CN106955112A - Brain wave Emotion recognition method based on Quantum wavelet neural networks model - Google Patents

Abstract

Brain wave Emotion recognition method disclosed by the invention based on Quantum wavelet neural networks model, including step：Brain wave data collection carries out regularization, the data after regularization is pre-processed, carrying out feature extraction to pretreated data, using classifying before three layers to type Quantum wavelet neural networks to data, wherein the hidden layer of Quantum wavelet neural networks adapts to basic function and wakes up degree analyzing using Mexican hat wavelet functions, progress evaluation of estimate.The present invention realizes faster convergence rate and Geng Gao cognitive precision.

Description

Brain wave Emotion recognition method based on Quantum wavelet neural networks model

Technical field

It is particularly a kind of to be managed based on quantum nerve network and small echo the present invention relates to a kind of Emotion recognition method of brain wave By the brain wave data Emotion recognition method being combined.

Background technology

In the last few years, the Digital Media of some new models was being entertained with interacting for human body, and the field such as study has huge Potentiality and market.But mood plays important role in the life of the mankind, so in the application of computer interface, base It is growing day by day in the demand of Emotion recognition.The mankind have six kinds of basic moods, i.e., happy, sad, frightened, surprised, angry, jealous It is jealous of.This six kinds of moods can be mutually combined, and derive remaining various compound mood, such as melancholy, nervous, anxiety. Emotion recognition is the key technology in the fields such as artificial intelligence, man-machine interaction.The arriving of information age, it is desirable to which machine can be more friendly Understanding and express the mankind mood, in actual life, Emotion recognition have been applied to medical treatment, education, business etc. field, but It is due to that mood is an extremely complex cognitive process, if Emotion recognition is wanted to obtain better effects, depends on what is deepened continuously Research.EEG signals are a kind of electricity physiological signals, with objectivity and accuracy, can directly reflect the activity of brain, thus quilt It is widely applied in Emotion recognition.

Brain wave (Electroencephalogram, EEG) has with mood closely to be contacted, and brain wave is brain in work When dynamic, what the postsynaptic potential that a large amount of neurons synchronously occur was formed after summation.It records electric wave change during brain activity, It is overall reflection of the bioelectrical activity in cerebral cortex or scalp surface of cranial nerve cell, its frequency variation scope is per second Between 1-30 times, four wave bands, i.e. δ (1-3Hz), θ (4-7Hz), α (8-13Hz), β (14-30Hz) can be divided into.Its In, alpha index (i.e. α ripples account for whole E.E.G percentages, it is quiet, when closing mesh alpha index for 75%) can as Emotion expression index, It is emotionally stable and the extensive people of thinking, alpha index is higher, emotional instability and narrow extreme people's alpha index are what for low.α ripples easily by Environmental stimuli is disturbed, and when opening eyes, α ripples can weaken or disappear, even in dark environment, and eye opening also can be such.β ripples are not Influenceed by opening, closing one's eyes.When opening eyes depending on thing, nervous, anxious, surprised and bewildered fear or taking the medicine such as stable, β ripples are lived It is dynamic drastically to increase.β activities are also relevant with some psychological traits of people.The people of beta response advantage often shows as：Nervous, mood Unstable, emotion is strong, the characteristics of be easy to get excited, stubbornly adhere one's opinion.In addition, when awakening and being absorbed in a certain thing, Chang Kejian A kind of frequency γ ripple higher compared with β ripples, its frequency is 30~80Hz, and wave amplitude scope is indefinite；And in sleep it may also occur that another A little waveforms more special normal brain wave, such as hump ripple, σ ripples, λ ripples, κ-complex wave, μ ripples.

The collection of current brain wave is main by the way of brain-computer interface, brain-computer interface (brain-computer Interface, BCI), sometimes referred to as " brain port " direct neural interface or " fusion of brain machine is perceived " Brain-machine interface, it is to be set up in human or animal's brain (or culture of brain cell) between external equipment Be directly connected to path.In the case of unidirectional brain-computer interface, computer either receives order or the sending signal that brain is transmitted To brain (such as video reconstruction), but it can not send and receive signal simultaneously.And two-way brain-computer interface allows between brain and external equipment Bi-directional exchanges of information.

Quantum nerve network (Quantum Neural Network, QNN) is that quantum calculation is mutually tied with neural network theory The new computation model produced is closed, is one of current emerging forward position cross discipline.Quantum nerve network is by the base of quantum calculation This concept and principle are incorporated into neural network model, quantization transformation are carried out to traditional neural network model, so as to strengthen god Performance and speed through network, are that the further development of neutral net proposes new thinking.Quantum nerve network is in EEG signal Emotion recognition in terms of, the deficiency of EEG signal Emotion recognition is carried out using traditional neural network before compensate for, with it is high surely Calmly, the features such as high accuracy and Fast Convergent.

But not only efficiency is low for the EEG Emotion recognitions that at present, quantum nerve network is calculated, and precision is not high.

For appeal problem, the present inventor devise speed faster, precision it is higher based on Quantum wavelet neural networks mould The brain wave Emotion recognition method of type.

The content of the invention

It is an object of the invention to provide the brain wave Emotion recognition method based on Quantum wavelet neural networks model, realize The cognitive precision of faster convergence rate and Geng Gao.

To achieve these goals, the present invention is adopted the following technical scheme that：

Brain wave Emotion recognition method based on Quantum wavelet neural networks model, comprises the following steps：

Step one, the brain wave data collection of collection is subjected to regularization, the value after regularization is between 0 to 1；

Data after regularization are pre-processed by step 2, specifically will be dry using independent PCA (ICA) Data are disturbed to remove from brain wave data；

Pretreated data are carried out feature extraction by step 3, specifically specific by selecting using clustering technique (CT) Feature reduce the size of data set；

Step 4, using classifying before three layers to type Quantum wavelet neural networks (QWNN) to data, wherein quantum is small The hidden layer of ripple neutral net adapts to basic function and uses Mexican-hat wavelet functions, and the specific expression formula of the function is：

Wherein t is time variable；

Step 5, carries out evaluation of estimate and wakes up degree analyzing.

Interference data are removed using the independent PCA and specifically include following steps：

(1) data by regularization in step one are designated as K₂(t),K₃(t)...K_n(t) and with desired signal K₁(t) together It is used as input signal K (t)=[K₁(t)；K₂(t)；K₃(t)；...K_n(t) independent principal component analysis] is carried out, separation signal S is obtained (t)=[S₁(t)；S₂(t)；S₃(t)；...S_n(t)]；

(2) separation signal being subjected to spectrum analysis respectively, according to ideal signal S'(t) feature determines echo signal S₁(t)；

(3) using spectrum correction method to the separation signal S after independent principal component analysis₁(t) spectrum recovery is carried out, finally To Y (t) signals for removing interference signal.

The step of carrying out feature extraction using the clustering technique (CT) includes：

1) each brain wave data is divided into n group, each group is referred to as special time interval cluster；

2) each cluster is divided into the submanifold of m special periods；

3) eight statistical natures, the statistical nature difference of these brain wave datas are extracted from the data point of each submanifold It is：Minimum value, maximum, average, intermediate value, the first quartile scope, the 3rd quartile scope, quartile range and standard deviation.

Comprise the following steps before described three layers to the training of type Quantum wavelet neural networks：

1. all training input vectors, are input to Quantum wavelet neural networks, Quantum wavelet neural networks use three Layer Feed-forward neural networks, input layer includes n_iIndividual node, hidden layer includes n_hIndividual many node layers, output layer includes n_oIndividual node.w_jl Represent l-th of input node to the weights of j-th of hidden node, v_ijRepresent that j-th of hidden node exports node layer to i-th Weights.Note

Here m represents the number of data set X Mean Vector；Assuming that a multilayer hidden node has n_sIndividual discrete quantum Layer, then excitation function can be write as by n_sThe superposition of individual sub- excitation function, and by θ^sChanged, the specific table of its hypothesized model It is as follows up to formula：

Wherein,The excitation function of hidden node is represented, β represents Slope Parameters, θ^sBetween expression quantum Every λ_jRepresent contraction-expansion factor, τ_jRepresent transfer factor；

The output vector of output layer is：

Wherein

The general expression of QWNN models is：

Wherein k represents the quantity of sample；

2., the renewal of QWNN weights：

Cost function isWherein k=1,2 ..., m, m be training sample sum,Parameter w_jl, v_ij, λ_j, τ_jCarried out by minimizing cost function Update：

Wherein

For any h iteration, parameter w_jl, v_ij, λ_j, τ_jAll it is adjusted according to following formula：

Wherein α is learning rate, 0<α<1.

3., the renewal at quantum interval：

c_iThe output variance of class is：

Wherein,Expression belongs to class c_iHidden node output summation.

Quantum intervalIt is adjusted by minimizing cost function G：

Calculate G gradient：

Wherein α_θForLearning rate, 0<α_θ<1；

Wherein,

Order

Finally calculate its renewal equation：

It is updated according to following formula：

The step 5 carries out evaluation of estimate and wakes up degree analyzing, is specially：

By changing the control parameter of facial muscles, six Haptek mood factors are defined, these mood factors are respectively： Happy, sad, frightened, surprised, angry, envy, these mood factors are classified by algorithm described in step 4；

Do evaluation of estimate and wake up the mapping of degree, wherein the degree of wake-up can only be chosen from 0, the 1,2 of mark, and evaluation of estimate It can only be chosen from the 0 of mark and 1.

After such scheme, present invention has the advantages that：Wavelet theory is applied into quantum nerve network algorithm, to amount Sub-neural network algorithm is improved, and has obtained Quantum wavelet neural networks algorithm, and apply to the Emotion recognition of brain wave In, in terms of EEG Emotion recognitions are handled, Quantum wavelet neural networks algorithm is calculated with traditional neutral net and quantum nerve network Method Comparatively speaking, the precision with faster convergence rate and Geng Gao.

The present invention is described further below in conjunction with the accompanying drawings.

Brief description of the drawings

Fig. 1 is brain wave data process chart of the present invention.

Fig. 2 is Quantum wavelet neural networks structure chart of the present invention.

Fig. 3 is that six big basic emotions are waking up the mark of degree-evaluation of estimate coordinate.

Embodiment

The brain wave Emotion recognition method based on Quantum wavelet neural networks model that the present embodiment as shown in Figure 1 is disclosed, Specifically include following steps：

Step one, the brain wave data collection of collection is subjected to regularization, the value after regularization is between 0 to 1；

Data after regularization are pre-processed by step 2, specifically will be dry using independent PCA (ICA) Disturb data to remove from brain wave data, interference data are mainly noise data；

Interference data are removed using the independent PCA and specifically include following steps：

(1) data by regularization in step one are designated as K₂(t),K₃(t)...K_n(t) and with desired signal K₁(t) together It is used as input signal K (t)=[K₁(t)；K₂(t)；K₃(t)；...K_n(t) independent principal component analysis] is carried out, separation signal S is obtained (t)=[S₁(t)；S₂(t)；S₃(t)；...S_n(t)]；

(2) separation signal being subjected to spectrum analysis respectively, according to ideal signal S'(t) feature determines echo signal S₁(t)；

(3) using spectrum correction method to the separation signal S after independent principal component analysis₁(t) spectrum recovery is carried out, finally To Y (t) signals for removing interference signal.

Pretreated data are carried out feature extraction by step 3, specifically specific by selecting using clustering technique (CT) Feature reduce the size of data set；

The step of carrying out feature extraction using the clustering technique includes：

1) each brain wave data is divided into n group, each group is referred to as special time interval cluster；

2) each cluster is divided into the submanifold of m special periods；

3) eight statistical natures, the statistical nature difference of these brain wave datas are extracted from the data point of each submanifold It is：Minimum value, maximum, average, intermediate value, the first quartile scope, the 3rd quartile scope, quartile range and standard deviation.

Step 4, using classifying before three layers to type Quantum wavelet neural networks (QWNN) to data, wherein quantum is small The hidden layer of ripple neutral net adapts to basic function and uses Mexican-hat wavelet functions, and the specific expression formula of the function is：

Wherein t is time variable；

Step 5, carries out evaluation of estimate and wakes up degree analyzing.

Step 5 carries out evaluation of estimate and wakes up degree analyzing, is specially：

By changing the control parameter of facial muscles, six Haptek mood factors are defined, these mood factors are respectively： Happy, sad, frightened, surprised, angry, envy, these mood factors are classified by algorithm described in step 4；

Do evaluation of estimate and wake up the mapping of degree, wherein the degree of wake-up can only be chosen from 0, the 1,2 of mark, and evaluation of estimate It can only be chosen from the 0 of mark and 1, mapping structure is referring to Fig. 3.

In step 4, comprise the following steps before three layers to the training of type Quantum wavelet neural networks：

1. all training input vectors, are input to Quantum wavelet neural networks, as shown in Fig. 2 quantum wavelet neural Network includes n using three layers of Feed-forward neural networks, input layer_iIndividual node, hidden layer includes n_hIndividual many node layers, output layer is included n_oIndividual node.w_jlRepresent l-th of input node to the weights of j-th of hidden node, v_ijRepresent j-th of hidden node to i-th Export the weights of node layer.Note

Here m represents the number of data set X Mean Vector；Assuming that a multilayer hidden node has n_sIndividual discrete quantum Layer, then excitation function can be write as by n_sThe superposition of individual sub- excitation function, and by θ^sChanged, the specific table of its hypothesized model It is as follows up to formula：

Wherein,The excitation function of hidden node is represented, β represents Slope Parameters, θ^sBetween expression quantum Every λ_jRepresent contraction-expansion factor, τ_jRepresent transfer factor；

The output vector of output layer is：

Wherein

The general expression of QWNN models is：

Wherein k represents the quantity of sample；

2., the renewal of QWNN weights：

Cost function isWherein k=1,2 ..., m, m be training sample sum,Parameter w_jl, v_ij, λ_j, τ_jCarried out by minimizing cost function Update：

Wherein

For any h iteration, parameter w_jl, v_ij, λ_j, τ_jAll it is adjusted according to following formula：

Wherein α is learning rate, 0<α<1.

3., the renewal at quantum interval：

c_iThe output variance of class is：

Wherein,Expression belongs to class c_iHidden node output summation.

Quantum intervalIt is adjusted by minimizing cost function G：

Calculate G gradient：

Wherein α_θForLearning rate, 0<α_θ<1；

Wherein,

Order,

Finally calculate its renewal equation：

It is updated according to following formula：

The preferred embodiments of the present invention have shown and described in described above, it should be understood that the present invention is not limited to this paper institutes The form of disclosure, is not to be taken as the exclusion to other embodiment, and can be used for various other combinations, modification and environment, and energy Enough in invention contemplated scope herein, it is modified by the technology or knowledge of above-mentioned teaching or association area.And people from this area The change that is carried out of member and change do not depart from the spirit and scope of the present invention, then all should appended claims of the present invention protection In the range of.

Claims (5)

1. the brain wave Emotion recognition method based on Quantum wavelet neural networks model, it is characterised in that comprise the following steps：

Step one, the brain wave data collection of collection is subjected to regularization, the value after regularization is between 0 to 1；

Data after regularization are pre-processed by step 2, specifically using independent PCA will disturb data from Removed in brain wave data；

Pretreated data are carried out feature extraction by step 3, it is specific using clustering technique by select specific feature come Reduce the size of data set；

Step 4, using classifying before three layers to type Quantum wavelet neural networks to data, wherein Quantum wavelet neural networks Hidden layer adapt to basic function and use Mexican-hat wavelet functions, the specific expression formula of the function is：

Wherein t is time variable；

Step 5, carries out evaluation of estimate and wakes up degree analyzing.

2. the brain wave Emotion recognition method as claimed in claim 1 based on Quantum wavelet neural networks model, its feature exists In removing interference data using the independent PCA and specifically include following steps：

(1) data by regularization in step one are designated as K₂(t),K₃(t)...K_n(t) and with desired signal K₁(t) together as defeated Enter signal K (t)=[K₁(t)；K₂(t)；K₃(t)；...K_n(t)] carry out independent principal component analysis, obtain separation signal S (t)= [S₁(t)；S₂(t)；S₃(t)；...S_n(t)]；

(2) separation signal being subjected to spectrum analysis respectively, according to ideal signal S'(t) feature determines echo signal S₁(t)；

(3) using spectrum correction method to the separation signal S after independent principal component analysis₁(t) spectrum recovery is carried out, is finally obtained Except Y (t) signals of interference signal.

3. the brain wave Emotion recognition method as claimed in claim 1 based on Quantum wavelet neural networks model, its feature exists In the step of carrying out feature extraction using the clustering technique includes：

1) each brain wave data is divided into n group, each group is referred to as special time interval cluster；

2) each cluster is divided into the submanifold of m special periods；

3) eight statistical natures are extracted from the data point of each submanifold, the statistical nature of these brain wave datas is respectively：Most Small value, maximum, average, intermediate value, the first quartile scope, the 3rd quartile scope, quartile range and standard deviation.

4. the brain wave Emotion recognition method as claimed in claim 1 based on Quantum wavelet neural networks model, its feature exists In comprising the following steps before described three layers to the training of type Quantum wavelet neural networks：

1. all training input vectors, are input to Quantum wavelet neural networks, Quantum wavelet neural networks are used before three layers To type neutral net, input layer includes n_iIndividual node, hidden layer includes n_hIndividual many node layers, output layer includes n_oIndividual node.w_jlRepresent L-th of input node is to the weights of j-th of hidden node, v_ijRepresent j-th of hidden node to the power of i-th of output node layer Value, note

$\begin{array}{cc}{x}^k={\&lsqb;x_1^k,x_2^k,x_3^k,\mathrm{...},x_{ni}^k\&rsqb;}^T& \&ForAll;k=1,\mathrm{...},m\end{array}$

Here m represents the number of data set X Mean Vector；Assuming that a multilayer hidden node has n_sIndividual discrete quantum layer, then Excitation function can be write as by n_sThe superposition of individual sub- excitation function, and by θ^sChanged, the expression of its hypothesized model is such as Under：

$h\left(x\right)=\frac{1}{n^s}{\mathrm{\&Sigma;}}_{s=1}^{n_s}{h}_0\left(\mathrm{\&beta;}\right(x-{\mathrm{\&theta;}}^s\left)\right)$

Wherein,The excitation function of hidden node is represented, β represents Slope Parameters, θ^sRepresent quantum interval, λ_jTable Show contraction-expansion factor, τ_jRepresent transfer factor；

The output vector of output layer is：

$\begin{array}{cc}{C}_i^k={\mathrm{\&Sigma;}}_{j=1}^{n_h}{v}_{ji}{B}_j,& i=1,2,...,n_0\end{array}$

Wherein

$\begin{array}{ccc}{B}_j=\frac{1}{n_s}{\mathrm{\&Sigma;}}_{s=1}^{n_s}h\&lsqb;\mathrm{\&beta;}({\hat{h}}_j^k-{\mathrm{\&theta;}}_j^s)\&rsqb;& s=1,2,...,n_s& j=1,2,...,n_h\end{array}$

${\hat{h}}_j^k={\mathrm{\&Sigma;}}_{l=1}^{n_i}{w}_{lj}{x}_l^k$

The general expression of QWNN models is：

$C_i^k={\mathrm{\&Sigma;}}_{j=1}^{n_h}{v}_{ji}\left(\frac{1}{n_s}{\mathrm{\&Sigma;}}_{s=1}^{n_s}h\right(\frac{\left(\mathrm{\&beta;}\right({\mathrm{\&Sigma;}}_{l=1}^{n_j}{w}_{lj}{x}_l)-{\mathrm{\&theta;}}_j^s)-{\mathrm{\&tau;}}_j}{{\mathrm{\&lambda;}}_j}\left)\right)$

Wherein k represents the quantity of sample；

2., the renewal of QWNN weights：

Cost function isWherein k=1,2 ..., m, m be training sample sum,Parameter w_jl, v_ij, λ_j, τ_jCarried out by minimizing cost function Update：

$\frac{\&part;E^k}{\&part;V_{ji}}=e_i*B_j$

$\frac{\&part;E^k}{\&part;V_{ji}}=e_i{v}_{ji}\frac{1}{n_s}{\mathrm{\&Sigma;}}_{s=1}^{n_s}\frac{\&part;h}{\&part;{x^{\&prime;}}_l}{\mathrm{\&beta;x}}_l$

${x^{\&prime;}}_j=\mathrm{\&beta;}({\mathrm{\&Sigma;}}_{l=1}^{n_i}{w}_{lj}{x}_l-{\mathrm{\&theta;}}_j^s)$

Wherein

${t^{\&prime;}}_j=\frac{{x^{\&prime;}}_j-{\mathrm{\&tau;}}_j}{{\mathrm{\&lambda;}}_j},e_i=C_i^k-Y_i^k$

$\frac{\&part;h}{\&part;w_{ir}}=1.373*4{\mathrm{\&pi;te}}^{-{\mathrm{\&pi;t}}^2}-2\mathrm{\&pi;}t*1.373(1+2{\mathrm{\&pi;t}}^2)e^{-{\mathrm{\&pi;t}}^2}$

For any h iteration, parameter w_jl, v_ij, λ_j, τ_jAll it is adjusted according to following formula：

$v_{ji}(h+1)=v_{ji}\left(h\right)-\mathrm{\&alpha;}\frac{\&part;E^k}{\&part;V_{ji}}$

$w_{li}(h+1)=w_{li}\left(h\right)-\mathrm{\&alpha;}\frac{\&part;E^k}{\&part;w_{li}}$

Wherein α is learning rate, 0<α<1.

3., the renewal at quantum interval：

c_iThe output variance of class is：

Wherein,Expression belongs to class c_iHidden node output summation.

Quantum intervalIt is adjusted by minimizing cost function G：

$G=\frac{1}{2}{\mathrm{\&Sigma;}}_{j=1}^{n_h}{\mathrm{\&Sigma;}}_{i=1}^{n_0}{\mathrm{\&sigma;}}_{j,i}^2=\frac{1}{2}{\mathrm{\&Sigma;}}_{j=1}^{n_h}{\mathrm{\&Sigma;}}_{i=1}^{n_0}{\mathrm{\&Sigma;}}_{x_k\&Element;c_i}{(<B_{j,ci}>-B_{j,k})}^2$

Calculate G gradient：

${\mathrm{\&Delta;\&theta;}}_j^s=-{\mathrm{\&alpha;}}_{\mathrm{\&theta;}}\frac{\&part;G}{\&part;{\mathrm{\&theta;}}_j^s}={\&Sigma;}_{i=1}^{n_0}{\&Sigma;}_{x_k\&Element;c_i}(<B_{j,ci}>-B_{j,k})\&lsqb;\frac{\&part;<B_{j,ci}>}{\&part;{\mathrm{\&theta;}}_j^s}-\frac{B_{j,k}}{\&part;{\mathrm{\&theta;}}_j^s}\&rsqb;$

Wherein α_θForLearning rate, 0<α_θ<1；

$\frac{\&part;<B_{j,ci}>}{\&part;{\mathrm{\&theta;}}_j^s}=\frac{1}{\left|c_i\right|}{\mathrm{\&Sigma;}}_{x_k\&Element;c_i}\frac{\&part;B_{j,k}}{\&part;{\mathrm{\&theta;}}_j^s}$

$\frac{\&part;B_{j,k}}{\&part;{\mathrm{\&theta;}}_j^s}-\frac{1}{n_s}\frac{\&part;h}{{\mathrm{\&theta;}}_j^s}=-\frac{1}{n_s}\frac{\&part;h}{\&part;{X^{\&prime;}}_i}\mathrm{\&beta;}$

Wherein,

Order

Finally calculate its renewal equation：

${\mathrm{\&Delta;\&theta;}}_j^s=-a_{\mathrm{\&theta;}}\frac{\mathrm{\&beta;}}{n_s}{\mathrm{\&Sigma;}}_{i=1}^{n_0}{\mathrm{\&Sigma;}}_{x_k\&Element;c_i}(<B_{j,ci}>-B_{j,k})*(<{\mathrm{DB}}_{j,k}>-{\mathrm{DB}}_{j,k})$

It is updated according to following formula：

${\left({\mathrm{\&theta;}}_j^s\right)}^{h+1}={\left({\mathrm{\&theta;}}_j^s\right)}^h-{\mathrm{\&alpha;}}_{\mathrm{\&theta;}}\frac{\&part;G}{\&part;{\mathrm{\&theta;}}_j^s}.$

5. the brain wave Emotion recognition method as claimed in claim 1 based on Quantum wavelet neural networks model, its feature exists In the step 5 carries out evaluation of estimate and wakes up degree analyzing, is specially：

By changing the control parameter of facial muscles, six Haptek mood factors are defined, these mood factors are respectively：It hurry up Happy, sad, frightened, surprised, angry, envy, these mood factors are classified by algorithm described in step 4；

Do evaluation of estimate and wake up the mapping of degree, wherein the degree of wake-up can only be chosen from 0, the 1,2 of mark, and evaluation of estimate can only Chosen from the 0 of mark and 1.