensional features to the input data. By minimizing the reconstruction error, the autoencoder is trained so that the low-dimensional features can restore the original data as much as possible.

[0024] S3. Material recognition: build a material recognition model, input the reduced-dimensional hyperspectral features into the model, learn the differences in spectral features of different materials through training, and accurately identify the materials of waste textiles.

[0025] Specifically, Figure 2 As shown, in the material identification described in step S3, the high-spectral characteristic peaks of the polyester material content in the waste textiles greater than 60% are mainly at 1130nm and 1660nm. This band is selected for data acquisition and processing to improve the high-spectral data processing speed and efficiency. By selecting the band acceleration program, the amount of data can be reduced while retaining key spectral features, thereby improving processing speed and efficiency.

[0026] Specifically, the material recognition model is built based on the fusion of convolutional neural network and visual transformer. The convolutional neural network is used to extract local features of waste textiles, and the visual transformer is used to capture global features and long-distance dependencies. The convolutional neural network includes multiple convolutional layers and pooling layers, which are used to gradually extract local features of the image. The visual transformer divides the image into multiple small blocks, and then captures the global features between these blocks through the self-attention mechanism. The local features extracted by the convolutional neural network and the global features extracted by the visual transformer are spliced ​​to form comprehensive features, and these features are further fused through fully connected layers or convolutional layers to extract richer feature representations.

[0027] S4. Surface identification: record the surface condition of the waste textile through the monitoring device and analyze its influence on spectral identification.

[0028] Specifically, the monitoring device is a high-definition camera installed above the waste textile identification assembly line, which can capture the surface condition of the waste textile in real time; the camera records video at a frequency of 30 frames per second with a resolution of 1920×1080 pixels to ensure that the stains, wrinkles and damage on the surface of the waste textile can be clearly captured. For the video data taken by the camera, the OpenCV and NumPy libraries are first used for image preprocessing, and the convolutional neural network algorithm in deep learning is used for image recognition; the image recognition algorithm is used to identify the size of the stains, wrinkles and damage areas, so as to obtain the impact of stains, wrinkles and damage on material identification.

[0029] Specifically, the surface condition of the waste textile in step S4 is represented by waste textile surface characteristic parameters, and the waste textile surface characteristic parameters are affected by stains, wrinkles and damage on the surface of the waste textile.

[0030] Specifically, the calculation formula of the waste spinning surface characteristic parameters is: waste spinning surface characteristic parameters , where α is the coefficient of influence of stains on reflectivity, indicating the proportion of reflectivity reduction caused by stains; S indicates the severity of stains, ranging from 0 to 1, and increases with the severity of stains. No stains are 0, and severe stains are 1; β is the coefficient of influence of wrinkles on reflectivity, indicating the proportion of reflectivity reduction caused by wrinkles; W indicates the severity of wrinkles, ranging from 0 to 1, and increases with the severity of wrinkles. No wrinkles are 0, and severe wrinkles are 1; γ is the coefficient of influence of damage on reflectivity, indicating the proportion of reflectivity reduction caused by damage; D indicates the severity of damage, ranging from 0 to 1, and increases with the severity of damage. No damage is 0, and severe damage is 1; θ is the coefficient of influence of ambient light on reflectivity, indicating the reflectivity caused by the dimming of ambient light Reduction ratio; T represents the severity of the ambient light dimming, ranging from 0 to 1. As the severity of the ambient light dimming increases, normal light is 0 and too dark light is 1. For example, through experiments, it is determined that the reflectivity R0 directly measured by the material is 0.75, the influence coefficient of stains on reflectivity is α=0.3, the influence coefficient of wrinkles on reflectivity is β=0.2, the influence coefficient of damage on reflectivity is γ=0.4, and the influence coefficient of ambient light on reflectivity is θ=0.2; for a polyester product with a stain parameter S=0.5, a moderate stain range, a wrinkle parameter W=0.3, a light wrinkle, a damage parameter D=0.2, a light damage, an ambient light parameter T=0, and a suitable ambient light intensity, the waste surface characteristic parameter V can be calculated as follows: .

[0031] S5. Sorting control, controlling the sorting equipment to perform sorting operations on polyester textiles in waste textiles according to the material recognition results and the surface recognition results.

[0032] Specifically, the material identification result and the surface identification result in step S5 are presented through the final reflectivity R. The final reflectivity has a characteristic peak, that is, the final reflectivity R=R0×V, wherein V represents the surface characteristic parameter of the waste spinning, and V=0.735 is calculated according to step S4, and R0 represents the directly measured reflectivity, R0=0.75; then the calculation formula of the final reflectivity is =0.75×0.735=0.551; at this time, the reflectivity R=0.551 is lower than the reflectivity R0=0.75 directly measured by the material, so that the characteristic peak formed is more obvious.

[0033] Specifically, the recyclable waste textiles in step S5 are separated by air blowing or a robot sorting device to obtain a primary sorted product, and the primary sorted product is cut and fed and sorted again to obtain a secondary sorted product, and the secondary sorted product is recycled.

[0034] S6. Feedback and optimization of results: Feedback the sorting results to the control system, record the sorting accuracy, and evaluate and optimize the performance of the recognition model and sorting equipment.

[0035] Specifically, the sorting accuracy rate in step S6 is the ratio of the number of waste textiles whose polyester content percentage meets the sorting requirements to the total number of waste textiles sorted; the calculation formula is: sorting accuracy rate = the number of waste textiles whose polyester content percentage meets the sorting requirements / the total number of waste textiles sorted × 100%; by analyzing the changing trend of the sorting accuracy rate for five consecutive times, the calculation formula for the average sorting accuracy rate for five consecutive times is: the average sorting accuracy rate for five consecutive times (x1+x2+x3+x4+x5) / 5; where x 1-5 Respectively represent the sorting accuracy obtained from the 1st to the 5th sorting; analyze and calculate the variance of the sorting accuracy for 5 consecutive times, where the variance calculation formula for the sorting accuracy for five consecutive times is: sorting accuracy variance s 2 = 2 ) / 5, where x 1-5 Respectively represent the sorting accuracy obtained from the 1st to 5th sorting, represents the average sorting accuracy of five consecutive times; z is set as the variance threshold, when s 2 When it is less than z, that is, the change range of the accuracy rate of five consecutive sortings is small, and sorting can be continued; when s 2When it is greater than z, that is, the sorting accuracy changes greatly for five consecutive times, first determine whether the sorting accuracy increases or decreases. If it decreases, it is possible that one or more values ​​of the experimentally measured stain parameter S, wrinkle parameter W, damage parameter D or ambient light parameter T are too large. By dynamically reducing the settings of the stain parameter S, wrinkle parameter W, damage parameter D or ambient light parameter T, the waste textile surface characteristic parameter V increases, thereby increasing the final reflectivity R, thereby reducing the difference in characteristic peaks, and avoiding when the sorting accuracy is high. Waste textiles with unqualified material identification can pass the sorting; if it increases, it is possible that one or more of the experimentally measured stain parameter S, wrinkle parameter W, damage parameter D or ambient light parameter T are too small. By dynamically increasing the stain parameter S, wrinkle parameter W, damage parameter D or ambient light parameter T, the surface characteristic parameter V of the waste textile is reduced, and then the final reflectivity R is reduced, thereby increasing the difference in characteristic peaks, so as to avoid waste textiles with qualified material identification from failing to pass the sorting due to the influence of surface characteristic parameters when the sorting accuracy is low.

[0036] The above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for intelligent identification and sorting of waste textiles, characterized in that: The following steps are involved: S1. Data collection: using a hyperspectral camera to scan the recycled waste textiles and obtain hyperspectral image data of the waste textiles; S2. Feature extraction and dimensionality reduction: preprocess the acquired hyperspectral image data, including noise removal and correction operations to improve data quality, and then use a method combining principal component analysis and autoencoder technology to extract features and reduce the dimensionality of the hyperspectral data; S3. Material recognition: construct a material recognition model, input the hyperspectral features after dimensionality reduction into the model, learn the spectral feature differences of different materials through training, and accurately identify the materials of waste textiles; S4. Surface identification, recording the surface condition of the waste textile through the monitoring device and analyzing its influence on spectral identification; S5. Sorting control, controlling the sorting equipment to perform sorting operations on polyester textiles in waste textiles according to the material recognition results and the surface recognition results; S6. Feedback and optimization of results: Feedback the sorting results to the control system, record the sorting accuracy, and evaluate and optimize the performance of the recognition model and sorting equipment.

2. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: Step S1 collects several samples with polyester contents ranging from 0-100% in sequence, and collects data from 5 positions for each sample to provide rich spectral features for subsequent identification and analysis.

3. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: The principal component analysis described in step S2 is used to extract the main components in the data and reduce the data dimension, while the autoencoder further optimizes the feature expression and retains the key spectral feature information.

4. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: In the material identification described in step S3, the high-spectral characteristic peaks of the polyester material content in the waste textiles greater than 60% are at 1130nm and 1660nm. This band is selected to accelerate data processing and reduce material identification time.

5. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: The surface condition of the waste textile in step S4 is represented by the waste textile surface characteristic parameters, and the waste textile surface characteristic parameters are affected by the stains, wrinkles and damage on the waste textile surface.

6. The method for intelligent identification and sorting of waste textiles according to claim 5, characterized in that: The calculation formula of the waste spinning surface characteristic parameter is: waste spinning surface characteristic parameter , where α is the influence coefficient of stains on reflectivity, indicating the proportion of reflectivity reduction caused by stains; S indicates the severity of stains, ranging from 0 to 1, and increases with the severity of stains; β is the influence coefficient of wrinkles on reflectivity, indicating the proportion of reflectivity reduction caused by wrinkles; W indicates the severity of wrinkles, ranging from 0 to 1, and increases with the severity of wrinkles; γ is the influence coefficient of damage on reflectivity, indicating the proportion of reflectivity reduction caused by damage; D indicates the severity of damage, ranging from 0 to 1, and increases with the severity of damage; θ is the influence coefficient of ambient light on reflectivity, indicating the proportion of reflectivity reduction caused by ambient light dimming; T indicates the severity of ambient light dimming, ranging from 0 to 1, and increases with the severity of ambient light dimming.

7. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: The material identification result and the surface identification result in step S5 are presented through the final reflectivity R. The final reflectivity has a characteristic peak, that is, the final reflectivity R=R0×V, wherein V represents the surface characteristic parameter of the waste textile, and R0 represents the directly measured reflectivity.

8. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: In step S5, the waste textiles are separated by air blowing or a robot sorting device to obtain a primary sorted product, and the primary sorted product is cut and fed and sorted again to obtain a secondary sorted product, and the secondary sorted product is recycled.

9. The method for intelligent identification and sorting of waste textiles according to claim 1, characterized in that: The sorting accuracy in step S6 is the ratio of the number of waste textiles whose polyester content percentage meets the sorting requirements to the total number of sorted waste textiles; by analyzing the changing trend of the sorting accuracy for five consecutive times, a basis is provided for the setting and optimization of subsequent stain parameters, wrinkle parameters, damage parameters or ambient light parameters.

Images