Method and device for discriminating amylopectin and amylose based on mid-infrared spectrum
Through the deep learning model based on mid-infrared spectroscopy, the problem of poor discrimination effect of amylopectin and amylose is solved, and efficient and automated starch type recognition is achieved.
Patent Information
- Application Number
- CN202510554498.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-29
- Publication Date
- 2025-08-12
AI Technical Summary
The existing methods for discrimination of amylose and amylose have poor results and are difficult to achieve high-throughput and automated detection in complex environments.
A deep learning model based on mid-infrared spectroscopy is adopted, including one-dimensional convolutional layer, batch normalization layer, activation function layer, residual convolution block, global average pooling layer and fully connected layer. By pre-processing the mid-infrared spectral data, the starch category discrimination model is input, and the cross-entropy loss function is trained to obtain the starch category discrimination results.
It improves the accuracy of discrimination between amylopectin and amylose, and achieves efficient and automated starch type recognition.
Smart Images
Figure CN120473024A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of starch category identification, and in particular to a method and device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy. Background Art
[0002] Starch is a common and important polysaccharide used in food, medicine, and chemical industries. Amylopectin and amylose differ in their molecular structure and physicochemical properties. The ratio of these different starch types significantly influences food texture, viscosity, gel strength, and performance in pharmaceutical and materials applications. In the winemaking industry, the amylopectin and amylose content in grains directly impacts the quality and flavor of the resulting wine. Therefore, the rapid and accurate identification of amylopectin and amylose is crucial for production process optimization and quality control.
[0003] Determining starch types primarily relies on chemical reagent testing or polarizing microscopy. However, these methods are often cumbersome and subjective, and struggle to achieve high-throughput, automated detection in complex environments. Mid-infrared spectroscopy is widely used in compositional analysis due to its high sensitivity to molecular vibrational characteristics. However, starch spectral signals are susceptible to factors such as environmental noise and baseline drift, and identifying and effectively distinguishing the characteristic absorption peaks of different starch types is challenging. Consequently, traditional numerical analysis methods, such as partial least squares, are less effective in determining starch types based on mid-infrared spectroscopy. Summary of the Invention
[0004] Technical problem solved by the present invention: The present invention provides a method and device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, which solves the problem of poor discrimination effect of existing amylopectin and amylose.
[0005] The present invention solves the above technical problems by adopting a technical solution: a method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, comprising the following steps:
[0006] S1. Obtain mid-infrared spectrum data of the starch sample to be evaluated;
[0007] S2. Inputting the mid-infrared spectrum data of the starch sample to be evaluated into the starch classification discrimination model to obtain the starch classification discrimination result.
[0008] Furthermore, in S1, the mid-infrared spectral data includes wavenumber and transmittance, and the wavenumber range is 4000cm -1 Up to 675cm -1 .
[0009] Furthermore, S1 also includes preprocessing the mid-infrared spectral data, and the preprocessing includes denoising, baseline correction, smoothing and standardization.
[0010] Furthermore, in S2, the starch category discrimination model is a deep learning model, including a one-dimensional convolution layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks, a global average pooling layer and a fully connected layer connected in sequence, wherein the one-dimensional convolution layer is used to extract features, the one-dimensional batch normalization layer is used for stable training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output.
[0011] Furthermore, the starch classification model includes a full spectrum discrimination model, a COH bending vibration characteristic region discrimination model, and a COC sugar ring vibration characteristic region discrimination model. The input data of the full spectrum discrimination model is a wave number range of 4000 cm -1 Up to 675cm -1 The input data of the COH bending vibration characteristic region discrimination model is in the wave number range of 1300 cm -1 Up to 1200cm -1 The input data of the COC sugar ring vibration characteristic region discrimination model is the wave number range of 1150cm -1 Up to 950cm -1 spectral data.
[0012] Furthermore, the starch category discrimination model is weighted according to the prediction results of the full spectrum discrimination model, the COH bending vibration characteristic region discrimination model and the COC sugar ring vibration characteristic region discrimination model to obtain the discrimination result of the starch category.
[0013] Furthermore, the full spectrum discrimination model, the COH bending vibration feature area discrimination model and the COC sugar ring vibration feature area discrimination model all include a one-dimensional convolution layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks and a global average pooling layer, and share the same fully connected layer. The one-dimensional convolution layer is used to extract features, the one-dimensional batch normalization layer is used for stable training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output.
[0014] Furthermore, in S2, the starch classification model uses the cross entropy loss function, and the expression of the cross entropy loss function is: Among them, Loss represents the cross entropy loss function value, M is the number of samples; P ij is the true category label of the j-th starch category corresponding to the i-th sample; is the probability that the i-th sample is predicted to be the j-th starch category.
[0015] Furthermore, when training the starch category discrimination model, the mid-infrared spectral data and the corresponding starch category are used as data sets and divided into a training set and a test set. The training set is used to train the starch category discrimination model, and the test set is used to evaluate the training effect of the starch category discrimination model. When the accuracy of the starch category discrimination model on the test set is higher than the preset value, the starch category discrimination model training is completed.
[0016] The present invention also provides a device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, which implements the above-mentioned method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy. The device includes a data acquisition module, a model training module and a category evaluation module. The data acquisition module is used to obtain mid-infrared spectral data of the starch sample to be evaluated and the data required for model training. The data required for model training includes mid-infrared spectral data and corresponding starch categories. The model training module is used to train the model using the data required for model training to obtain a starch category discrimination model. The category evaluation module is used to determine the category of the starch sample to be evaluated using the starch category discrimination model based on the mid-infrared spectral data of the starch sample to be evaluated.
[0017] Beneficial effects of the present invention: The present invention provides a method and device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy. By inputting the mid-infrared spectral data of the starch sample to be evaluated into a starch category discrimination model, which is a deep learning model, the starch category discrimination result is obtained, which solves the problem of poor discrimination effect of amylopectin and amylose in the existing field. BRIEF DESCRIPTION OF THE DRAWINGS
[0018] Figure 1 This is a schematic flow chart of a method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention;
[0019] Figure 2 This is a comparison diagram of mid-infrared spectra of amylopectin and amylose collected in a method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention;
[0020] Figure 3 This is a schematic structural diagram of a starch classification model in a method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention;
[0021] Figure 4 This is a schematic diagram of the classification results of the test set using the partial least squares method;
[0022] Figure 5 It is the distribution diagram of the classification results of the test set using the partial least squares method;
[0023] Figure 6This is a curve diagram of the change of the cross entropy loss function during the training process of the starch classification discrimination model in the method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention;
[0024] Figure 7 This is a schematic diagram of the classification results of a test set by a starch classification model in a method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention. DETAILED DESCRIPTION
[0025] In response to the problem of poor discrimination effect between amylopectin and amylose in existing technologies, the present invention provides a method and device for discriminating amylopectin and amylose based on mid-infrared spectroscopy. By inputting the mid-infrared spectral data of the starch sample to be evaluated into a starch classification discrimination model, the starch classification discrimination result is obtained. Compared with the partial least squares method, the method has higher accuracy in starch type determination based on mid-infrared spectroscopy.
[0026] like Figure 1 As shown, the method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy provided by the present invention comprises the following steps:
[0027] S1. Obtain mid-infrared spectral data of the starch sample to be evaluated.
[0028] Specifically, a Fourier transform infrared spectrometer was used to obtain mid-infrared spectral data of the starch sample to be evaluated. The mid-infrared spectral data included wavenumber and transmittance, and the wavenumber range was 4000 cm -1 Up to 675cm -1 , the wavelength range corresponding to the wave number is 2500nm to 14800nm, and the mid-infrared spectra of amylopectin and amylose are obtained, such as Figure 2 As shown, the two are quite similar and it is difficult to distinguish them only by naked eyes. Therefore, the present invention adopts a starch category discrimination model for discrimination.
[0029] S2. Inputting the mid-infrared spectrum data of the starch sample to be evaluated into the starch classification discrimination model to obtain the starch classification discrimination result.
[0030] Specifically, the starch classification model is a deep learning model, such as Figure 3 As shown, it includes a one-dimensional convolution layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks, a global average pooling layer and a fully connected layer connected in sequence, wherein the one-dimensional convolution layer is used to extract features, the one-dimensional batch normalization layer is used to stabilize training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output.
[0031] The starch classification model uses the cross entropy loss function, and the expression of the cross entropy loss function is:
[0032] Among them, Loss represents the cross entropy loss function value, M is the number of samples; P ij is the true category label of the j-th starch category corresponding to the i-th sample; is the probability that the i-th sample is predicted to be the j-th starch category.
[0033] When training the starch category discrimination model, the mid-infrared spectral data and the corresponding starch category are used as data sets and divided into a training set and a test set. The training set is used to train the starch category discrimination model, and the test set is used to evaluate the training effect of the starch category discrimination model. When the accuracy of the starch category discrimination model on the test set is higher than the preset value, the starch category discrimination model training is completed.
[0034] Furthermore, the starch classification model can also include a full spectrum discrimination model, a COH bending vibration characteristic region discrimination model, and a COC sugar ring vibration characteristic region discrimination model. The input data of the full spectrum discrimination model is a wave number range of 4000 cm -1 Up to 675cm -1 The input data of the COH bending vibration characteristic region discrimination model is in the wave number range of 1300 cm -1 to
[0035] 1200cm -1 The input data of the COC sugar ring vibration characteristic region discrimination model is the wave number range of 1150cm -1 Up to 950cm -1 The starch classification discrimination model is weighted according to the prediction results of the full spectrum discrimination model, the COH bending vibration characteristic region discrimination model, and the COC sugar ring vibration characteristic region discrimination model to obtain the discrimination result of the starch classification.
[0036] The full spectrum discrimination model, COH bending vibration characteristic area discrimination model and COC sugar ring vibration characteristic area discrimination model are sub-models of the starch category discrimination model. Their model structures are similar to those of the starch category discrimination model. The only difference is that different sub-models process different mid-infrared spectral data segments.
[0037] Furthermore, the full spectrum discrimination model, COH bending vibration feature area discrimination model, and COC sugar ring vibration feature area discrimination model all include a one-dimensional convolution layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks, and a global average pooling layer, and share the same fully connected layer. The one-dimensional convolution layer is used to extract features, the one-dimensional batch normalization layer is used for stable training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output. In this way, the full spectrum discrimination model, COH bending vibration feature area discrimination model, and COC sugar ring vibration feature area discrimination model are fused through the same fully connected layer to obtain the discrimination result of starch category.
[0038] The present invention also provides a device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, which implements the above-mentioned method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy. The device includes a data acquisition module, a model training module and a category evaluation module. The data acquisition module is used to obtain mid-infrared spectral data of the starch sample to be evaluated and the data required for model training. The data required for model training includes mid-infrared spectral data and corresponding starch categories. The model training module is used to train the model using the data required for model training to obtain a starch category discrimination model. The category evaluation module is used to determine the category of the starch sample to be evaluated using the starch category discrimination model based on the mid-infrared spectral data of the starch sample to be evaluated.
[0039] With 157 amylose data and 143 amylopectin data, a total of 300 sample data, if the partial least squares method is used for fitting analysis, 35 amylose data and 25 amylopectin data are taken as the test set, and the rest are used as the training set. The categories are mapped to 0 for amylose and 1 for amylopectin. After the training is completed, the predicted value is judged by a threshold of 0.5 to achieve binary classification. The classification results of the test set are as follows Figure 4 As shown, the accuracy is only 60%, and the distribution of the test set is as follows Figure 5 shown.
[0040] Using the mid-infrared spectroscopy-based amylopectin and amylose discrimination method provided by the present invention, during the starch classification discrimination model training process, the cross entropy loss function change curve is as follows: Figure 6 As shown, the classification results of the test set are as follows Figure 7 As shown, the accuracy is 100%, which is better than the partial least squares method.
Claims
1. A method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, characterized in that: The following steps are involved: S1. Obtain mid-infrared spectrum data of the starch sample to be evaluated; S2. Inputting the mid-infrared spectrum data of the starch sample to be evaluated into the starch classification discrimination model to obtain the starch classification discrimination result.
2. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 1, characterized in that: In S1, the mid-infrared spectral data includes wave number and transmittance, and the wave number range is 4000cm -1 Up to 675cm -1 .
3. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 1, characterized in that: S1 also includes preprocessing the mid-infrared spectral data, which includes denoising, baseline correction, smoothing and standardization.
4. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 1, wherein: In S2, the starch category discrimination model is a deep learning model, including a one-dimensional convolutional layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks, a global average pooling layer and a fully connected layer connected in sequence. The one-dimensional convolutional layer is used to extract features, the one-dimensional batch normalization layer is used for stable training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output.
5. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 1, characterized in that: The starch classification model includes the full spectrum discrimination model, the COH bending vibration characteristic area discrimination model and the COC sugar ring vibration characteristic area discrimination model. The input data of the full spectrum discrimination model is the wave number range of 4000 cm -1 Up to 675cm -1 The spectral data, The input data of the COH bending vibration characteristic region discrimination model is in the wave number range of 1300 cm -1 Up to 1200cm -1 The input data of the COC sugar ring vibration characteristic region discrimination model is the wave number range of 1150cm -1 Up to 950cm -1 spectral data.
6. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 5, characterized in that: The starch category discrimination model is weighted according to the prediction results of the full spectrum discrimination model, the COH bending vibration characteristic region discrimination model and the COC sugar ring vibration characteristic region discrimination model to obtain the discrimination result of the starch category.
7. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to claim 5, characterized in that: The full spectrum discrimination model, COH bending vibration feature area discrimination model and COC sugar ring vibration feature area discrimination model all include a one-dimensional convolution layer, a one-dimensional batch normalization layer, an activation function layer, multiple one-dimensional residual convolution blocks and a global average pooling layer, and share the same fully connected layer. The one-dimensional convolution layer is used to extract features, the one-dimensional batch normalization layer is used for stable training, the activation function layer is used for nonlinear activation, the multiple one-dimensional residual convolution blocks are used for deep feature modeling, the global average pooling layer is used for dimensionality reduction, and the fully connected layer is used for classification output.
8. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to any one of claims 1 to 7, characterized in that: In S2, the starch classification model uses the cross entropy loss function, and the expression of the cross entropy loss function is: Among them, Loss represents the cross entropy loss function value, M is the number of samples; P ij is the true category label of the j-th starch category corresponding to the i-th sample; is the probability that the i-th sample is predicted to be the j-th starch category.
9. The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy according to any one of claims 1 to 7, characterized in that: When training the starch category discrimination model, the mid-infrared spectral data and the corresponding starch category are used as data sets and divided into a training set and a test set. The training set is used to train the starch category discrimination model, and the test set is used to evaluate the training effect of the starch category discrimination model. When the accuracy of the starch category discrimination model on the test set is higher than the preset value, the starch category discrimination model training is completed.
10. A device for distinguishing amylopectin and amylose based on mid-infrared spectroscopy, characterized in that: The method for distinguishing amylopectin and amylose based on mid-infrared spectroscopy as described in claim 1 is implemented, and the device includes a data acquisition module, a model training module and a category evaluation module, the data acquisition module is used to obtain mid-infrared spectral data of the starch sample to be evaluated and the data required for model training, the data required for model training includes mid-infrared spectral data and the corresponding starch category, the model training module is used to train the model using the data required for model training to obtain a starch category discrimination model, and the category evaluation module is used to determine the category of the starch sample to be evaluated based on the mid-infrared spectral data of the starch sample to be evaluated using the starch category discrimination model.