CN111950427A - Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model - Google Patents
Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model Download PDFInfo
- Publication number
- CN111950427A CN111950427A CN202010785928.0A CN202010785928A CN111950427A CN 111950427 A CN111950427 A CN 111950427A CN 202010785928 A CN202010785928 A CN 202010785928A CN 111950427 A CN111950427 A CN 111950427A
- Authority
- CN
- China
- Prior art keywords
- garnet
- reflection peak
- neural network
- network model
- thermal infrared
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
- G06F2218/10—Feature extraction by analysing the shape of a waveform, e.g. extracting parameters relating to peaks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic 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
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/048—Activation functions
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/12—Classification; Matching
Abstract
The invention discloses a method for identifying a garnet subclass based on thermal infrared spectrum characteristics and a BP neural network model. The method for identifying the sub-types of the garnet comprises the following steps: acquiring thermal infrared spectrum characteristic data of a garnet sample of a known sub type; constructing a BP neural network model by utilizing the subclass type and thermal infrared spectrum characteristic data of the garnet sample; and acquiring thermal infrared spectrum characteristic data of the garnet sample to be detected, inputting the constructed BP neural network model, and identifying the subclass type of the garnet sample to be detected. The method is based on rich information in the garnet thermal infrared spectrum and the nonlinear automatic mapping capability of the BP neural network model, performs the garnet subclass identification, provides a quick and effective technical support for the garnet subclass identification, and provides a technical inspiration for the quick identification of other minerals.
Description
Technical Field
The present invention relates to the field of garnet subclass classification. And more particularly, to a garnet subclass identification method based on thermal infrared spectroscopy and a BP neural network model.
Background
Garnet is a general name of silicate mineral with island structure and has a chemical formula X3Y2[ZO4]3Wherein X represents the divalent cations calcium, iron, magnesium, manganese, Y represents the trivalent cations aluminum, iron, chromium, manganese, and Z generally represents the tetravalent cation silicon. The presently known garnet group minerals can be classified into two types according to the characteristics of the cations at the X-position and the Y-position: calcium series garnet and aluminum series garnet. The calcium series garnet comprises calcium garnet, calcium iron garnet, calcium aluminum garnet, etc., and the aluminum series garnet comprises manganese aluminum garnet, iron aluminum garnet, magnesium aluminum garnet, etc. The traditional discrimination of the subclass classification of the garnet is mainly carried out through laboratory test analysis, and the analysis method has the defects of time and labor consumption, destructiveness, high cost and the like, so that a new method for rapidly and accurately discriminating the type of the garnet is urgently needed.
The hyperspectrum can capture a large amount of information, can reflect the complex inherent characteristics of a research object, has the advantages of nondestructive detection, time saving and labor saving, and is gradually applied to mineral identification in recent years. Hyperspectral research shows that the change of mineral compositions caused by metal ion substitution can be reflected on the spectrum. The sub-types of the garnet can be mutually converted through metal ion substitution, the hyperspectral reflection is the difference of absorption and reflection positions of the thermal infrared band of the garnet, and the wavelength of the absorption and reflection peak is regularly changed along with the component change of the garnet. The reflection peak wavelength characteristics of the calcium-chromium garnet and the manganese-aluminum garnet are obvious, and the reflection peak wavelengths of the iron-aluminum garnet, the magnesium-aluminum garnet, the calcium-iron garnet and the calcium-aluminum garnet have large overlapping regions and cannot be directly distinguished, so that a method capable of determining the mapping relation between the reflection peak wavelength of the thermal infrared spectrum section of the garnet and the type of the subclass is needed to be found.
Disclosure of Invention
The invention aims to provide a method for identifying a garnet subclass based on thermal infrared spectrum characteristics and a BP neural network model.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention provides a method for identifying a garnet subclass based on a thermal infrared spectrum characteristic and a BP neural network model, which comprises the following steps: the method is characterized in that the thermal infrared spectrum characteristics are combined with a BP neural network model for use, and the nonlinear automatic mapping capability of the BP neural network model is utilized to find a complex mapping relation between the wavelength of a reflection peak of a thermal infrared spectrum band of the garnet and the subclass type of the garnet, so that the subclass identification of the garnet is carried out, and the method specifically comprises the following steps (figure 1):
acquiring thermal infrared spectrum characteristic data of a garnet sample of a known sub type;
constructing a BP neural network model for identifying the sub-type of the garnet based on the thermal infrared spectrum characteristics by utilizing the thermal infrared spectrum characteristic data and the sub-type of the garnet sample;
and acquiring thermal infrared spectrum characteristic data of the garnet sample to be detected, inputting the constructed BP neural network model, and identifying the subclass type of the garnet sample to be detected.
Further, the thermal infrared spectral characteristic data is wavelength data of a reflection peak of the garnet sample in a spectrum band of 9 to 13 μm, including a first reflection peak (reflection peak 1) wavelength, a second reflection peak (reflection peak 2) wavelength, a third reflection peak (reflection peak 3) wavelength, and a wavelength difference between the third reflection peak and the second reflection peak, the third reflection peak wavelength being greater than the second reflection peak wavelength and greater than the first reflection peak wavelength; the sub-types of garnet include calcium garnet, manganese garnet, iron garnet, magnesium garnet, calcium iron garnet, and calcium aluminum garnet.
The invention further provides a construction method of a BP neural network model for identifying the sub-types of the garnet on the basis of the thermal infrared spectrum characteristics, wherein the BP neural network model comprises an input layer, a hidden layer and an output layer; the construction method comprises the following steps:
dividing garnet samples with known garnet subclass types and thermal infrared spectrum characteristic data into training samples and verification samples, training the BP neural network model by using the training samples and verifying the precision of the BP neural network model by using the verification samples to obtain the BP neural network model for identifying the garnet subclasses based on the thermal infrared spectrum characteristic;
the training of the BP neural network model by using the training samples comprises the following steps:
inputting thermal infrared spectrum characteristic data of the training sample as neurons of an input layer and garnet subclass types of the training sample as neurons of an output layer into a BP neural network model;
initializing parameters of the BP neural network model;
forward propagation and calculation of output values of the output layer, and increase of nonlinearity of the BP neural network model by using an activation function;
if the end condition is not met, calculating the error between the output value of the output layer and the expected output value, and converting the error into error reverse feedback;
correcting the weight value by using an optimization method to perform error reverse feedback, and terminating the training when a termination condition is reached;
and processing the output value by adopting an output layer function.
In a specific embodiment of the present invention, when constructing the BP neural network model, the neurons of the infrared spectrum characteristic data or the input layer are wavelengths of reflection peaks of the garnet sample in a spectrum band of 9 to 13 μm, and include a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength, and a wavelength difference between the third reflection peak and the second reflection peak, where the third reflection peak wavelength is greater than the second reflection peak wavelength and is greater than the first reflection peak wavelength;
the neurons of the garnet subclass type or output layer are calcium-chromium-garnet, manganese-aluminum garnet, iron-aluminum garnet, magnesium-aluminum garnet, calcium-iron garnet, and calcium-aluminum garnet.
Further, the initializing the parameters of the BP neural network model includes setting the number of layers of the hidden layer, the number of neurons of the hidden layer, an activation function, an optimization method, a termination condition, and an output layer function.
Further, initializing parameters of the BP neural network model comprises setting an optimization method to be an L-BFGS algorithm.
Further, initializing parameters of the BP neural network model includes setting an activation function to a ReLU function.
Further, initializing parameters of the BP neural network model includes setting an output layer function to a softmax function.
Further, initializing parameters of the BP neural network model includes setting the neuron number of the hidden layer to 10. The neuron number of the hidden layer is calculated by formula
Determining, wherein m is the neuron number of the hidden layer, n is the neuron number of the input layer, a is the neuron number of the output layer, and b is an arbitrary constant of 0-10; as can be seen from the above, the neuron number of the input layer is 4, and the neuron number of the output layer is 6; tests show that the neuron number of the hidden layer is 10, and the effect is the best.
Further, initializing parameters of the BP neural network model includes setting the number of hidden layers to 1.
Further, initializing parameters of the BP neural network model includes setting a termination condition to a maximum number of iterations of 200.
In a specific embodiment of the present invention, the constructing the BP neural network model further includes performing normalization processing on the thermal infrared spectrum characteristic data.
The invention further provides a method for identifying the sub-types of the garnet, which comprises the following steps:
acquiring thermal infrared spectrum characteristic data of the garnet to be detected, wherein the characteristic data is the wavelength of a reflection peak of the garnet to be detected in a spectrum band of 9-13 μm, and comprises a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength and the wavelength difference between the third reflection peak and the second reflection peak, the third reflection peak wavelength is greater than the second reflection peak wavelength and greater than the first reflection peak wavelength,
and inputting the characteristic data into the BP neural network model for identifying the sub-type of the garnet to be detected based on the thermal infrared spectrum characteristics, and identifying the sub-type of the garnet to be detected.
The invention has the following beneficial effects:
according to the invention, the thermal infrared spectrum characteristic data (namely, the reflection peak wavelength in a spectrum band of 9-13 μm) of the garnet is taken as a research object, the nonlinear automatic mapping capability of a BP neural network model is utilized to find the complex mapping relation between the reflection peak wavelength of the thermal infrared spectrum band of the garnet and the subclass type, the judgment research of the subclass of the garnet is developed, a quick and effective technical support is provided for identifying the subclass of the garnet, and meanwhile, a good technical inspiration is provided for quickly and effectively identifying other minerals.
Drawings
The following describes embodiments of the present invention in further detail with reference to the accompanying drawings.
Fig. 1 shows a flow chart of a method for constructing a BP neural network model for identifying a garnet subclass based on thermal infrared spectral characteristics according to the present invention.
Fig. 2 shows a thermal infrared spectrum of garnet.
Fig. 3 shows a three-dimensional distribution diagram of the reflection peak wavelength of the 6-type garnet.
Fig. 4A is a wavelength distribution diagram of reflection peaks 1 and 2 of 6 types of garnet.
Fig. 4B is a wavelength distribution diagram of reflection peaks 1 and 3 of 6 types of garnet.
Fig. 4C is a wavelength distribution diagram of reflection peak 2 and reflection peak 3 of the 6-type garnet.
Detailed Description
In order to more clearly illustrate the invention, the invention is further described below with reference to preferred embodiments and the accompanying drawings. Similar parts in the figures are denoted by the same reference numerals. It is to be understood by persons skilled in the art that the following detailed description is illustrative and not restrictive, and is not to be taken as limiting the scope of the invention.
Embodiment 1 BP neural network model construction method for identifying garnet sub-type based on thermal infrared spectrum characteristics
The method for constructing the BP neural network model based on the thermal infrared spectrum characteristics and the garnet subclass type is shown in figure 1, and comprises the following specific steps:
firstly, acquiring thermal infrared spectrum characteristic data of a known sub-type garnet sample
Extracting thermal infrared spectrum characteristic data (table 1) of 85 garnet samples from a thermal infrared spectrum library, wherein the thermal infrared spectrum characteristic data comprise reflection peak wavelength data of 18 iron aluminum garnet, 15 calcium iron garnet, 25 calcium aluminum garnet, 18 magnesium aluminum garnet, 6 manganese aluminum garnet and 3 calcium chromium garnet in a spectrum band of 9-13 μm, namely the wavelength of a reflection peak 1, the wavelength of a reflection peak 2, the wavelength of a reflection peak 3 and reflection peak wavelength difference information, and the wavelength of the reflection peak 3 minus the wavelength of the reflection peak 2, wherein the wavelength of a third reflection peak is greater than the wavelength of a second reflection peak and is greater than the wavelength of a first reflection peak. The spectral feature data of garnet in table 1 were plotted by Python software into 6 garnet reflection peak wavelength distribution diagrams, as shown in fig. 3 and 4, from which: the reflection peak wavelengths of the calcium-chromium-garnet and the manganese-aluminum-garnet have obvious characteristics and are easy to distinguish from other types, the iron-aluminum-garnet, the magnesium-aluminum-garnet, the calcium-iron-garnet and the calcium-aluminum-garnet have larger overlapping at each reflection peak wavelength, and the types of punicalcites cannot be distinguished through simple threshold division.
TABLE 1 garnet thermal infrared spectral characteristic data
Secondly, constructing a BP neural network model for identifying the garnet subclasses based on the thermal infrared spectrum characteristics by utilizing the subclass types of the garnet samples and the thermal infrared spectrum characteristic data
A typical three-layer neural network is established through software python, the BP neural network model in the embodiment includes a three-layer neural network model of an input layer, an output layer and a hidden layer, the neurons of the input layer are determined by the thermal infrared spectrum characteristic data (namely, the wavelength of a reflection peak 1, the wavelength of a reflection peak 2, the wavelength of a reflection peak 3 and the wavelength difference of the reflection peak 3 and the reflection peak 2 in a spectrum band of 9-13 μm) and the neurons of the garnet subclass type (namely, calcium chromium garnet, manganese aluminum garnet, iron aluminum garnet, magnesium aluminum garnet, calcium iron garnet and calcium aluminum garnet) are determined as the neurons of the output layer, namely, the number of the neurons of the input layer is 4, and the number of the neurons of the output layer is 6.
In this embodiment, parameters when constructing the BP neural network model are set as: the number of layers of the hidden layer is 1, the number of neurons of the hidden layer is 10, the ReLU function is selected as the activation function, the L-BFGS algorithm is selected as the optimization method, the softmax function is selected as the output layer function, and the termination condition is set to be the maximum iteration number of 200 times. Wherein the neuron number of the hidden layer is represented by formula
Determining, wherein m is the neuron number of the hidden layer, n is the neuron number of the input layer, a is the neuron number of the output layer, and b is an arbitrary constant of 0-10; tests show that the neuron number of the hidden layer is 10, and the effect is the best.
Dividing 85 garnet samples into 68 training samples and 17 verification samples according to the ratio of 4: 1, training the BP neural network model by using the training samples, and verifying the precision of the BP neural network model by using the verification samples to finally obtain the BP neural network model:
one) training method of BP neural network model includes the following steps:
1. data information of the training sample is preferably normalized and input into the BP neural network model, that is, data after normalization processing is performed on thermal infrared spectrum characteristic data (i.e., wavelength of reflection peak 1, wavelength of reflection peak 2, wavelength of reflection peak 3, and difference between wavelengths of reflection peak 2 and reflection peak 3) in table 1 is input into the BP neural network model as neurons of an input layer, and garnet subclasses (i.e., calcium-chromium-garnet, manganese-aluminum-garnet, iron-aluminum-garnet, magnesium-aluminum-garnet, calcium-iron-garnet, and calcium-aluminum-garnet) are input into the BP neural network model as neurons of an output layer.
2. Initializing parameters of the BP neural network model, namely setting the number of layers of a hidden layer of the BP neural network model as 1, the number of neurons of the hidden layer as 10, an activation function as a ReLU function, an optimization method as an L-BFGS algorithm, an output layer function as a softmax function and a termination condition as the maximum iteration number of 200 times.
3. And (4) forward propagation and calculation of an output value of an output layer, and increasing the nonlinearity of the BP neural network model by adopting a ReLU function. During forward propagation, a BP neural network model is randomly given a connection weight, and layer-by-layer calculation is carried out from a hidden layer.
The introduction of the activation function may increase the non-linearity of the neural network model, such that the neural network may be applied to a multitude of non-linear models. The activation function selected in this embodiment is a ReLU function, which is of the form: f (x) max (0, x). The ReLU function is a piecewise linear function, with the function value being 0 when the argument x is negative and 0, and the function value being the x value itself when the argument is positive, an operation known as one-sided suppression. The property of unilateral inhibition makes the ReLU function a more biological activation model. sigmoid and tanh functions are saturated, their derivatives are both increasing and then decreasing, the closer to the target, the smaller the corresponding derivative, while the derivative of ReLu is always 1 for the part greater than 0. Therefore, the ReLU function is more consistent with biological characteristics than the sigmoid function and the tanh function, and has the advantages of overcoming gradient disappearance, accelerating training speed and relieving overfitting.
4. If the end condition is not reached, the error between the output value of the output layer and the expected output value is calculated and converted into the error reverse feedback.
5. And correcting the weight value by using an L-BFGS algorithm to perform error reverse feedback (back propagation), and terminating the training of the BP neural network model after the termination condition is met. The BP neural network model attempts to minimize the error signal at each training iteration. And during error reverse feedback, repeatedly correcting the weight from the output layer to the first hidden layer to continuously reduce the error between the output value and the expected output value until the iteration number reaches the maximum iteration number, and terminating training.
The BP neural network attempts to minimize the error at each training. When the error is minimized, a proper optimization method needs to be selected to correct the weight, and common optimization methods include: gradient descent method, newton method, conjugate gradient method, quasi-newton method. In the optimization method, an L-BFGS algorithm, which is one of quasi-Newton methods, is selected, namely, the weight is corrected by using the L-BFGS algorithm. The gradient descent method for solving the gradient matrix generally has the defects that the iteration direction is zigzag, the direct optimization cannot be carried out to a minimum value point, the more the iteration times are, the slower the convergence speed is, and the like; newton's method of solving the sea plug matrix results in a large amount of computation. The quasi-Newton method represented by the L-BFGS algorithm adopts a method of calculating a positive definite matrix similar to a sea plug matrix to optimize the cost function, and overcomes the defects of a gradient descent method and a Newton method. In addition, the quasi-newton method requires a larger storage space and a longer time per iteration than the conjugate gradient method, but requires a smaller number of iterations for convergence than the conjugate gradient method, and is therefore more suitable for use in a small network.
6. The output values are normalized by a softmax function to calculate probabilities that the sample is discriminated as each of the subclass types, the output values of the output layer are converted into a probability distribution in which the sum is 1 in the range of [0, 1], and the subclass discrimination is performed by the magnitude of the probability. And performing softmax function normalization processing on the output value is favorable for accurately judging the significance and the error of the output value, and finally obtaining the trained BP neural network model.
The output of the softmax function follows a multinomial distribution, suitable for the multi-classification case, while the output of the sigmoid function follows a bernoulli distribution, commonly used for the two-classification case.
And secondly) verifying the precision of the trained BP neural network model by using a verification sample to finally obtain the BP neural network model.
In the embodiment, 17 verification samples are used for evaluating the model prediction result, and the accuracy of the model is measured by using the accuracy, the recall ratio and the F1 value, and the result is shown in Table 2.
TABLE 2 BP neural network model prediction result evaluation
From the precision, recall and F1 values of the validation samples in table 2, it can be seen that: the evaluation results of the precision rate, the recall rate and the F1 value reach 100%, so the model created by using the BP neural network is effective and good in effect.
Acquiring thermal infrared spectrum characteristic data of a garnet sample to be detected, namely a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength and a wavelength difference between the third reflection peak and the second reflection peak of the garnet sample to be detected in a 9-13 mu m spectrum band, wherein the third reflection peak wavelength is greater than the second reflection peak wavelength and is greater than the first reflection peak wavelength;
inputting the characteristic data into the BP neural network model constructed as described in embodiment 1, the subclass type of the garnet sample to be tested can be identified.
As can be seen from the identification result obtained by verifying the sample in embodiment 1, the present invention obtains the thermal infrared spectrum characteristic data of the garnet sample to be detected, that is, obtains the first reflection peak wavelength, the second reflection peak wavelength and the third reflection peak wavelength of the thermal infrared spectrum of the garnet sample to be detected in the 9-13 μm spectrum band, the third reflection peak wavelength is greater than the second reflection peak wavelength and greater than the first reflection peak wavelength, calculates the wavelength difference between the third reflection peak and the second reflection peak, and inputs 4 pieces of thermal infrared spectrum characteristic data into the BP neural network model constructed in the present invention, so as to accurately and rapidly identify which garnet subclass the sample belongs to.
Comparative example 1 conventional K-means clustering recognized garnet subclasses
Similarly, 85 samples in table 1 were classified into garnet subclasses by using the thermal infrared spectral characteristic data of 85 samples shown in table 1, i.e., the data of reflection peak 1 wavelength, reflection peak 2 wavelength, reflection peak 3 wavelength, and reflection peak 3 wavelength minus reflection peak 2 wavelength in the spectrum band of 9-13 μm, and the results of the clustering classification are shown in table 3. The accuracy, the recall rate and the F1 value of the K-means clustering method for identifying the sub-types of the garnet are respectively 86.1%, 80% and 79.2%, except that the calcium-chromium garnet is completely and correctly classified, other types of garnet are wrongly classified.
TABLE 3K means Cluster analysis results and evaluation
Comparative example 2 identification of garnet subclasses by multivariate Linear discrimination method
Similarly, 85 samples in Table 1 were classified by the multivariate linear discriminant method using the thermal infrared spectral feature data of 85 samples shown in Table 1, i.e., the data of the wavelength of reflection peak 1, the wavelength of reflection peak 2, the wavelength of reflection peak 3, and the wavelength of reflection peak 3 minus the wavelength of reflection peak 2 in the 9-13 μm band, and the results are shown in Table 4. The accuracy, recall rate and F1 values of the classification of the multivariate linear discriminant method are respectively 84.2%, 80% and 79.5%, except that the calcium iron garnet, the manganese aluminum garnet and the calcium chromium garnet are completely and correctly classified, and other types of pomegranate stones are wrongly classified.
TABLE 4 multivariate Linear discriminant analysis results and evaluation
The k-means clustering analysis in the comparative example 1 is an unsupervised classification method for classifying from the angle of distance, and the multivariate linear discriminant calculation in the comparative example 2 is low in complexity, so that the mapping relation between the reflection peak wavelength of the garnet thermal infrared spectrum and the subclass type is difficult to determine by the two methods. The BP (Back propagation) neural network is a nonlinear supervised classification method, and the connection rule is obtained through self training, so that the problem that many traditional information analysis methods cannot solve can be solved, and therefore, a complex mapping relation is found by using a BP neural network model, and the classification of the garnet subclasses is accurately realized.
Comparative example 3 a method for identifying a sub-type of garnet
In the process of constructing the BP neural network model by using the subclass type of the garnet sample and the thermal infrared spectrum characteristic data in the second step of example 1, only the wavelength of the reflection peak 1, the wavelength of the reflection peak 2 and the wavelength of the reflection peak 3 are used as neurons of the input layer (that is, the number of neurons of the input layer is 3), and training for constructing the BP neural network model is performed in the same manner as in example 1 in other steps, so that the trained BP neural network model is obtained.
And verifying the precision of the trained BP neural network model by using the verification sample to finally obtain the BP neural network model.
The comparative example uses 17 verification samples for evaluation of model prediction results, and measures the model accuracy using the accuracy, recall and F1 values, and the results are shown in table 5.
TABLE 5 BP neural network model prediction result evaluation
From the precision, recall and F1 values in table 5, it can be seen that: the evaluation results of the precision rate, the recall rate and the F1 value can reach more than 80 percent and not reach 100 percent.
It can be seen that based on the BP neural network model constructed by the invention, the sub-type types of the garnet can be accurately and rapidly identified by obtaining the thermal infrared spectrum characteristic data of the garnet sample in the spectrum band of 9-13 μm without destructive chemical analysis means.
It should be understood that the above-mentioned embodiments of the present invention are only examples for clearly illustrating the present invention, and are not intended to limit the embodiments of the present invention, and it will be obvious to those skilled in the art that other variations or modifications may be made on the basis of the above description, and all embodiments may not be exhaustive, and all obvious variations or modifications may be included within the scope of the present invention.
Claims (10)
1. The identification method of the garnet subclasses based on the thermal infrared spectrum characteristics and the BP neural network model is characterized by comprising the following steps of:
acquiring thermal infrared spectrum characteristic data of a garnet sample of a known sub type;
constructing a BP neural network model for identifying the sub-type of the garnet based on the thermal infrared spectrum characteristics by utilizing the thermal infrared spectrum characteristic data and the sub-type of the garnet sample;
and acquiring thermal infrared spectrum characteristic data of the garnet sample to be detected, inputting the constructed BP neural network model, and identifying the subclass type of the garnet sample to be detected.
2. The identification method according to claim 1, wherein the infrared spectral feature data is the wavelength of the reflection peak of the garnet sample in the 9-13 μm spectrum band, including a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength and a wavelength difference of the third reflection peak and the second reflection peak, the third reflection peak wavelength being greater than the second reflection peak wavelength and greater than the first reflection peak wavelength; the sub-types of the garnet sample include calcium garnet, manganese garnet, iron garnet, magnesium garnet, calcium iron garnet, and calcium aluminum garnet.
3. The construction method of the BP neural network model for identifying the sub-types of the garnet on the basis of the thermal infrared spectrum features is characterized in that the BP neural network model comprises an input layer, a hidden layer and an output layer; the construction method comprises the following steps:
dividing garnet samples with known garnet subclass types and thermal infrared spectrum characteristic data into training samples and verification samples, training the BP neural network model by using the training samples and verifying the precision of the BP neural network model by using the verification samples to obtain the BP neural network model for identifying the garnet subclasses based on the thermal infrared spectrum characteristic;
the training of the BP neural network model by using the training samples comprises the following steps:
inputting thermal infrared spectrum characteristic data of the training sample as neurons of an input layer and garnet subclass types of the training sample as neurons of an output layer into a BP neural network model;
initializing parameters of the BP neural network model;
forward propagation and calculation of output values of the output layer, and increase of nonlinearity of the BP neural network model by using an activation function;
if the end condition is not met, calculating the error between the output value of the output layer and the expected output value, and converting the error into error reverse feedback;
correcting the weight value by using an optimization method to perform error reverse feedback, and terminating the training when a termination condition is reached;
and processing the output value by adopting an output layer function.
4. The constructing method according to claim 3, wherein the infrared spectrum characteristic data is the wavelength of the reflection peak of the garnet sample in the 9-13 μm spectrum band, including a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength and the wavelength difference of the third reflection peak and the second reflection peak, the third reflection peak wavelength is greater than the second reflection peak wavelength and is greater than the first reflection peak wavelength;
the sub-types of garnet are calcium garnet, manganese garnet, iron garnet, magnesium garnet, calcium iron garnet and calcium aluminum garnet.
5. The method according to claim 3, wherein the initializing parameters of the BP neural network model comprises setting the number of hidden layers, the number of neurons in the hidden layers, an activation function, an optimization method, a termination condition, and an output layer function.
6. The construction method according to claim 3, wherein initializing parameters of the BP neural network model comprises setting an optimization method to an L-BFGS algorithm.
7. The method of claim 3, wherein initializing parameters of the BP neural network model includes setting an activation function to a ReLU function.
8. The building method according to claim 3, wherein initializing parameters of the BP neural network model comprises setting an output layer function to a softmax function.
9. The construction method according to claim 3, wherein initializing parameters of the BP neural network model comprises setting the number of hidden layers to 1;
preferably, initializing parameters of the BP neural network model includes setting the number of neurons of the hidden layer to 10.
10. A method for identifying a garnet subclass, comprising the steps of:
acquiring thermal infrared spectrum characteristic data of the garnet to be detected, wherein the characteristic data is the wavelength of a reflection peak of the garnet to be detected in a spectrum band of 9-13 μm, and comprises a first reflection peak wavelength, a second reflection peak wavelength, a third reflection peak wavelength and the wavelength difference between the third reflection peak and the second reflection peak, the third reflection peak wavelength is greater than the second reflection peak wavelength and greater than the first reflection peak wavelength,
inputting the characteristic data into a BP neural network model for identifying the sub-type of the garnet to be detected based on the thermal infrared spectrum characteristics, identifying the sub-type of the garnet to be detected,
wherein the BP neural network model identifying a garnet subclass based on thermal infrared spectral characteristics is constructed according to the method of claim 3.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010785928.0A CN111950427B (en) | 2020-08-06 | 2020-08-06 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010785928.0A CN111950427B (en) | 2020-08-06 | 2020-08-06 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111950427A true CN111950427A (en) | 2020-11-17 |
| CN111950427B CN111950427B (en) | 2022-02-15 |
Family
ID=73331759
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010785928.0A Active CN111950427B (en) | 2020-08-06 | 2020-08-06 | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111950427B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113343576A (en) * | 2021-06-22 | 2021-09-03 | 燕山大学 | Prediction method of calcium yield in calcium treatment process based on deep neural network |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704883A (en) * | 2016-11-21 | 2018-02-16 | 东北大学 | A kind of sorting technique and system of the grade of magnesite ore |
| CN110161075A (en) * | 2019-05-05 | 2019-08-23 | 中国地质科学院矿产资源研究所 | Garnet analysis method utilizing thermal infrared spectrum and application thereof |
| CN111141698A (en) * | 2019-12-30 | 2020-05-12 | 中国地质大学（北京） | Lithology classification method based on thermal infrared emissivity |
| CN111460947A (en) * | 2020-03-15 | 2020-07-28 | 吉林大学 | Method and system for identifying metal minerals under microscope based on BP neural network |
-
2020
- 2020-08-06 CN CN202010785928.0A patent/CN111950427B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107704883A (en) * | 2016-11-21 | 2018-02-16 | 东北大学 | A kind of sorting technique and system of the grade of magnesite ore |
| CN110161075A (en) * | 2019-05-05 | 2019-08-23 | 中国地质科学院矿产资源研究所 | Garnet analysis method utilizing thermal infrared spectrum and application thereof |
| CN111141698A (en) * | 2019-12-30 | 2020-05-12 | 中国地质大学（北京） | Lithology classification method based on thermal infrared emissivity |
| CN111460947A (en) * | 2020-03-15 | 2020-07-28 | 吉林大学 | Method and system for identifying metal minerals under microscope based on BP neural network |
Non-Patent Citations (1)
| Title |
|---|
| 罗跃平: "石榴石的品种及鉴定", 《宝石和宝石学杂志》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113343576A (en) * | 2021-06-22 | 2021-09-03 | 燕山大学 | Prediction method of calcium yield in calcium treatment process based on deep neural network |
| CN113343576B (en) * | 2021-06-22 | 2022-03-11 | 燕山大学 | Prediction method of calcium yield in calcium treatment process based on deep neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111950427B (en) | 2022-02-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108509935B (en) | Radar working mode identification method based on random forest algorithm | |
| Guh | A hybrid learning-based model for on-line detection and analysis of control chart patterns | |
| CN108647272B (en) | Method for predicting concentration of butane at bottom of debutanizer by expanding small samples based on data distribution | |
| CN109240274B (en) | Industrial process fault diagnosis method based on high-order correlation | |
| CN107844751B (en) | Method for classifying hyperspectral remote sensing images of guide filtering long and short memory neural network | |
| US20040002928A1 (en) | Pattern recognition method for reducing classification errors | |
| JP3169715B2 (en) | Gas identification method and gas identification device | |
| KR100795227B1 (en) | Method and apparatus for analyzing signal pattern of sensor array | |
| CN112904810B (en) | Process industry nonlinear process monitoring method based on effective feature selection | |
| Yun | Prediction model of algal blooms using logistic regression and confusion matrix | |
| CN107203750B (en) | Hyperspectral target detection method based on combination of sparse expression and discriminant analysis | |
| CN111950427B (en) | Garnet subclass identification method based on thermal infrared spectrum characteristics and BP neural network model | |
| CN115640526A (en) | Drilling risk identification model, building method, identification method and computer equipment | |
| CN116204831A (en) | Road-to-ground analysis method based on neural network | |
| CN113095354B (en) | Unknown radar target identification method based on radiation source characteristic subspace knowledge | |
| CN112990125B (en) | Method for judging whether radiation source radar belongs to target platform | |
| CN110659767A (en) | Stock trend prediction method based on LSTM-CNN deep learning model | |
| CN103778342A (en) | Emitter recognition algorithm based on interval number and evidence theory | |
| CN103076595A (en) | Abnormal type identifying method for multivariate discrete radar emitter signal | |
| CN107067034B (en) | Method and system for rapidly identifying infrared spectrum data classification | |
| CN109766522B (en) | Non-probability reliability sensitivity analysis method for scraper conveyor chain wheel | |
| CN116930042B (en) | Building waterproof material performance detection equipment and method | |
| CN114580295B (en) | Pressure working condition identification method based on elastic BP random forest fusion | |
| Zhan et al. | Long Short-term Memory modeling method with monotonicity analysis as constraints base on Spearman coefficient | |
| CN117541626B (en) | Remote sensing image change detection method based on deep learning |
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 |