CN114088657B - Textile fiber component analysis method based on depth encoder/decoder - Google Patents
Textile fiber component analysis method based on depth encoder/decoder Download PDFInfo
- Publication number
- CN114088657B CN114088657B CN202111176920.5A CN202111176920A CN114088657B CN 114088657 B CN114088657 B CN 114088657B CN 202111176920 A CN202111176920 A CN 202111176920A CN 114088657 B CN114088657 B CN 114088657B
- Authority
- CN
- China
- Prior art keywords
- depth
- textile
- data
- textile fiber
- codec
- 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.)
- Active
Links
- 239000004753 textile Substances 0.000 title claims abstract description 76
- 239000000835 fiber Substances 0.000 title claims abstract description 43
- 238000004458 analytical method Methods 0.000 title claims abstract description 33
- 238000001228 spectrum Methods 0.000 claims abstract description 23
- 230000004927 fusion Effects 0.000 claims abstract description 10
- 238000013507 mapping Methods 0.000 claims abstract description 4
- 238000000034 method Methods 0.000 claims description 23
- 238000002329 infrared spectrum Methods 0.000 claims description 19
- 230000003595 spectral effect Effects 0.000 claims description 19
- 238000012545 processing Methods 0.000 claims description 13
- 238000007781 pre-processing Methods 0.000 claims description 12
- 238000004445 quantitative analysis Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 238000001914 filtration Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 7
- 238000004140 cleaning Methods 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 5
- 238000000605 extraction Methods 0.000 claims description 5
- 230000009467 reduction Effects 0.000 claims description 5
- 238000012937 correction Methods 0.000 claims description 4
- 238000013519 translation Methods 0.000 claims description 4
- 239000011159 matrix material Substances 0.000 claims description 2
- 230000000452 restraining effect Effects 0.000 claims description 2
- 238000011897 real-time detection Methods 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 9
- 239000004744 fabric Substances 0.000 abstract description 5
- 239000000463 material Substances 0.000 abstract description 5
- 238000005259 measurement Methods 0.000 abstract description 4
- 230000006870 function Effects 0.000 description 12
- 238000004422 calculation algorithm Methods 0.000 description 10
- 238000004451 qualitative analysis Methods 0.000 description 8
- 239000000126 substance Substances 0.000 description 6
- 238000009499 grossing Methods 0.000 description 4
- 238000002156 mixing Methods 0.000 description 4
- 230000008569 process Effects 0.000 description 4
- 238000004497 NIR spectroscopy Methods 0.000 description 3
- 239000003153 chemical reaction reagent Substances 0.000 description 3
- 229920000742 Cotton Polymers 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000002045 lasting effect Effects 0.000 description 2
- 239000000203 mixture Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- 244000025254 Cannabis sativa Species 0.000 description 1
- 235000012766 Cannabis sativa ssp. sativa var. sativa Nutrition 0.000 description 1
- 235000012765 Cannabis sativa ssp. sativa var. spontanea Nutrition 0.000 description 1
- 229920000433 Lyocell Polymers 0.000 description 1
- 239000004677 Nylon Substances 0.000 description 1
- 229920000297 Rayon Polymers 0.000 description 1
- 229920002334 Spandex Polymers 0.000 description 1
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 235000009120 camo Nutrition 0.000 description 1
- 210000000085 cashmere Anatomy 0.000 description 1
- 235000005607 chanvre indien Nutrition 0.000 description 1
- 239000002131 composite material Substances 0.000 description 1
- 238000001212 derivatisation Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000004090 dissolution Methods 0.000 description 1
- 239000011487 hemp Substances 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000002372 labelling Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000001320 near-infrared absorption spectroscopy Methods 0.000 description 1
- 229920001778 nylon Polymers 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000003960 organic solvent Substances 0.000 description 1
- 230000002085 persistent effect Effects 0.000 description 1
- 238000004940 physical analysis method Methods 0.000 description 1
- 238000000053 physical method Methods 0.000 description 1
- 229920000728 polyester Polymers 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 239000002964 rayon Substances 0.000 description 1
- 239000004759 spandex Substances 0.000 description 1
- 241000894007 species Species 0.000 description 1
- 238000004611 spectroscopical analysis Methods 0.000 description 1
- 210000002268 wool Anatomy 0.000 description 1
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
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Spectroscopy & Molecular Physics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Chemical & Material Sciences (AREA)
- Biochemistry (AREA)
- Analytical Chemistry (AREA)
- Pathology (AREA)
- Immunology (AREA)
- Investigating Or Analysing Materials By Optical Means (AREA)
- Treatment Of Fiber Materials (AREA)
Abstract
The invention discloses a textile fiber component analysis method based on a depth codec, which combines a spectrum unmixing principle to design an end-to-end depth codec model so as to realize the aim of textile fiber component measurement. The invention converts the textile fiber component measurement problem into a problem of solving the pure material and the corresponding ratio contained in the fabric, designs a depth encoder model, learns the space mapping after the high-dimensional input data is reduced in dimension, reconstructs the original data by the depth encoder, and particularly designs a fusion loss function combining reconstruction loss and regression loss to improve component analysis performance. The analysis method is rapid, nondestructive and clean in detection, high in accuracy and easy to operate.
Description
Technical Field
The invention relates to the technical field of textile component analysis, in particular to a qualitative and quantitative analysis method for solving near infrared textile fiber component nondestructive cleaning by a deep codec model.
Background
Along with the improvement of the living standard of people, the requirements on the quality of textiles are also continuously improved. The phenomena of unclear labeling, poor quality, and the like of related textile components existing in a large number on the market bring about high demands for analysis of textile components. At present, a textile fiber component analysis method mainly adopts a chemical method and a physical method traditionally. The chemical method needs to send the textile to be detected to a related detection mechanism for detection, and the method for detecting with the assistance of the chemical reagent not only can cause loss on the textile, but also is complex in operation and long in detection period. The physical analysis method adopts a micro-projection method at first, the accuracy of quantitative analysis is largely determined by the capability of an analyst for identifying various fibers, and the prior two methods are difficult to simultaneously extract enough characteristic materialization information of each fiber, so that accurate and reliable component analysis work cannot be performed.
The near infrared spectrometry of FZ/T01144-2018 textile fiber quantitative analysis is formally implemented on 1 month 7 in 2019, which marks that the textile component analysis method based on the near infrared spectrum enters an application stage from the research field. The method for analyzing the fiber components of the textile by using the near infrared spectroscopy has the advantages of being rapid and free of damage, and can realize accurate identification of different types of fibers by preparing a standard sample in the early stage and establishing a detection model. Near infrared spectroscopy analysis of textile fiber components mainly covers two tasks: qualitative analysis and quantitative analysis, wherein the qualitative analysis refers to determining the composition type of the target fabric, and the quantitative analysis refers to determining the mixing proportion of the composition materials of the mixed materials.
In spectroscopic analysis, the purpose of mixed signal unmixing is to obtain a pure spectrum of the contained species (i.e. end-member extraction) and its corresponding content (i.e. end-member abundance estimation). After near infrared spectrum signals are determined as research objects, the problem that the essence of qualitative and quantitative analysis of textiles is to solve pure materials (spectrum end members) and corresponding ratios (abundance values) contained in fabrics is considered, spectral unmixing is creatively applied to textile component analysis, a textile fiber component analysis method based on a depth codec is designed, a textile fiber component analysis scheme based on the depth codec is provided aiming at defects and technical difficulties of a traditional detection method, and a fusion loss function combining reconstruction loss and regression loss is specifically designed to improve component analysis performance. The method is nondestructive, clean, efficient, quick and low in cost, and can be used for laboratory analysis, field analysis and the like.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the existing detection algorithm in the prior art, and provides a textile fiber component analysis method based on a depth codec by combining a spectrum unmixing theory, wherein the method relates to near infrared spectrum equipment and a core depth codec module. The method comprises a data marking module, a plurality of preprocessing mode selection modules based on near infrared spectrum data, a depth coding module, a depth decoding module, a constraint module and a qualitative and quantitative result output module. The near infrared equipment is used for collecting the spectrum information of the current textile; the data mark refers to that according to the existing text mark, a script is used for converting the text mark into required data information and a data label; the spectrum data preprocessing method comprises data denoising processing, including median filtering denoising, wiener filtering denoising and wavelet denoising, multi-element scattering correction for eliminating baseline translation and drift, bilinear interpolation and Bezier curve interpolation for expanding data dimension processing, and data enhancement processing based on generation of an countermeasure network; the depth encoder module is used for learning the space mapping of the high-dimensional input data after the dimension reduction; the depth decoder module is used for reconstructing original data; the constraint module is used for non-negative constraint summation and constraint of the prediction result; the qualitative and quantitative result output module is used for predicting the components of the textile end to end, is clean without damage and can directly output qualitative and quantitative results.
In order to achieve the above purpose, the present invention adopts the following scheme: the method has the core function of solving the problem of component mixing by using a depth codec model, can be applied to qualitative and quantitative analysis of textile components, and can be used for carrying out high-precision component prediction on textiles. The user can provide a textile, and can test the type and content of the textile without the need for chemical reagents and without the need to destroy the sample. Firstly, collecting data in a large scale, carrying out preprocessing related works such as smoothing and derivatization on the data, then using the data in a training depth codec model, specifically designing a fusion loss function combining reconstruction loss and regression loss in the training process to improve component analysis performance, and carrying out non-negative constraint and constraint on components to meet the reality condition; and entering a test link, wherein the spectral information of the test data can be used as a target for testing, the spectral data can be collected again for preprocessing, and the test data is put into a well-trained depth codec model from end to obtain a qualitative and quantitative analysis result.
The textile fiber component analysis method based on the depth encoder can effectively detect the textile content of different specifications. The property content of twelve pure materials and mixed materials such as cotton, hemp, human cotton (rayon), polyester fiber, nylon, wool, cashmere, spandex, tencel, silk and the like is predicted and evaluated. The qualitative analysis and the quantitative analysis of the method comprise the following steps:
S1: establishing a near infrared spectrum characteristic database of textile fiber components;
S11: based on the theory of spectral unmixing (Spectral unmixing), the correspondence of spectral information to textile components is defined. The aim of spectral unmixing is to obtain the end-member components and their content contained in the mixed pixels or signals, i.e. end-member extraction and end-member abundance estimation. In general, the principle of operation of the unmixed model may be defined as:
Wherein, Reconstructed signal or pixel, η e R d is noise, in pixel unmixing represents the spectral value of a pixel point on d channels, in signal represents a sequence data signal ;E＝(e1,e2,...,eM)＝g(z)＝ (g(z1),g(z2),...,g(zM))∈RM×d with length d is the abundance of M end members e m∈Rd;a＝ (a1,a2,...,aM)T as each end member, generally,/>And each term is non-negative;
When the spectrum information database of the textile fiber components is established, the category information of the textile can be regarded as end members e m in spectrum unmixing, the content information of the textile can be regarded as abundance of each end member, and the connection point of the two is that the near infrared spectrum information of the textile fiber components can be regarded as near infrared spectrum information in spectrum unmixing, so that the near infrared spectrum characteristic database of the textile fiber components is established.
S2: through a series of preprocessing such as data cleaning, data noise reduction, data enhancement, band selection and the like;
S21: the following preprocessing is carried out on the collected and marked data: the data denoising processing comprises median filtering denoising, wiener filtering denoising and wavelet denoising, multi-element scattering correction for eliminating baseline translation and drift, bilinear interpolation and Bezier curve interpolation for expanding data dimension processing, and data enhancement processing based on generation of an countermeasure network.
S3: creating a depth codec model;
S31: creating a depth coding layer, and the purpose of a depth coder learns a space mapping of high-dimensional input data after dimension reduction, and converts an input x into a low-dimensional characteristic representation v:
v＝fE(x)
Using a conventional depth network as an encoder, the four fully connected layers taper from the input layer to deeper layers. The number of nodes of the last hidden layer is equal to the dimension of the end member. The output of the input layer and the last hidden layer may be any value, such as when 228 bands and 4 components (end members) of near infrared spectral data are used, the dimensions of the input layer and the last hidden layer may be set to 224 and 4, respectively. The first three layers, except the last hidden layer, employ an activation function Is a rectifying linear unit (ReLU), leakage ReLU (LeayReLU), or Sigmoid;
s32: creating a depth decoding layer, a depth decoder reconstructing the original data
The depth decoder includes a linear portion and a nonlinear portion. The linear part of the depth decoder represents a linear model of spectral unmixing, the relevant weights are used for extracting end members, the nonlinear part of the decoder represents nonlinear iteration between the end members, noise in textile spectral information is resisted, in textile fiber component analysis, the linear model is reasonable, and the influence of the nonlinear part is regarded as fluctuation;
S33: interpreting the output v network of the encoder as an estimate of the abundance vector a and the W of the linear part of the weight decoder as the extracted end-member matrix M, the unmixed model, can be expressed as:
abundance estimation:
And (3) end member extraction:
s4: constructing a fusion loss function of the depth coder;
s41: creating a first part of a fusion loss function, minimizing the input x of the encoder and the output of the decoder Implementing constraints on the reconstruction function:
s42: creating a second part of the fusion loss function, and restraining textile components of the predicted result, wherein the restraint mainly comprises a non-negative restraint sum and a restraint, and considering that the real textile quantitative analysis result is a non-negative value of the sum;
The non-negative constraint is embodied by, for x= [ X 1,x2,…,xn]T,
And a constraint is embodied in that, for x= [ X 1,x2,…,xn]T,
Taking the absolute value of the output to satisfy the non-negative constraint and then normalizing this non-negative quantity to satisfy a constraint may be expressed as:
s43: the accuracy of the reconstructed end-members and abundance estimates is assessed by Root Mean Square Error (RMSE):
s5: training, testing and storing a depth codec model;
S6: and detecting the textile fiber end to end in real time.
Compared with the prior art, the invention has the following advantages: the invention creatively combines spectral unmixing with qualitative and quantitative analysis of textiles, solves the problems of difficult measurement of blending ratio of blended products with similar structures and similar chemical dissolution performance in the textile industry, pollution problem of using a large amount of organic solvents and acid-base reagents in the measurement of blending ratio of the blended products of insoluble fibers and soluble fibers, and digital and intelligent detection of textile components, and provides a nondestructive clean textile component analysis method. The near infrared spectrum data is used, and based on a depth codec network, a fusion loss function combining reconstruction loss and regression loss is specifically designed through content sum being a constraint and content non-negative constraint, so that nondestructive rapid detection of partial textile components is realized.
Drawings
The application will be described in further detail with reference to the drawings and the detailed description.
FIG. 1 is a flow chart of a near infrared spectroscopy method used in the present invention.
Fig. 2 is a flow chart of the textile composition analysis of the present invention.
Fig. 3 is a technical framework of the present invention.
Fig. 4 shows a codec structure according to the present invention.
Detailed Description
The following describes in further detail the embodiments of the present invention with reference to the drawings and examples. The following examples are illustrative of the invention and are not intended to limit the scope of the invention.
As shown in fig. 1, the near infrared spectrum analysis method establishes a prediction model between spectrum data of a sample and sample target value data, and can accurately predict the content of a substance by acquiring spectrum data of an unknown sample according to the established prediction model. The prediction model not only depends on a strict model algorithm with strong data processing and expression capability, but also depends on the characteristics of the acquired sample, such as the accuracy of sample data, whether the data characteristics can reflect the properties of the sample, and the like. In general, developing a substance content prediction model based on near infrared spectrum mainly includes:
1. obtaining experimental data;
2. Identifying and removing abnormal samples;
3. the sample set is divided into a training set and a testing set;
4. preprocessing spectral data;
5. Extracting characteristic wavelengths;
6. A predictive model of the target content of the substance is established by means of a machine learning algorithm.
The method uses the general flow of the near infrared spectrum analysis method, improves the method based on the linear spectrum unmixing method, and develops a steady and high-precision textile component analysis model.
As shown in fig. 2, fig. 2 is a flowchart of the textile inspection provided by the present invention, in which the textile fiber component analysis process is described in detail.
The algorithm preparation work of the invention is carried out according to the following steps:
step 1: collecting near infrared spectrum characteristics of textile fiber components and performing data cleaning;
step 2: preprocessing the original acquired data, denoising, dimension expansion and data enhancement;
step 3: creating a depth codec network, performing non-negative constraint and a constraint, and training the network by using the data obtained in the step 2;
step 4: lasting the trained model;
The data cleaning work in the first step uses a segmented Savitzky-Golay algorithm of the signal and information standard deviation characteristics of the signal. The segmented Savitzky-Golay algorithm of the signal is a polynomial smoothing algorithm based on the least squares principle, also known as convolutional smoothing. The principle is that 5 points with equal wavelength intervals in a section of a spectrum are marked as an X set, polynomial smoothing is to replace m points by using polynomial fitting values of data of which the wavelength points are m points left two, m points left one, m points right one and m points right two, and then the polynomial fitting values are sequentially moved until the spectrum signal is traversed. Noise data is cleaned by calculating the curve smoothness and information standard deviation of the overall signal.
The data preprocessing in the second step is to perform the following preprocessing on the collected and marked data: and the data denoising processing comprises median filtering denoising, wiener filtering denoising and wavelet denoising, multi-element scattering correction for eliminating baseline translation and drift, bilinear interpolation and Bezier curve interpolation for expanding data dimension processing, and data enhancement processing based on generation of an countermeasure network. In the use process, the data noise reduction and data enhancement algorithm can be selected and combined by oneself according to the data characteristics, and parameters are adjusted to achieve the optimal prediction effect.
In the third step, a depth codec network is created, and four fully-connected layers are gradually reduced from an input layer to a deeper layer by using a conventional depth network as an encoder. The number of nodes of the last hidden layer is equal to the dimension of the end member. The output of the input layer and the last hidden layer may be any value, such as 228 and 4 dimensions of the input layer and the last hidden layer when 228 bands and 4 components (end elements) of near infrared spectrum data are used, respectively. The first three layers, except the last hidden layer, employ an activation functionMay be a rectifying linear unit (ReLU), leakage ReLU (LeayReLU), and Sigmoid.
The depth decoder includes a linear portion and a nonlinear portion. The linear part of the depth decoder represents a linear model of spectral unmixing, the associated weights are used to extract the end members, and the nonlinear part of the decoder represents nonlinear iterations between the end members for later prediction.
In the fourth step, the weights and parameters of the obtained depth codec network are persisted to facilitate subsequent calls.
The textile component prediction work of the invention is carried out according to the following steps:
step 1: collecting near infrared spectrum characteristics of textile fiber components;
step 2: preprocessing the original acquired data, denoising, dimension expansion and data enhancement;
step 3: a persistent depth encoder network;
Step 4: performing non-negative constraint and a constraint;
step 5: and obtaining the qualitative and quantitative result of the textile fiber.
The step 1 requires the user to have uniform collection environment and no interference of external environments such as excessive illumination. And randomly selecting a plurality of characteristic points of each piece of cloth, and recording the spectral data characteristics of each piece of cloth.
Wherein, the step 3 and the step 4 embody the principle of combining a spectrum unmixing algorithm with the analysis of textile components, and the training process of the spectrum unmixing algorithm mainly comprises the steps of reconstructing signalsModeling optimization of the difference function with the actual value x, the abundance a is presumed, namely:
s.t.||f(E,a)||1＝1,a≥0
Wherein ||f (E, a) || 1 =1 means and a constraint, and a+.gtoreq.0 means a non-negative constraint. The actual input value x is reconstructed through a coder and decoder, and the hidden layer output and the network parameters can be regarded as end member abundance and end member information obtained by unmixing.
In step 5, since the model is end-to-end, real-time testing can be realized, and the prediction result can be obtained without damage and cleaning. After a user uses near infrared equipment to scan the surface of the textile to be tested, spectral data is uploaded to a terminal, and a trained and lasting depth codec network model is subjected to reasoning calculation to obtain a qualitative and quantitative result of the textile fiber.
Fig. 3 is a technical frame diagram provided by the invention.
Table 1 shows the model architecture of the depth codec used in the present invention. Using a conventional depth network as an encoder, the four fully connected layers taper from the input layer to deeper layers. The number of nodes of the last hidden layer is equal to the dimension of the end member. The output of the input layer and the last hidden layer may be any value, such as 228 and 4 dimensions of the input layer and the last hidden layer, respectively, when using 228 bands and 4 components (end members) of near infrared spectral data. The first three layers, except the last hidden layer, employ an activation function phi, which may be a rectifying linear unit (ReLU), leakage ReLU (LeayReLU), and Sigmoid. The decoder is divided into a linear part and a nonlinear part and is used for reconstructing waveforms, and the full-connection layer combined activation function is also used, so that the network structure is simple, the interpretability is strong, and the qualitative and quantitative effect of the textile fiber is good.
The foregoing is merely a preferred embodiment of the present application, and it should be noted that modifications and substitutions will now occur to those skilled in the art without departing from the technical principles of the present application, and these modifications and substitutions should also be considered to be within the scope of the present application.
Claims (1)
1. A method for analyzing textile fiber components based on a depth codec, the method comprising the steps of:
S1: establishing a near infrared spectrum characteristic database of textile fiber components;
s2: data cleaning, data noise reduction, data enhancement and band selection;
S3: creating a depth codec model;
s4: constructing a fusion loss function of the depth coder;
s5: training, testing and storing a depth codec model;
s6: the end-to-end real-time detection of the textile fiber,
Step S1 in the depth codec-based textile fiber component analysis method further comprises:
S11: based on a spectrum unmixing theory, defining a corresponding relation between spectrum information and textile components, wherein the spectrum unmixing aim is to obtain end member components and contents thereof contained in mixed pixels or signals, namely end member extraction and end member abundance estimation, and the working principle of an unmixing model is defined as follows:
Wherein, The reconstructed signal or pixel, eta epsilon R d is noise, the spectrum value of a pixel point on d channels is represented in pixel unmixing, and a sequence data signal with the length of d is represented in the signal; e= (E 1,e2,…,eM) =g (z) =
(G (z 1),g(z2),…,g(zM))∈RM×d is the abundance of M end members e m∈Rd;a＝(a1,a2,…,aM)T is each end member, and each term is non-negative;
When the spectrum information database of the textile fiber components is established, the category information of the textile can be regarded as end members e m in spectrum unmixing, the content information of the textile can be regarded as abundance of each end member, the connection point of the two is that the near infrared spectrum information of the textile fiber components can be regarded as near infrared spectrum information in spectrum unmixing, thereby establishing a near infrared spectrum characteristic database of the textile fiber components,
Step S2 in the depth codec-based textile fiber component analysis method further includes:
S21: the following preprocessing is carried out on the collected and marked data: data denoising processing, including median filtering denoising, wiener filtering denoising and wavelet denoising, multi-element scattering correction for eliminating baseline translation and drift, bilinear interpolation and Bezier curve interpolation for expanding data dimension processing, data enhancement processing based on generation of an countermeasure network,
Step S3 in the depth codec-based textile fiber component analysis method further includes:
S31: creating a depth coding layer, and the purpose of a depth coder learns a space mapping after the high-dimensional input data is reduced in dimension, and converts an input x into a low-dimensional characteristic representation v:
v＝fE(x)
Using conventional depth network as encoder, four fully connected layers gradually shrink from input layer to deeper layer, the node number of the last hidden layer is equal to the dimension of end member, the output of input layer and last hidden layer is set according to actual requirement, except the last hidden layer, the first three layers adopt activation function Rectifying linear units (ReLU), leakage ReLU (LeayReLU), and Sigmoid;
s32: creating a depth decoding layer, a depth decoder reconstructing the original data
The depth decoder comprises a linear part and a nonlinear part, the linear part of the depth decoder represents a linear model of spectral unmixing, related weights are used for extracting end members, the nonlinear part of the decoder represents nonlinear iteration among the end members, noise in textile spectral information is resisted, in textile fiber component analysis, the linear model is a reasonable assumption, and the influence of the nonlinear part is regarded as fluctuation;
S33: interpreting the output u-network of the encoder as an estimate of the abundance vector a, the W of the linear portion of the weight decoder as the extracted end-member matrix M, i.e. the unmixed model, can be expressed as:
abundance estimation:
And (3) end member extraction:
Step S4 in the depth codec-based textile fiber component analysis method further includes:
S41: creating a first part of the fusion loss function, minimizing the input x of the encoder and the output of the decoder Implementing constraints on the reconstruction function:
S42: creating a second part of the fusion loss function, and restraining the textile components of the predicted result, wherein the restraint mainly comprises a non-negative restraint sum and a restraint, and considering that the real textile quantitative analysis result is a non-negative value of the sum;
The non-negative constraint is embodied as, for
And a constraint is embodied as to
Taking the absolute value of the output to satisfy the non-negative constraint and then normalizing this non-negative quantity to satisfy a constraint may be expressed as:
S43: the accuracy of the reconstructed end-members and abundance estimates is assessed by Root Mean Square Error (RMSE):
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111176920.5A CN114088657B (en) | 2021-10-09 | 2021-10-09 | Textile fiber component analysis method based on depth encoder/decoder |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111176920.5A CN114088657B (en) | 2021-10-09 | 2021-10-09 | Textile fiber component analysis method based on depth encoder/decoder |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114088657A CN114088657A (en) | 2022-02-25 |
| CN114088657B true CN114088657B (en) | 2024-05-14 |
Family
ID=80296669
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111176920.5A Active CN114088657B (en) | 2021-10-09 | 2021-10-09 | Textile fiber component analysis method based on depth encoder/decoder |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114088657B (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003077263A2 (en) * | 2002-03-06 | 2003-09-18 | Advanced Photometrics, Inc. | Method and apparatus for radiation encoding and analysis |
| CN107179291A (en) * | 2017-05-10 | 2017-09-19 | 中国科学院合肥物质科学研究院 | Soil Heavy Metal Elements Content Forecasting Methodology based on tera-hertz spectra Yu depth autocoder |
| 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 |
| CN112529865A (en) * | 2020-12-08 | 2021-03-19 | 西安科技大学 | Mixed pixel bilinear deep layer de-mixing method, system, application and storage medium |
| CN112699838A (en) * | 2021-01-13 | 2021-04-23 | 武汉大学 | Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting |
-
2021
- 2021-10-09 CN CN202111176920.5A patent/CN114088657B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2003077263A2 (en) * | 2002-03-06 | 2003-09-18 | Advanced Photometrics, Inc. | Method and apparatus for radiation encoding and analysis |
| CN107179291A (en) * | 2017-05-10 | 2017-09-19 | 中国科学院合肥物质科学研究院 | Soil Heavy Metal Elements Content Forecasting Methodology based on tera-hertz spectra Yu depth autocoder |
| 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 |
| CN112529865A (en) * | 2020-12-08 | 2021-03-19 | 西安科技大学 | Mixed pixel bilinear deep layer de-mixing method, system, application and storage medium |
| CN112699838A (en) * | 2021-01-13 | 2021-04-23 | 武汉大学 | Hyperspectral mixed pixel nonlinear blind decomposition method based on spectral diagnosis characteristic weighting |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114088657A (en) | 2022-02-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113686804B (en) | Textile fiber component nondestructive cleaning analysis method based on deep regression network | |
| CN110376139A (en) | Soil organic matter content quantitative inversion method based on ground high-spectrum | |
| CN112597038B (en) | Software defect prediction method and system | |
| CN107451760B (en) | Rolling bearing fault diagnosis method based on time window slip limited Boltzmann machine | |
| CN101915744A (en) | Near infrared spectrum nondestructive testing method and device for material component content | |
| Andrade et al. | Procrustes rotation in analytical chemistry, a tutorial | |
| Branikas et al. | A novel data augmentation method for improved visual crack detection using generative adversarial networks | |
| CN113970532B (en) | Fabric fiber component detection system and prediction method based on near infrared spectrum | |
| Camacho et al. | An analysis of spectral variability in hyperspectral imagery: a case study of stressed oil palm detection in Colombia | |
| CN114088657B (en) | Textile fiber component analysis method based on depth encoder/decoder | |
| CN113970528B (en) | Textile component mixing method based on complete constraint least square method | |
| CN111309958A (en) | Spectrum reconstruction method based on interpolation operation | |
| CN114994069A (en) | Hyperspectrum-based method and system for detecting contamination components and content of insulator | |
| CN114359229A (en) | Defect detection method based on DSC-UNET model | |
| CN110658156B (en) | Near infrared spectrum feature extraction method and device | |
| CN112966789A (en) | Tobacco maturity identification method, device and equipment | |
| CN113902695A (en) | Detection method for yarn-dyed fabric cut piece defect area | |
| CN113033623A (en) | Pollution source identification method and system based on ultraviolet-visible absorption spectrum | |
| Ramesh et al. | Identification of Timber Defects Using Convolution Neural Network | |
| Tu et al. | A Python-OpenCV based software for processing single-bacterium tracking microscopy videos | |
| Vitale et al. | Hyperspectral Video Analysis by Motion and Intensity Preprocessing and Subspace Autoencoding | |
| CN114324294A (en) | Multi-probe gas Raman spectrum system and mixed gas quantitative analysis method | |
| CN113656441B (en) | Printing and dyeing color matching data set construction method, device, equipment and medium | |
| CN117871459A (en) | Mutton crude fat content determination method and system | |
| CN117269074A (en) | Method and system for predicting chlorophyll content of fir based on hyperspectrum |
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 |