CN113970532A - Fabric fiber component detection system and prediction method based on near infrared spectrum - Google Patents
Fabric fiber component detection system and prediction method based on near infrared spectrum Download PDFInfo
- Publication number
- CN113970532A CN113970532A CN202111178122.6A CN202111178122A CN113970532A CN 113970532 A CN113970532 A CN 113970532A CN 202111178122 A CN202111178122 A CN 202111178122A CN 113970532 A CN113970532 A CN 113970532A
- Authority
- CN
- China
- Prior art keywords
- data
- near infrared
- infrared spectrum
- textile
- filtering
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 238000002329 infrared spectrum Methods 0.000 title claims abstract description 48
- 239000000835 fiber Substances 0.000 title claims abstract description 40
- 239000004744 fabric Substances 0.000 title claims abstract description 26
- 238000001514 detection method Methods 0.000 title claims abstract description 16
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 29
- 238000001914 filtration Methods 0.000 claims abstract description 25
- 239000000463 material Substances 0.000 claims abstract description 25
- 238000012549 training Methods 0.000 claims abstract description 20
- 238000004458 analytical method Methods 0.000 claims abstract description 16
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 13
- 230000009467 reduction Effects 0.000 claims abstract description 13
- 238000004140 cleaning Methods 0.000 claims abstract description 10
- 238000002156 mixing Methods 0.000 claims abstract description 8
- 238000000605 extraction Methods 0.000 claims abstract description 7
- 238000005070 sampling Methods 0.000 claims abstract description 6
- 239000004753 textile Substances 0.000 claims description 67
- 244000025254 Cannabis sativa Species 0.000 claims description 10
- 235000012766 Cannabis sativa ssp. sativa var. sativa Nutrition 0.000 claims description 10
- 235000012765 Cannabis sativa ssp. sativa var. spontanea Nutrition 0.000 claims description 10
- 229920000742 Cotton Polymers 0.000 claims description 10
- 229920002334 Spandex Polymers 0.000 claims description 10
- 235000009120 camo Nutrition 0.000 claims description 10
- 235000005607 chanvre indien Nutrition 0.000 claims description 10
- 239000011487 hemp Substances 0.000 claims description 10
- 239000004759 spandex Substances 0.000 claims description 10
- 239000004677 Nylon Substances 0.000 claims description 9
- 210000000085 cashmere Anatomy 0.000 claims description 9
- 229920001778 nylon Polymers 0.000 claims description 9
- 238000007781 pre-processing Methods 0.000 claims description 9
- 210000002268 wool Anatomy 0.000 claims description 9
- 229920000433 Lyocell Polymers 0.000 claims description 8
- 229920000297 Rayon Polymers 0.000 claims description 8
- 229920004933 Terylene® Polymers 0.000 claims description 8
- -1 modal Polymers 0.000 claims description 8
- 239000005020 polyethylene terephthalate Substances 0.000 claims description 8
- 239000002964 rayon Substances 0.000 claims description 8
- 238000001228 spectrum Methods 0.000 claims description 7
- 230000006870 function Effects 0.000 claims description 6
- 229920002972 Acrylic fiber Polymers 0.000 claims description 5
- 238000009499 grossing Methods 0.000 claims description 5
- 238000010801 machine learning Methods 0.000 claims description 5
- 230000003595 spectral effect Effects 0.000 claims description 5
- NIXOWILDQLNWCW-UHFFFAOYSA-N acrylic acid group Chemical group C(C=C)(=O)O NIXOWILDQLNWCW-UHFFFAOYSA-N 0.000 claims description 4
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000004497 NIR spectroscopy Methods 0.000 claims description 3
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 2
- 241000196324 Embryophyta Species 0.000 claims 3
- 238000000354 decomposition reaction Methods 0.000 claims 1
- 238000002310 reflectometry Methods 0.000 abstract description 7
- 238000013135 deep learning Methods 0.000 abstract description 2
- 230000008569 process Effects 0.000 description 9
- 238000012360 testing method Methods 0.000 description 7
- 238000013473 artificial intelligence Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 6
- 238000007635 classification algorithm Methods 0.000 description 5
- 239000000126 substance Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000001965 increasing effect Effects 0.000 description 4
- 239000000203 mixture Substances 0.000 description 4
- 238000005457 optimization Methods 0.000 description 4
- 229920000728 polyester Polymers 0.000 description 4
- 238000013145 classification model Methods 0.000 description 3
- 238000002790 cross-validation Methods 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011176 pooling Methods 0.000 description 3
- 230000001737 promoting effect Effects 0.000 description 3
- ORILYTVJVMAKLC-UHFFFAOYSA-N Adamantane Natural products C1C(C2)CC3CC1CC2C3 ORILYTVJVMAKLC-UHFFFAOYSA-N 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 208000035475 disorder Diseases 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 239000004615 ingredient Substances 0.000 description 2
- 238000010606 normalization Methods 0.000 description 2
- 238000000053 physical method Methods 0.000 description 2
- 238000000611 regression analysis Methods 0.000 description 2
- 239000002699 waste material Substances 0.000 description 2
- 208000037170 Delayed Emergence from Anesthesia Diseases 0.000 description 1
- JHWNWJKBPDFINM-UHFFFAOYSA-N Laurolactam Chemical compound O=C1CCCCCCCCCCCN1 JHWNWJKBPDFINM-UHFFFAOYSA-N 0.000 description 1
- 229920000299 Nylon 12 Polymers 0.000 description 1
- 238000012952 Resampling Methods 0.000 description 1
- 239000012237 artificial material Substances 0.000 description 1
- 238000012512 characterization method Methods 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 238000013434 data augmentation Methods 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 239000006185 dispersion Substances 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 238000003912 environmental pollution Methods 0.000 description 1
- 238000013210 evaluation model Methods 0.000 description 1
- 238000004880 explosion Methods 0.000 description 1
- 238000007477 logistic regression Methods 0.000 description 1
- 238000003333 near-infrared imaging Methods 0.000 description 1
- 238000001579 optical reflectometry Methods 0.000 description 1
- 238000003908 quality control method Methods 0.000 description 1
- 238000011002 quantification Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/25—Colour; Spectral properties, i.e. comparison of effect of material on the light at two or more different wavelengths or wavelength bands
- G01N21/31—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry
- G01N21/35—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light
- G01N21/359—Investigating relative effect of material at wavelengths characteristic of specific elements or molecules, e.g. atomic absorption spectrometry using infrared light using near infrared light
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
- G01N21/55—Specular reflectivity
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N2201/00—Features of devices classified in G01N21/00
- G01N2201/12—Circuits of general importance; Signal processing
- G01N2201/129—Using chemometrical methods
- G01N2201/1296—Using chemometrical methods using neural networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Abstract
The invention provides a fabric fiber component detection system and a prediction method based on near infrared spectrum, wherein the system collects the waveform of the reflectivity and the absorptivity of the near infrared spectrum of a fabric, performs noise reduction through a filtering method of median filtering, mean filtering, wiener filtering and S-V filtering, enhances data through a generative countermeasure network, a difference value, a sampling algorithm and the like, increases the feature extraction capability by using a convolutional neural network in deep learning, performs model training by using smooth and derivative data through a deep regressor model aiming at the characteristics of the fabric near infrared spectrum data, learns and distinguishes the fabric near infrared spectrum characteristics, realizes the lossless cleaning analysis of the fabric fiber components, and obtains the fiber component type of a target fabric and the mixing proportion of all component materials in the mixed material.
Description
Technical Field
The invention relates to a textile fiber component analysis method, in particular to a near-infrared textile fiber component nondestructive cleaning analysis method based on a depth regressor, which realizes quantitative analysis aiming at a target fabric and designs a fabric fiber component detection system and a prediction method based on near-infrared spectrum.
Background
With the continuous development of economy in China, various industrial industries are developed vigorously, Artificial Intelligence (AI) is hot and popular, industrial intelligence is used as an important foothold of AI, and a plurality of application fields with considerable prospects exist. Although the traditional industry is greatly compressed in the national economy proportion, the industries such as textile industry and the like are still important industries for the national economy development, and still play a very important role in promoting the social harmony, solving the employment of the graduates and driving the brisk development of the traditional industry. The textile market is still red fire, but consumers and textile recyclers often cannot effectively and quickly distinguish fabrics, components and the like of textiles, and due to disorder and disorder of the market, a plurality of bad enterprises are speculated and skillful, so that the consumers are cheated, and the daily life of people is influenced. In the textile industry, identification of the types of textile ingredients is essential, and methods for classifying textile ingredients in the prior art are mainly classified into physical methods, chemical methods and conventional intelligent methods.
Physical and chemical methods are time-consuming and labor-consuming methods that also pollute the environment. Physical methods are the most commonly used methods for the classification of textile fibre components, in which textile material is cut up and separated into fibres, which are then subjected to operations such as combustion, chemical dissolution, etc., in which a large amount of manpower and material resources are consumed and a large amount of chemical pollution is caused, and in which the operation is often performed by specialized personnel and there are some safety problems
With the development of digital technology, conventional intelligent methods, that is, methods for detecting textiles by combining algorithms in some institutions and research departments, generally use devices such as a spectrometer and a microscope, and combine traditional mathematical statistics or artificial intelligence methods to perform component classification and characterization.
The existing textile component analysis method based on near infrared spectrum data has certain limitations: firstly, when data is collected, most of the spectrometers use large-scale spectrometers, the microscopes have large volume and need to be mastered in a certain time
Apparatus for use of the method. Secondly, the features of near infrared spectrum data are mostly acquired by using a traditional mathematical statistics method or a conventional machine learning method, and then the textiles are qualified by simple machine learning methods such as random forest, decision tree, logistic regression and the like, so that the data feature extraction capability is limited; in addition, the classification of each component of the textile is not clear enough in the current method, and the textile containing a certain type of pure material and the blended material thereof are often combined into the same category, so that when the material is increased, the scale of the related problems is exponentially increased, and a large amount of calculation power is consumed; finally, the existing near-infrared method can only predict the sample classes contained in the training set samples, has no universality, cannot identify when the components which never appear in the sample to be tested appear, and has low model generalization. The convolutional neural network combined with deep learning designed by the invention can better extract the depth characteristics of the near infrared spectrum sequence data. The whole set of system and method improves the qualitative accuracy of the textile components, has certain effect on quantification, can not only distinguish each component in the blended fabric, but also judge the contained proportion, greatly improves the efficiency of textile component classification, and has certain expansibility on the problem of complicated multi-material.
Disclosure of Invention
The object of the present invention is to overcome the problems of the prior art and to provide.
In order to overcome the defects and the defects of the existing classification algorithm in the prior art, the invention provides a fabric fiber component detection system and a prediction method based on near infrared spectrum, the system uses the reflectivity and the absorptivity of the near infrared spectrum of a sample acquired by a near infrared spectrum acquisition instrument, after data cleaning, preprocessing and data enhancement, a convolutional neural network is used for extracting depth sequence data characteristics from a textile near infrared spectrum sequence, and a multi-label classification method is combined to judge whether the characteristics of the current sample belong to one or more of various natural and artificial materials, such as cotton, hemp, rayon, modal, tencel, acrylic fiber, terylene, spandex, wool, real silk, cashmere and nylon 12, and the included proportions are judged.
According to the invention, the portable near-infrared equipment is used for scanning the surface of the textile sample to be detected to acquire the near-infrared spectrum data of the sample, so that the problem of sample damage in the detection of the traditional method is avoided, and the operation difficulty and time are greatly reduced; the light-weight convolutional neural network has better extraction capability on the near infrared spectrum sequence of the textile, and the near infrared spectrum sequence data is input into a light-weight convolutional neural network model more quickly in calculation, so that the model has better extraction capability and calculation speed on the textile spectrum information; the extracted information is classified in a multi-label mode, the complex classification problem of unit groups or multiple unit groups on 12 materials of cotton, hemp, rayon, modal, tencel, acrylic fiber, terylene, spandex, wool, real silk, cashmere and nylon is solved, and judgment can be made on the mutual proportion of the blended materials according to the difference of spectrums on different specific pure materials or blended materials.
The method for detecting the fabric fiber component by the near infrared spectrum comprises the following steps:
s1: data acquisition using near infrared spectroscopy data acquisition instrument
S11: using a near-infrared spectrometer to collect near-infrared spectrum data of the sample according to the spectrum collection specification, and recording the reflectivity and the absorptivity of the sample and the real label of the sample
S2: using various machine learning methods to clean, enhance and preprocess data, and classifying and storing the data; the method comprises the following specific steps:
s21: cleaning the data, and according to a quality control program, when the data is wrong, the system can screen out the wrong data and remind an acquirer to repeat the step of S1 for resampling;
s22: preprocessing data, denoising the data by a filtering method comprising median filtering, mean filtering, wiener filtering and S-V filtering, and enhancing the data by a generative countermeasure network, a difference value, a sampling algorithm and the like;
s23: and storing the cleaned, preprocessed and enhanced data into a database, and performing testing and division of a training set according to a certain proportion.
S3: the construction of the textile fiber component analysis model comprises the following specific processes:
s31: selecting the data band as 500-1600, selecting the used spectrum type, and generally using the reflectivity and the absorptivity;
s32: constructing a model main body network, adapting one-dimensional reflectivity data of a near infrared spectrum by using a convolutional neural network, and extracting depth characteristics of the data; convolutional neural networks used include, but are not limited to, Resnets, inclusion, TapNet, Transformer, specific for textile component analysis;
s4: the method comprises the following steps of training and verifying a textile component classification model, and specifically comprises the following steps:
s41: continuously optimizing to obtain a minimized loss function as much as possible through an optimization algorithm;
s42: the accuracy is used for evaluating the performance of the model, namely according to a hierarchical algorithm, firstly calculating the accurate class of the classification of the major classes, then calculating the accurate class of the minor classes at the bottom layer, calculating the accuracy of the classification of the components of the model, and simultaneously evaluating the accuracy of the proportion of the components by using the parameters such as MAE, MSE and the like to obtain the accuracy of the judgment of the proportion of the components
S43: putting the training set and the test set into a model for training, and repeating the processes of S41-S42 on the model in S3 until the model has good accuracy on the training set and the test set;
s44: performing cross validation on the overall data of the training set and the test set, and repeating the steps of S41-S42 to verify whether the model can be well fitted under the condition that the training set and the test set are changed
S45: taking an average result evaluation model of multiple cross validation;
s5: and (3) carrying out real-time qualitative and quantitative detection on the components of the textile fibers.
S51: collecting the surface of a sample to be detected according to a collection standard by using a near infrared spectrum data collector:
s52: selecting a wireless or wired mode to transmit data to a terminal;
s53: the terminal sends the data to the server, and the server further performs noise reduction, preprocessing and enhancement;
s54: the server puts the processed data into a model for prediction to obtain the corresponding proportion among various predicted components;
s55: the terminal will display the corresponding result, and the operator can compare with the existing label.
The invention provides a fabric fiber component detection system and a prediction method based on near infrared spectrum, and designs a complete system and method applying the method. Firstly, noise reduction is carried out by using filtering methods of median filtering, mean filtering, wiener filtering and S-V filtering, data are enhanced through a generating countermeasure network, a difference value, a sampling algorithm and the like, and finally features are extracted from textile spectrum sequence data by using an enhanced convolutional neural network near-infrared technology, so that the feature extraction capability of the textile spectrum sequence data is improved; secondly, the method covers 12 natural or artificial pure materials of cotton, hemp, rayon, modal, tencel, acrylic, terylene, spandex, wool, real silk, cashmere and nylon and material combinations thereof in a classified manner through multiple labels, has wide coverage range, solves the qualitative and quantitative problems of multiple pure materials of textile components and material combinations thereof, and solves the problems of time and labor waste and environmental pollution of the traditional physical and chemical methods and the problem of complex classification algorithm of the conventional artificial intelligence method; and finally, accurately identifying the components of the textile fibers in a fast, environment-friendly, clean and nondestructive mode by using a small and light near-infrared spectrometer and various light convolutional neural networks.
Drawings
The present application will be described in further detail with reference to the following drawings and detailed description.
FIG. 1 is a flow chart of a near infrared spectrum based fabric fiber composition detection system and prediction method of the present invention.
Detailed Description
The following detailed description of embodiments of the present invention is provided in connection with the accompanying drawings and examples. The following examples are intended to illustrate the invention but are not intended to limit the scope of the invention.
In the following description of the embodiments, for purposes of clearly illustrating the structure and operation of the present invention, directional terms are used, but the terms "front", "rear", "left", "right", "outer", "inner", "outward", "inward", "axial", "radial", and the like are to be construed as words of convenience and are not to be construed as limiting terms.
The relevant terms are explained as follows:
data preprocessing, cleaning and enhancing: the near infrared spectrum preprocessing system based on curve smoothing focuses on the problems of noise, distortion, inter-platform difference and the like of textile near infrared spectrum original data, data cleaning and noise reduction smoothing can be carried out on collected textile near infrared spectrum data, and the difficulty of subsequent textile specification analysis is greatly reduced. Data cleansing is the process of re-examining and verifying data with the aim of deleting duplicate information, correcting existing errors, and providing data consistency. Data enhancement, also called data augmentation, is the generation of more equivalent data volume on the original data as a deeply learned data set by operations such as cropping, rotation, scaling, symmetry, etc., without substantially increasing the original data volume.
Near infrared spectroscopy: near Infrared (NIR) is an electromagnetic wave between visible (vis) and mid-Infrared (MIR) and is defined by ASTM (american society for testing and materials testing) as an electromagnetic wave having a wavelength in the range of 780 to 2526 nm. The near infrared spectrum has information such as reflectivity and absorptivity of a sample, and feature extraction can be performed on the near infrared spectrum sequence data through the convolutional neural network.
Analysis of the textile fiber composition: the method is characterized in that textile fibers are treated differently to obtain a corresponding contained component table and a corresponding contained component ratio, and the textile fibers are usually covered with 12 natural or artificial pure materials of cotton, hemp, rayon, modal, tencel, acrylic, terylene, spandex, wool, real silk, cashmere and nylon and combinations of various component ratios of the materials.
Specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.
Referring to fig. 1, a flow chart of a textile fiber component detection system and a prediction method based on near infrared spectrum comprises the following steps S1-S4:
1. s1, acquiring data by using a near infrared spectrum data acquisition instrument;
1.1, a near-infrared spectrometer is used, 228 units of near-infrared light reflectivity and absorptivity between 900-1700 nm wave bands are selected for collection, a sample to be detected is opaque when the near-infrared spectrometer is used, collection points are smooth, and reflectivity and absorptivity data of one 1-dimensional and 228-feature sampling point can be obtained.
1.2, successfully collecting the non-sawteeth sample data, correctly marking the qualified sample data, and marking the components and the proportion of the sample data according to a standard component table of the textile, wherein the sample data comprises pure materials or multi-component blending.
2. Step S2, data cleaning, preprocessing, grouping and storing;
s21: data cleaning is carried out on the data, if the data which are collected wrongly are prompted during data collection, the data need to be deleted and collected again
S22: carrying out data preprocessing and enhancing: the method comprises the steps of carrying out median filtering, mean filtering, wiener filtering and S-V filtering for noise reduction, and then carrying out data enhancement through a generating type countermeasure network, a difference value, a sampling algorithm and the like.
3. Step S3, constructing a textile component classification model; the model main network comprises a residual convolutional neural network module, a branch convolutional neural network module, an attention module, a middle convolutional layer and a pooling layer.
The definition of the full convolution is:
the residual convolutional neural network is used for solving the problems that the traditional convolutional neural network brings large parameters and gradient diffusion when the number of the networks is continuously increased.
To solve the problem of gradient dispersion/explosion, a skip/shortcut connection is added, adding the input x to the output after several weight layers. The output is h (x) ═ f (x) + x, and the weighting layer actually learns a residual map: f (x) h (x) -x, we always have the flag x can transition back to the earlier layer even if the gradient of the weight layer disappears.
Specifically, the model is a residual convolutional neural network module, specifically a residual convolutional neural network module, in which Resnet9, that is, Blocks are [1,1,1,1 ]; replacing all convolution calculations in ResNet9 with a one-dimensional convolution operation of 1x1, wherein the number of convolution kernels and the number of channels per residual block need to be adjusted to accommodate the size of one-dimensional spectral sequence data, and the number of channels can be set to [8 × 8,16 × 8,32 × 8,64 × 8] for spectral sequence data of one size (1, 1, 190) in 3.1; after the first layer of convolution, 4 convolution blocks are respectively named as Block1, Block2, Block3 and Block4, and a GCBlock attention module att is added after each convolution Block; the deep features are extracted and then subjected to a convolution, Batch Normalization and ReLu intermediate convolution structure mediaConv to adapt the feature size to the next module.
Then, the feature simultaneously passes through 4 different Branch convolution neural network modules consisting of convolution, Batch Normalization and ReLu, which are respectively named Branch1, Branch2, Branch3 and Branch4, wherein the 1 st Branch structure Branch1 comprises a convolution operation of a one-dimensional convolution kernel with the size of 1, the 2 nd Branch structure Branch2 comprises a convolution operation of two one-dimensional convolution kernels with the sizes of 1 and 3 respectively, the 3 rd Branch structure Branch3 comprises a convolution operation of three one-dimensional convolution kernels with the sizes of 1, 3 and 3 respectively, and the 4 th Branch structure Branch4 comprises a convolution operation of a pooling layer and a one-dimensional convolution kernel with the size of 1; connecting the extracted 4 features to improve the network width; finally, a feature vector with the size of 2034 is finally extracted through a one-dimensional pooling GapPooling operation.
The features extracted by the convolutional neural network are classified by multiple labels, each classifier respectively judges whether 12 components including cotton, hemp, rayon, modal, tencel, acrylic fiber, polyester, spandex, wool, real silk, cashmere and nylon exist or not, namely the classification result is formed by the existence or not of the various components including the cotton, the hemp, the polyester, the spandex and the like, and meanwhile regression analysis is carried out on the proportion of each component in a sample.
The definition of the effective value convolution is:
output Z of the virtual value convolution layer(l)(m-n +1) is:
the features extracted by the convolutional neural network are classified by multiple labels, each classifier respectively judges whether 12 components including cotton, hemp, rayon, modal, tencel, acrylic, polyester, spandex, wool, silk, cashmere and nylon exist or not, namely the classification result is formed by the existence or not of the multiple components including cotton, hemp, polyester, spandex and the like.
And finally, scoring through a Softmax function, connecting and outputting simultaneous scoring to obtain a predicted value of the data on each material, wherein the sum of the probabilities of existence or nonexistence of each material is 1, the sum of prediction multi-label sets formed by all the existing materials is judged to be the prediction type of the fabric sample, and meanwhile, regression analysis is carried out on the proportion of each textile component in the sample.
4. Step S4-training and verifying the classification model of textile components:
model losses are calculated. Predictive tagset for all training samples
From the true label set Y of all training samples. A commonly used loss function for an optimized classifier for data imbalance problems is therefore:
FL(pt)＝-(1-pt)γlog(pt)
wherein p istThe probability that the training samples are predicted to be correct, gamma is a gamma factor, so that the loss of easily classified samples can be reduced, and the samples are more concerned with difficult samples. By passingThe method for minimizing the loss function is continuously optimized to obtain the loss function as small as possible.
And verifying the model on a verification set, determining classification and regression results by using the accuracy and the MAE, determining parameters for controlling the complexity of the model or the network structure according to the verification set, and verifying the training effect and progress of the current model.
The training process repeats the model construction process in the S3 process and the training and optimization process in the S4, iteratively trains on the model based on an Adam (attachment motion optimization algorithm, Adam) optimization algorithm, and verifies on the verification set after each training fixed turn until the model converges.
And finally, taking the average result of the cross validation to control the structure and the complexity of the network.
After the above process, we can get: a fabric fiber component detection system and a prediction method based on near infrared spectrum.
The method comprises the steps that after a user collects data by using a near-infrared spectrometer according to a collection standard, the data collected by the spectrometer are uploaded to a server through a terminal, the server cleans, preprocesses and enhances the data, then the data are put into a model for prediction, whether a sampled sample belongs to 12 textile components of cotton, hemp, rayon, modal, tencel, acrylic fiber, terylene, spandex, wool, real silk, cashmere and nylon or blends thereof is judged through a label-based classification algorithm, and the proportions of the various components are judged through a regression algorithm.
A fabric fiber component detection system and a prediction method based on near infrared spectrum are provided, and a model applying the method comprises a textile component analysis model based on a multi-label classification convolutional neural network and a convolutional network specially used for extracting textile component characteristics. Wherein, the convolutional neural network is a convolutional neural network adapting to spectral data and an enhancement or a variant thereof; the multi-label classification is a classification method for converting the exponential combination relation existing among different textile components into a plurality of dichotomy relations by using a multi-head classifier method.
With the rise of artificial intelligence in the industrial field, the combination of various industries and AI is a necessary way for promoting the realization efficiency of the traditional industry and saving energy, although the traditional industry is greatly compressed in the national economy proportion, the industries such as textile industry and the like are still important industries for the national economy development, and play a very important role in promoting the social harmony, solving the employment of graduation college students and driving the brisk development of the traditional industry. The textile market is still red fire, but the prior art wastes time and labor for the textile component classification method, has damage to a detection sample and pollutes the environment, and the invention provides a fabric fiber component detection system and a prediction method based on near infrared spectrum in order to overcome the defects and shortcomings of the prior art in the current classification algorithm.
Compared with the prior art, the invention realizes the following effects: the method comprises the steps that firstly, a convolutional neural network based on a deep learning model is used as a main framework of a model used by a system, and the convolutional neural network has the advantages of being fast, light and accurate; secondly, the problem that textile fiber components comprise a plurality of groups and the content of the groups is different can be effectively solved based on a multi-label classification algorithm model, and the universality is higher; thirdly, the proportion of each component in the textile fiber component calculated based on the regression algorithm can correctly help the user to clarify the component composition of the sample, and the user can conveniently process the sample in the next step.
Claims (5)
1. The system and the method for detecting and predicting the textile fiber components based on the near infrared spectrum are characterized by comprising the following steps of:
s1: collecting data by using a near infrared spectrum data collector;
s2: using various machine learning methods to clean and preprocess data, and performing classified group classification and storage;
s3: constructing a textile fiber component analysis model;
s4: training and verifying a textile fiber component analysis model;
s5: and (3) carrying out real-time qualitative and quantitative detection on the components of the textile fibers.
2. The system and method for detecting and predicting the fiber content of textile fabrics in the near infrared spectrum of claim 1 wherein: the step 2 of performing data cleaning, preprocessing and classified group classified storage by using various machine learning methods further comprises the following steps:
s21: the method comprises the steps of carrying out noise reduction data cleaning and preprocessing aiming at collected textile near infrared spectrum original data by using a filtering method of median filtering, mean filtering, wiener filtering and S-V filtering, and carrying out data noise reduction smoothing on normal waveform data, and relates to algorithms comprising Fourier transform noise reduction, wiener filtering noise reduction, segmented Savitzky-Golay smoothing noise reduction, spectral subtraction voice noise reduction, wavelet threshold noise reduction, multi-granularity empirical mode decomposition and the like, and the method can be used for carrying out aiming at noise reduction according to the actual condition of the near infrared spectrum data and realizing data smoothing;
s22: enhancing data through a generative countermeasure network, a difference value, a sampling algorithm and the like;
s23: the near infrared spectrum of the data is put into a model after being subjected to data processing by the method, the data is divided into 8 categories at the top layer at present, wherein the categories comprise natural plant fibers, acrylic fibers and blending thereof, terylene and natural plant fiber blending, terylene blending, nylon and natural plant fiber blending, nylon blending, wool, cashmere and silk blending and the like, and the category of the bottom layer is divided into 63 categories.
3. The system and method for detecting and predicting the fiber content of textile fabrics in the near infrared spectrum of claim 1 wherein: the construction of the textile fiber component analysis model in step 3 further comprises:
s31: constructing model input and output, and setting data as one-dimensional near infrared spectrum reflectance value vectors equivalent to wavelengths;
s32: a model backbone network is constructed, and a convolutional neural network is used as a feature extractor to adapt to sequence data of the near infrared spectrum.
4. The system and method for detecting and predicting the fiber content of textile fabrics in the near infrared spectrum of claim 1 wherein: in step S32, the method for detecting and predicting textile fiber components based on near infrared spectroscopy further comprises:
s321: the model backbone network is a convolutional neural network, including but not limited to: widened convolutional neural networks inclusion, residual neural networks ResNet specially used for textile component near infrared spectrum data to perform textile component analysis, convolutional neural networks capable of extracting spectral data characteristics of small sample learning networks TapNet and related variants;
s322: the method comprises the steps of improving the capability of a backbone network for extracting features, adding a channel attention SE module, a GC Block module and a self-attention module to a convolutional neural network which is specially used for carrying out component analysis on textile near infrared spectrum data;
s323: the method comprises the steps of optimizing a backbone network to improve the characteristic extraction capability, carrying out component analysis on textile near infrared spectrum data by using a convolutional neural network, using a multilayer deep residual neural network, and stacking the width of a model by using a plurality of branch convolutional network blocks to increase the receptive field of each dimension of a spectrum sequence.
5. The system and method for detecting and predicting the fiber content of textile fabrics in the near infrared spectrum of claim 1 wherein: the training and verification of the textile fiber component analysis model in step 4 further comprises:
s41, optimizing according to the loss function of the training set on the near infrared spectrum fabric fiber component detection system;
s42, the textile component categories include 12 common pure material categories of cotton, hemp, rayon, modal, tencel, acrylic, terylene, spandex, wool, real silk, cashmere and nylon and the combination category of blended fabrics composed of the materials.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111178122.6A CN113970532B (en) | 2021-10-09 | 2021-10-09 | Fabric fiber component detection system and prediction method based on near infrared spectrum |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111178122.6A CN113970532B (en) | 2021-10-09 | 2021-10-09 | Fabric fiber component detection system and prediction method based on near infrared spectrum |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113970532A true CN113970532A (en) | 2022-01-25 |
| CN113970532B CN113970532B (en) | 2024-04-19 |
Family
ID=79587208
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111178122.6A Active CN113970532B (en) | 2021-10-09 | 2021-10-09 | Fabric fiber component detection system and prediction method based on near infrared spectrum |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113970532B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117524340A (en) * | 2024-01-04 | 2024-02-06 | 南京信息工程大学 | Textile component quantitative characterization method based on multilayer one-dimensional CNN depth network |
| CN117740727A (en) * | 2024-02-19 | 2024-03-22 | 南京信息工程大学 | Textile component quantitative inversion method based on infrared hyperspectrum |
| CN117740727B (en) * | 2024-02-19 | 2024-05-14 | 南京信息工程大学 | Textile component quantitative inversion method based on infrared hyperspectrum |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102564966A (en) * | 2012-02-03 | 2012-07-11 | 江西出入境检验检疫局检验检疫综合技术中心 | Near infrared rapid non-destructive detection method for textile components |
| CN107219188A (en) * | 2017-06-02 | 2017-09-29 | 中国计量大学 | A kind of method based on the near-infrared spectrum analysis textile cotton content for improving DBN |
| WO2020109170A1 (en) * | 2018-11-27 | 2020-06-04 | BSH Hausgeräte GmbH | Textile identification apparatus and method for identifying a textile type |
| CN111369500A (en) * | 2020-02-21 | 2020-07-03 | 北京雪莲集团有限公司 | Textile classification and identification method based on infrared spectrum detection technology |
| CN111751318A (en) * | 2019-03-29 | 2020-10-09 | 因斯派克托里奥股份有限公司 | Fabric validation using spectral measurements |
-
2021
- 2021-10-09 CN CN202111178122.6A patent/CN113970532B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102564966A (en) * | 2012-02-03 | 2012-07-11 | 江西出入境检验检疫局检验检疫综合技术中心 | Near infrared rapid non-destructive detection method for textile components |
| CN107219188A (en) * | 2017-06-02 | 2017-09-29 | 中国计量大学 | A kind of method based on the near-infrared spectrum analysis textile cotton content for improving DBN |
| WO2020109170A1 (en) * | 2018-11-27 | 2020-06-04 | BSH Hausgeräte GmbH | Textile identification apparatus and method for identifying a textile type |
| CN111751318A (en) * | 2019-03-29 | 2020-10-09 | 因斯派克托里奥股份有限公司 | Fabric validation using spectral measurements |
| CN111369500A (en) * | 2020-02-21 | 2020-07-03 | 北京雪莲集团有限公司 | Textile classification and identification method based on infrared spectrum detection technology |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117524340A (en) * | 2024-01-04 | 2024-02-06 | 南京信息工程大学 | Textile component quantitative characterization method based on multilayer one-dimensional CNN depth network |
| CN117740727A (en) * | 2024-02-19 | 2024-03-22 | 南京信息工程大学 | Textile component quantitative inversion method based on infrared hyperspectrum |
| CN117740727B (en) * | 2024-02-19 | 2024-05-14 | 南京信息工程大学 | Textile component quantitative inversion method based on infrared hyperspectrum |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113970532B (en) | 2024-04-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110717368A (en) | Qualitative classification method for textiles | |
| CN113686804B (en) | Textile fiber component nondestructive cleaning analysis method based on deep regression network | |
| CN108256482A (en) | A kind of face age estimation method that Distributed learning is carried out based on convolutional neural networks | |
| Du et al. | Efficient recognition and automatic sorting technology of waste textiles based on online near infrared spectroscopy and convolutional neural network | |
| CN116363440B (en) | Deep learning-based identification and detection method and system for colored microplastic in soil | |
| CN111369500A (en) | Textile classification and identification method based on infrared spectrum detection technology | |
| CN103149210A (en) | System and method for detecting fabric cashmere content based on scale graphic features | |
| Saleh et al. | Palm oil classification using deep learning | |
| CN113970532B (en) | Fabric fiber component detection system and prediction method based on near infrared spectrum | |
| CN115205209A (en) | Monochrome cloth flaw detection method based on weak supervised learning | |
| CN114972342A (en) | Gearbox gear surface defect detection method | |
| CN113408616B (en) | Spectral classification method based on PCA-UVE-ELM | |
| CN117524340A (en) | Textile component quantitative characterization method based on multilayer one-dimensional CNN depth network | |
| CN117172430B (en) | Deep learning-based water body environment assessment and prediction method and system | |
| CN112927180B (en) | Cashmere and wool optical microscope image identification method based on generation of confrontation network | |
| CN108428234B (en) | Interactive segmentation performance optimization method based on image segmentation result evaluation | |
| CN113838040A (en) | Detection method for defect area of color texture fabric | |
| CN115406852A (en) | Fabric fiber component qualitative method based on multi-label convolutional neural network | |
| CN117491293A (en) | High-steep bank slope carbonate rock corrosion rapid evaluation method based on hyperspectrum | |
| CN111275131A (en) | Chemical image classification and identification method based on infrared spectrum | |
| Das et al. | Moment-based features of knitted cotton fabric defect classification by artificial neural networks | |
| Nie et al. | Machine vision-based apple external quality grading | |
| CN114414523A (en) | Textile fiber component qualitative method based on automatic waveband selection | |
| CN114264626A (en) | Fabric nondestructive quantitative analysis method based on time series residual error network | |
| CN115455407A (en) | Machine learning-based GitHub sensitive information leakage monitoring method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |