Thickness measurement and composition identification method of co-extruded composite films based on infrared spectroscopy

Abstract

The present invention provides a method for measuring the thickness and identifying the components of a co-extruded composite film based on infrared spectroscopy. The infrared spectrum of the co-extruded composite film is used as the object, and has the advantages of being fast, non-destructive, and easy to operate. At the same time, the infrared spectrum is rich in information. Based on the one-dimensional convolutional neural network (1D-CNN) model, a 1D-CNN quantitative model of the co-extruded composite film is first established, and high-precision thickness measurement is achieved by analyzing the infrared spectrum data of the co-extruded composite film. Then, on this basis, the 1D-CNN quantitative model of the co-extruded composite film is migrated to construct a 1D-CNN classification model of the co-extruded composite film. This method significantly improves the efficiency of constructing a component identification model without increasing the computational complexity through transfer learning technology. The entire process can not only effectively improve the recognition accuracy, but also optimize the model training speed, and realize the efficient and accurate dual functions of thickness measurement and component identification of the co-extruded composite film.

Description

Technical Field

[0001] The invention relates to the technical field of film thickness measurement and component identification, and in particular to a method for measuring the thickness of a co-extruded composite film and identifying its components based on infrared spectroscopy. Background Art

[0002] Composite plastic film is made of multiple films, combining the advantages of different materials, and is widely used in daily use, agriculture and industry. Common composite plastic films are made of multiple layers of different materials co-extruded into composite plastic films. Currently, there are hundreds of different components of common composite plastic films. For this type of composite plastic film finished product, each layer is a transparent polymer, and it is basically impossible to directly determine the type of each layer and the thickness of each component by visual inspection.

[0003] Classification and identification of such composite plastic films is of great significance for safety issues and recycling of waste composite plastic films. For example, if the wrong type of composite plastic film is used in the production, processing and consumption of plastic products, it will cause serious safety accidents. In addition, the component thickness of the composite plastic film is the basis for affecting other performance indicators of the film. If the thickness of a component of the produced composite plastic film does not meet the standard, on the one hand, it will have a negative impact on the performance of the film (such as air tightness, etc.), and on the other hand, the composite plastic film with unsatisfactory component thickness will flow into the next production, processing and consumption link, which may cause safety problems.

[0004] Plastic classification and discrimination methods usually use density discrimination and Raman spectroscopy. Before the sinking and floating test, the density discrimination method requires the sample to be sliced ​​and pre-processed, and this pre-processing process usually takes a lot of time and is difficult to meet the needs of efficient detection. The known problems are: 1. The pre-processing is cumbersome; 2. The classification and discrimination accuracy is poor. Raman spectroscopy analyzes and discriminates the differential Raman spectrum of the sample, combines the artificial neural network algorithm to build a classification model and classify. The known problems are: 1. It is susceptible to fluorescence interference; 2. The sample composition is prone to change.

[0005] Plastic thickness measurement methods usually use X-ray thickness measurement and laser thickness measurement. The detection principle of X-ray thickness measurement is based on the principle that the material to be measured will absorb radiation, and the thickness of the material to be measured can be inferred based on the amount of radiation absorbed. The known problems are: 1. Radiation of radiation will pollute the environment and threaten human health; 2. It is not suitable for use in crowded places. Laser thickness measurement uses the principle that infrared laser will attenuate after penetrating the object to be measured, and calculates the thickness information of the object to be measured by measuring the weakening of the laser intensity. The known problems are: 1. Low measurement accuracy; 2. Short service life of laser.

[0006] Today, infrared spectroscopy is often used as an analytical technique to identify and quantify chemical substances, with the advantages of extremely high sensitivity, fast analysis speed, low sample preparation requirements and high resolution. It is also a non-destructive method that can collect chemical information without damaging the sample. Therefore, infrared spectroscopy has the potential to be used to determine the chemical characteristics of composite plastic films. By establishing a calibration model for infrared spectroscopy, the thickness of each component of the composite plastic film can be quickly measured and the category of the composite plastic film can be identified.

[0007] In recent years, the application of convolutional neural network (CNN) in infrared spectroscopy analysis has gradually attracted attention. Different characteristic peaks in the infrared spectrum correspond to different chemical bonds, and the convolution layer can automatically capture these features through window sliding. Its good local perception and feature processing capabilities make one-dimensional convolutional neural network (1D-CNN) an effective tool for accurate identification and quantitative analysis of chemical components. In order to overcome the problems of cumbersome preprocessing, susceptibility to interference in the measurement process, poor classification and discrimination accuracy, radiation hazards in quantitative methods, and short service life of measuring instruments in traditional composite plastic film classification methods, this application proposes a 1D-CNN model that can quickly and efficiently quantify the thickness of each component of composite plastic film and a 1D-CNN model that can identify different types of composite plastic films. Summary of the invention

[0008] The purpose of the present invention is to provide a method for identifying the components of co-extruded composite films and measuring their thickness based on infrared spectroscopy, which combines infrared spectral data with a one-dimensional convolutional neural network (1D-CNN: One Dimensional ConvolutionalNeural Network) and is applied to the identification of the components of co-extruded composite films and the measurement of their thickness. The method can quickly and accurately identify different co-extruded composite films and accurately measure the thickness of co-extruded composite films, thereby overcoming the common problems existing in traditional identification and measurement methods, such as cumbersome pretreatment, susceptibility to fluorescence interference, radiation pollution, and low measurement accuracy.

[0009] The first aspect of the present invention provides a method for identifying components and measuring thickness of a co-extruded composite film based on infrared spectroscopy, which specifically comprises the following steps:

[0010] Step 1: Collect infrared spectral data of co-extruded composite films of different thicknesses, preprocess the infrared spectral data and divide them into training set, validation set and prediction set.

[0011] Step 2: Build a 1D-CNN quantitative model.

[0012] Step 3: Train, verify and test the quantitative model to obtain the optimal 1D-CNN quantitative model, and use the quantitative model to quantitatively predict the thickness of the co-extruded composite film components with unidentified thickness.

[0013] Step 4: In response to the classification requirements of different film types, the collected infrared spectral data are further divided according to the film types, and the 1D-CNN quantitative model is migrated to a 1D-CNN classification model.

[0014] Step 5: Divide the re-divided infrared spectrum data into training set, validation set and prediction set.

[0015] Step 6: Train, verify and test the 1D-CNN classification model to obtain the optimal 1D-CNN classification model, and classify and identify the co-extruded composite film of unidentified category through the 1D-CNN classification model.

[0016] Preferably, the specific method of collecting infrared spectrum data of co-extruded composite films of different thicknesses in step 1 is: fixing the co-extruded composite films of different thicknesses on a sample holder, and making their surfaces perpendicular to the optical path of the infrared spectrometer, and transmitting scanning the samples by the infrared spectrum detection system, and taking the spectral average of each sample as the final sample spectrum.

[0017] Preferably, in step 1, the infrared spectrum is preprocessed using a multivariate scattering correction method, and the conversion function calculation process is as follows:

[0018] First, calculate the average spectrum of all sample infrared spectra: ;

[0019] Then, the infrared spectrum of each sample was subjected to a linear regression operation with the average spectrum to obtain the intercept constant and the regression coefficient : ;

[0020] Finally, we get the conversion function: ;

[0021] in, For the Sample spectra, is the total number of sample spectra.

[0022] Preferably, in step 1, the training set and the prediction set are generated using the Kennard-Stone method, which is specifically divided into: using the KS method to select 20% of the preprocessed infrared spectral data as the test set of the 1D-CNN quantitative model, and the remaining 80% of the infrared spectral data are divided into a 20% validation set and an 80% training set using the KS method.

[0023] Preferably, the step 3 specifically includes:

[0024] Step 3.1: Set the 1D-CNN quantitative model hyperparameters and initialize the weights.

[0025] Step 3.2: Train the 1D-CNN quantitative model, calculate the error through forward propagation, and determine whether the set number of iterations has been reached. If not, backpropagate the error to calculate and update the parameters of each layer. When the number of iterations has been reached, end the training, solidify the parameters of each layer of the 1D-CNN quantitative model, and save them.

[0026] Step 3.3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters.

[0027] Step 3.4: Use the obtained quantitative model to quantitatively predict the thickness of the co-extruded composite film component with unidentified thickness.

[0028] Preferably, the specific operation of migrating the 1D-CNN quantitative model to the 1D-CNN classification model in step 4 is: freezing the 1D-CNN quantitative model training parameters of all layers except the output layer to ensure that the parameters of each layer except the output layer remain unchanged when training the 1D-CNN classification model. Changing the activation function of the output layer to a softmax function, and modifying the size of the output layer to correspond to the infrared spectrum data of different film types.

[0029] Preferably, in step 5, 20% of the preprocessed infrared spectral data are selected as the test set of the 1D-CNN classification model using the KS method, and the remaining 80% of the infrared spectral data are divided into a 20% validation set and an 80% training set using the KS method.

[0030] Preferably, the step 6 specifically includes:

[0031] Step 6.1: Set the hyperparameters of the 1D-CNN classification model and freeze the model parameters of all layers of the 1D-CNN quantitative model except the output layer.

[0032] Step 6.2: During the training process, the error is calculated through forward propagation to determine whether the set number of training iterations has been reached. If not, the error is back-propagated to calculate and update the parameters of each layer. When the number of training iterations has been reached, the training is terminated, and the parameters of each layer of the 1D-CNN classification model are solidified and saved.

[0033] Step 6.3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters.

[0034] Step 6.4: Classify and identify the co-extruded composite film of unidentified category through the obtained classification model.

[0035] The second aspect of the present invention provides an electronic device for measuring the thickness and identifying the components of co-extruded composite films based on infrared spectroscopy, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the above-mentioned method for measuring the thickness and identifying the components of co-extruded composite films based on infrared spectroscopy is implemented.

[0036] A third aspect of the present invention provides a computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program is used to execute the above-mentioned method for measuring the thickness and identifying the components of a co-extruded composite film based on infrared spectroscopy.

[0037] Beneficial effects of the present invention: The present invention uses the infrared spectrum of the co-extruded composite film as the research object, which has the advantages of being fast, non-destructive, and easy to operate, and the infrared spectrum is rich in information. Based on the one-dimensional convolutional neural network (1D-CNN) model, a 1D-CNN quantitative model of the co-extruded composite film is first established, and high-precision thickness measurement is achieved by analyzing the infrared spectrum data of the co-extruded composite film. Then, on this basis, the 1D-CNN quantitative model of the co-extruded composite film is migrated to construct a 1D-CNN classification model of the co-extruded composite film. This method significantly improves the efficiency of building the component identification model through transfer learning technology without increasing the computational complexity. The entire process can not only effectively improve the recognition accuracy, but also optimize the model training speed, and realize the efficient and accurate dual functions of thickness measurement and component identification of the co-extruded composite film. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a schematic diagram of the infrared spectrum sampling structure of the co-extruded composite film in an embodiment of the present application.

[0039] Figure 2 This is a schematic diagram of the core structure of the D-CNN model in Example 1 of the present application.

[0040] Figure 3 This is a flow chart of step S5 of an embodiment of the present application.

[0041] Figure 4 This is a flow chart of step S9 of an embodiment of the present application. DETAILED DESCRIPTION

[0042] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0043] like Figure 1-Figure 4 As shown, this application uses a one-dimensional convolutional neural network model (1D-CNN) to study the infrared spectrum data of co-extruded composite films, and designs a quantitative prediction and classification recognition scheme for the thickness of co-extruded composite films. The specific implementation steps are as follows:

[0044] S1: Collect infrared spectra of composite plastic films of different thicknesses

[0045] This preferred embodiment uses five kinds of films, PE, PP, PC, PVC and PET, to combine. Figure 1 As shown, the co-extruded composite film is fixed on the sample holder, and its surface is perpendicular to the optical path of the infrared spectrometer, and the infrared spectrum detection system transmits and scans the sample.

[0046] Set the system resolution to 4cm -1 , 32 spectra were scanned for each sample and the average was taken as the final sample spectrum. Finally, 875 infrared spectrum samples of co-extruded composite films were obtained as the infrared spectrum data of this embodiment. The thickness corresponding to 1 to 5 layers of each film is shown in Table 1.

[0047] Table 1. Film thickness correspondence table

[0048]

[0049] S2: Preprocessing the infrared spectrum data

[0050] This embodiment uses a multivariate scatter correction method to preprocess the collected data, and the conversion function calculation process is as follows:

[0051] First, calculate the average spectrum of all sample infrared spectra: ;

[0052] Then, the infrared spectrum of each sample was subjected to a linear regression operation with the average spectrum to obtain the intercept constant and the regression coefficient : ;

[0053] Finally, we get the conversion function: ;

[0054] in, For the Sample spectra, is the total number of sample spectra.

[0055] S3: Divide the preprocessed infrared spectral data into training set, validation set and prediction set

[0056] This embodiment uses the Kennard-Stone method to select 20% of the pre-processed data as the test set of the 1D-CNN quantitative model, and the remaining 80% of the infrared spectrum data is divided into a 20% validation set and an 80% training set using the KS method. Therefore, the sample data is divided into 3 data sets, namely, a training set, a validation set, and a test set. Among them, the training set contains 560 samples, the validation set contains 140 samples, and the prediction set contains 175 samples. The model is established using the training set data, the validation set data is used as the model selection, and the test set is used to finally verify the performance of the model.

[0057] S4: Building a 1D-CNN quantitative model

[0058] like Figure 2 As shown, the 1D-CNN (One Dimensional Convolutional Neural Network) quantitative model of this embodiment consists of an input layer, a hidden layer and an output layer. The hidden layer includes a convolution layer, a pooling layer, a flattening layer and a fully connected layer.

[0059] S5: Training, verification and testing of 1D-CNN quantitative model based on infrared spectroscopy data of co-extruded composite films

[0060] like Figure 3 As shown, the S5 specifically includes:

[0061] S5-1: Set the 1D-CNN quantitative model hyperparameters and initialize the weights.

[0062] S5-2: Train the 1D-CNN quantitative model, calculate the error through forward propagation, and determine whether the set number of iterations has been reached. If not, backpropagate the error to calculate and update the parameters of each layer. When the number of iterations has been reached, end the training, solidify the parameters of each layer of the 1D-CNN quantitative model, and save them.

[0063] S5-3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters.

[0064] S5-4: The obtained quantitative model is used to quantitatively predict the thickness of the co-extruded composite film component with unidentified thickness, and finally the quantitative result is output.

[0065] S6: According to the classification requirements of different film types, the collected infrared spectrum data is further divided by film type.

[0066] The infrared spectrum of the co-extruded composite film is collected. The co-extruded composite film is composed of five films of different thicknesses. According to the different types of films contained in each co-extruded composite film, the acquired infrared spectrum data of the co-extruded composite film can be divided into 31 categories.

[0067] S7: Migrating 1D-CNN quantitative model to 1D-CNN classification model

[0068] Freeze the training parameters of the 1D-CNN quantitative model in all layers except the output layer to ensure that the parameters of each layer except the output layer remain unchanged when training the 1D-CNN classification model.

[0069] The activation function of the output layer is changed to the softmax function, and the size of the output layer is adjusted to 31 to represent the combination of five different types of plastic films.

[0070] The number of trainable parameters of the 1D-CNN quantitative model is 3615717, and the number of trainable parameters after migration to the 1D-CNN classification model is 7967.

[0071] S8: Divide the re-partitioned infrared spectral data for the 1D-CNN classification model into training set, validation set and prediction set

[0072] In this embodiment, the KS method is used to select 20% of the preprocessed data as the test set of the 1D-CNN classification model, and the KS method is used to divide the remaining 80% of the infrared spectrum data into a 20% validation set and an 80% training set. Therefore, the sample data is divided into 3 data sets, namely, a training set, a validation set and a test set, wherein the training set contains 560 samples, the validation set contains 140 samples, and the prediction set contains 175 samples. The model is established using the training set data, the validation set data is used as the model selection, and the test set is used to finally verify the classification performance of the model.

[0073] S9: Training, validation and testing of 1D-CNN classification model based on infrared spectroscopy data of co-extruded composite films

[0074] like Figure 4 As shown, the S9 specifically includes:

[0075] S9-1: Set the hyperparameters of the 1D-CNN classification model and freeze the model parameters of all layers of the quantitative model except the output layer.

[0076] S9-2: During the training process, the error is calculated through forward propagation to determine whether the set number of training iterations has been reached. If not, the error is back-propagated to calculate and update the parameters of each layer. When the number of training iterations has been reached, the training is terminated, and the parameters of each layer of the 1D-CNN classification model are solidified and saved.

[0077] S9-3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters.

[0078] S9-4: Classify and identify the co-extruded composite film of unidentified category through the obtained classification model, and finally output the classification result.

[0079] The embodiment of the present application also discloses an electronic device for measuring the thickness of a co-extruded composite film and identifying its components based on infrared spectroscopy, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for measuring the thickness of a co-extruded composite film and identifying its components based on infrared spectroscopy are implemented. The memory may include a memory, such as a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk memory, etc. The processor, the network interface, and the memory are interconnected through an internal bus, which may be an industrial standard architecture bus, a peripheral component interconnection standard bus, an extended industrial standard architecture bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory is used to store programs. Specifically, the program may include a program code, and the program code includes computer operation instructions. The memory may include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0080] The embodiment of the present application also discloses a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the steps of the method for measuring the thickness of a co-extruded composite film and identifying its components based on infrared spectroscopy are implemented. Specifically, the computer-readable storage medium includes, but is not limited to, for example, volatile memory and / or non-volatile memory. The volatile memory may include a random access memory (RAM) and / or a cache memory, etc. The non-volatile memory may include a read-only memory (ROM), a hard disk, a flash memory, an optical disk, a magnetic disk, etc.

[0081] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for measuring the thickness of co-extruded composite films and identifying their components based on infrared spectroscopy, characterized in that: The specific steps include: Step 1: Collect infrared spectrum data of co-extruded composite films of different thicknesses, pre-process the infrared spectrum data and divide them into training set, verification set and prediction set; Step 2: Build a 1D-CNN quantitative model; Step 3: training, verifying and testing the quantitative model to obtain an optimal 1D-CNN quantitative model, and using the quantitative model to quantitatively predict the thickness of the co-extruded composite film components whose thickness has not been identified; Step 4: According to the classification requirements of different film types, the collected infrared spectrum data is further divided by film type, and the 1D-CNN quantitative model is migrated to a 1D-CNN classification model; Step 5: Divide the re-divided infrared spectrum data into a training set, a validation set, and a prediction set; Step 6: training, verifying and testing the 1D-CNN classification model to obtain an optimal 1D-CNN classification model, and classifying and identifying the co-extruded composite film of an unidentified category by using the 1D-CNN classification model; The specific method of collecting infrared spectrum data of co-extruded composite films of different thicknesses in step 1 is: Fix the co-extruded composite films of different thicknesses on the sample holder, and make their surfaces perpendicular to the optical path of the infrared spectrometer. The infrared spectrum detection system transmits and scans the samples, and takes the average spectrum of each sample as the final sample spectrum. In step 1, the infrared spectrum is preprocessed using a multivariate scattering correction method, and the conversion function calculation process is as follows: First, calculate the average spectrum of all sample infrared spectra: ; Then, the infrared spectrum of each sample was subjected to a linear regression operation with the average spectrum to obtain the intercept constant and the regression coefficient : ; Finally, we get the conversion function: ; in, is the spectrum of the th sample, is the total number of sample spectra; The specific operation of migrating the 1D-CNN quantitative model to the 1D-CNN classification model in step 4 is: Freeze the training parameters of the 1D-CNN quantitative model in all layers except the output layer to ensure that the parameters of each layer except the output layer remain unchanged when training the 1D-CNN classification model; The activation function of the output layer is changed to the softmax function, and the size of the output layer is modified to correspond to the infrared spectrum data of different film types.

2. The method for measuring thickness and identifying components of co-extruded composite films based on infrared spectroscopy according to claim 1, characterized in that: In step 1, the training set and the prediction set are generated using the Kennard-Stone method, which is specifically divided into: The KS method was used to select 20% of the preprocessed infrared spectral data as the test set of the 1D-CNN quantitative model, and the remaining 80% of the infrared spectral data was divided into a 20% validation set and an 80% training set using the KS method.

3. The method for measuring thickness and identifying components of co-extruded composite films based on infrared spectroscopy according to claim 1, characterized in that: The step 3 specifically includes: Step 3.1: Set the hyperparameters of the 1D-CNN quantitative model and initialize the weights; Step 3.2: Train the 1D-CNN quantitative model, calculate the error through forward propagation, and determine whether the set number of iterations has been reached. If not, backpropagate the error to calculate and update the parameters of each layer. When the number of iterations has been reached, end the training, solidify the parameters of each layer of the 1D-CNN quantitative model, and save them. Step 3.3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters; Step 3.4: Use the obtained quantitative model to quantitatively predict the thickness of the co-extruded composite film component with unidentified thickness.

4. The method for measuring thickness and identifying components of co-extruded composite films based on infrared spectroscopy according to claim 1, characterized in that: In step 5, 20% of the preprocessed infrared spectral data are selected as the test set of the 1D-CNN classification model using the KS method, and the remaining 80% of the infrared spectral data are divided into a 20% validation set and an 80% training set using the KS method.

5. The method for measuring thickness and identifying components of co-extruded composite films based on infrared spectroscopy according to claim 1, characterized in that: The step 6 specifically includes: Step 6.1: Set the hyperparameters of the 1D-CNN classification model and freeze the model parameters of all layers of the 1D-CNN quantitative model except the output layer; Step 6.2: During the training process, the error is calculated by forward propagation to determine whether the set number of training iterations has been reached. If not, the error is back-propagated to calculate and update the parameters of each layer. When the number of training iterations has been reached, the training is terminated, and the parameters of each layer of the 1D-CNN classification model are solidified and saved. Step 6.3: Enter the model testing phase, read the model file saved in the training phase, and restore the model parameters; Step 6.4: Classify and identify the co-extruded composite film of unidentified category through the obtained classification model.

6. An electronic device for measuring the thickness of co-extruded composite films and identifying their components based on infrared spectroscopy, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for measuring the thickness of a co-extruded composite film and identifying its components based on infrared spectroscopy as described in any one of claims 1 to 5 is implemented.

7. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and the computer program is used to execute the co-extruded composite film thickness measurement and component identification method based on infrared spectroscopy as described in any one of claims 1 to 5.

Images