CN104504716A - Solid wood floor paint automatic color blending method based on BP neural network - Google Patents

Abstract

The invention discloses a solid wood floor paint automatic color blending method based on BP neural network; the method comprises the following steps; designing a floor paint formula, and configuring corresponding paints, painting a solid wood floor plain panel for obtaining a floor sample; collecting color characteristic values of the floor sample and the plain panel thereof, establishing a database of the floor sample color characteristic difference value and the paint formula; establishing a three-layer BP neural network model, inputting the floor color characteristic difference value (img file='DDA0000649137980000011.TIF' wi='204' he='82'/) and (img file='DDA00006491379800012.TIF' wi='1.4' he='90'/), the output is the floor paint formula, i.e., the proportion of the main paint, the red color concentrate, the black color concentrate and the yellow color concentrate; using the sample to train the network model to obtain the trained BP neural network; inputting the color characteristic difference value of the painted floor sample into the trained BP neural network, thereby obtaining the paint formula of the floor sample. Solid wood floor paint automatic color blending based on computer is achieved; the conventional manual color blending mode of floor pain is changed; and the efficiency and precision of color blending are increased.

Description

Based on the solid wooden floor board paint Automatic color matching method of BP neural network

Technical field

The present invention relates to solid wooden floor board paint-color matching field, particularly relate to a kind of solid wooden floor board based on BP neural network paint Automatic color matching method.

Background technology

At present, the annual production of domestic solid wooden floor board more than 8,000 ten thousand square metres, and also along with the increase of export volume, annual production is in rising trend.Because each one hobby is different with preference, the color on floor is not unalterable, is generally to be undertaken customizing by the requirement of client.Floor enterprise is after obtaining the paint floor sample of user, painting technology teacher relies on personal experience to add look essence in main paint, then the floor of this paint and sample are compared analysis, change the ratio of look essence further, to be met the paint formulation that floor color sample requires.Because the relatively main paint proportion of look essence is less, the change of its trace all can produce larger aberration, and therefore, the artificial paint formulation obtaining floor needs the longer time.In order to meet floor color individual demand further, realize the High-efficient Production of solid wooden floor board, information system management in the urgent need to realizing the robotization color matching of solid wooden floor board paint, and is carried out in the floor produced and formula thereof by solid wooden floor board enterprise.

Artificial neural network is a kind of network model utilizing the 26S Proteasome Structure and Function of computer simulation human brain, has stronger Nonlinear Processing ability.BP neural network is a kind of Multi-layered Feedforward Networks by Back Propagation Algorithm training, and having self-organization, self study and adaptive ability, is one of current most widely used neural network model.Therefore, the present invention considers to set up BP neural network model to realize the computing machine Automatic color matching of solid wooden floor board paint.

Summary of the invention

Technical matters to be solved by this invention is the deficiency for existing solid wooden floor board paint Man-made Color Matching, provides a kind of solid wooden floor board based on BP neural network model paint Automatic color matching method.

The present invention is for solving the problems of the technologies described above by the following technical solutions:

Based on the solid wooden floor board paint Automatic color matching method of BP neural network, comprise following steps:

Step 1), set up some floor paint formulas according to the formation of main paint and look essence in solid wooden floor board paint and proportional range thereof, and each floor paint formula corresponding produces the sample that each paints solid wooden floor board;

Step 2), calculate the color characteristic difference of each paint solid wooden floor board sample relatively laminin plate, set up the database of sample floor color characteristic difference and paint formulation;

Step 3), set up BP network model, make it be input as floor color characteristic difference, export as floor paint formula;

Step 4), the data in the database of sample floor color characteristic difference and paint formulation are inputted in BP network model, BP network model is trained, until the network error of BP network model is less than or equal to default error threshold;

Step 5), extract the color characteristic difference of the paint floor sample needing color matching;

Step 6), will the color characteristic difference input BP neural network of the paint floor sample of color matching be needed, draw the paint formulation of its correspondence.

As the solid wooden floor board paint Automatic color matching method further prioritization scheme that the present invention is based on BP neural network, described step 1) detailed step as follows:

Step 1.1), in main paint and red, black, yellow three kinds of look essences two kinds of look essence ratio-dependents condition under, by another look essence ratio in the scope of 0 to 4%, the mode increased progressively with 0.3% from 0 forms each paint formulation;

Step 1.2), for each paint formulation, Manufacture of Floor priming paint and finish paint respectively;

Step 1.3), the process sequence ground electroplate primer made according to each paint formulation and finish paint painted below according to first priming paint over the ground laminin plate paints;

Step 1.4), drying process is carried out to each paint floor, obtains the sample of each paint solid wooden floor board.

As the solid wooden floor board paint Automatic color matching method further prioritization scheme that the present invention is based on BP neural network, described step 2) detailed step as follows:

Step 2.1), under same light line strength, the sample of laminin plate and each paint solid wooden floor board is taken over the ground;

Step 2.2), the image of each shooting is selected the region comprehensively can reflecting floor color, and calculates the mean value of its color of image components R, G, B respectively, to obtain the color feature value of ground laminin plate and each paint solid wooden floor board sample;

Step 2.3), the color feature value of each paint solid wooden floor board sample and the color feature value of ground laminin plate are carried out mathematic interpolation respectively, obtains the color characteristic difference of each paint solid wooden floor board sample relatively laminin plate;

Step 2.4), according to each paint solid wooden floor board sample relatively the color characteristic difference of laminin plate and its correspondence paint formulation, set up the database of sample floor color characteristic difference and paint formulation.

As the further prioritization scheme of solid wooden floor board paint Automatic color matching method that the present invention is based on BP neural network, described step 3) in BP network model be three layers of BP network model, the neuron number of its input layer, hidden layer, output layer is respectively 3,4,4, the excitation function of hidden layer neuron is S type function, and the neuronic excitation function of output layer is linear function.

As the solid wooden floor board paint Automatic color matching method further prioritization scheme that the present invention is based on BP neural network, described step 4) detailed step as follows:

Step 4.1), the weight w of initialization input layer and hidden layer _ij, hidden layer and output layer weight w _jk, hidden layer threshold value a _j, output layer threshold value b _k, learning rate η and computational accuracy ε, wherein, i=1,2,3; J=1,2,3,4; K=1,2,3,4;

Step 4.2), according to input vector x _i, w _ijand a _j, calculate hidden layer neuron and export y _j:

$y_j=f_1({\mathrm{\Σ}}_{i=1}^3{x}_i{w}_{\mathrm{ij}}-a_j)$

In formula, f ₁for the excitation function of hidden layer neuron;

Step 4.3), output layer exports and calculates: export y according to hidden layer _j, w _jkand b _k, calculate BP neural network and export o _k:

$o_k=f_2({\mathrm{\Σ}}_{j=1}^4{y}_j{w}_{\mathrm{jk}}-b_k)$

In formula, f ₂for the neuronic excitation function of output layer;

Step 4.4), export o according to network _kwith desired output d _k, calculate network error e _k:

e _k＝d _k-o _k；

Step 4.5), calculate the error E of whole sample _p:

$E_P=\frac{1}{2}{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4{e}_{\mathrm{pk}}^2$

In formula, number of training is P, p=1,2,3 ... P, for the training error of single sample p;

Step 4.6), according to error E _padjustment network weight w _ijand w _jk:

$w_{\mathrm{ij}}=w_{\mathrm{ij}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}}){f_2}^{\′}\left(S_k\right)w_{\mathrm{jk}}{f_1}^{\′}\left(S_j\right)x_i$

$w_{\mathrm{jk}}=w_{\mathrm{jk}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}}){f_2}^{\′}\left(S_k\right)y_j$

In formula, d _pkfor the desired output of sample p, o _pkfor the network of sample p exports, S _jfor the neuronic input of jth in hidden layer, S _kfor the neuronic input of kth in output layer;

Step 4.7), judge whether network error is less than or equal to default error threshold, if network error is greater than default error threshold, then repeat step 4.2) to step 4.6), until network error is less than or equal to default error threshold.

The present invention adopts above technical scheme compared with prior art, has following technique effect:

1. achieve computer based solid wooden floor board paint Automatic color matching, change traditional paint floor Man-made Color Matching pattern, save a large amount of human resources;

2. improve color matching efficiency and precision.

Accompanying drawing explanation

Fig. 1 is the structure principle chart of color characteristic acquisition system in floor in the present invention;

Fig. 2 is three layers of BP neural network model figure in the present invention;

Fig. 3 is the process flow diagram utilizing BP neural network model to realize solid wooden floor board paint Automatic color matching in the present invention.

Embodiment

Below in conjunction with accompanying drawing, technical scheme of the present invention is described in further detail:

As shown in Figure 3, the invention discloses a kind of solid wooden floor board based on BP neural network paint Automatic color matching method, comprise following steps:

Step 1), floor Sample Establishing:

Solid wooden floor board paint mixes according to a certain percentage primarily of main paint and red, black, yellow three kinds of look essences, and often kind of look essence in floor paint is relative to main paint, and its proportion is between 0 ~ 4%.Under the condition of main paint and wherein two kinds of looks essence ratio-dependents, by another look essence ratio in the scope of 0 to 4%, the mode increased progressively with 0.3% from 0 forms paint formulation.Utilize same formula, respectively Manufacture of Floor priming paint and finish paint, the process sequence then painted below according to first priming paint over the ground laminin plate paints, and the thickness of paint film is 0.2mm.After drying process is carried out to paint floor, obtain paint solid wooden floor board sample.

Step 2), sample data collection:

As shown in Figure 1, floor color characteristic acquisition system is made up of industrial digital video camera, LED light source and computing machine, the DH-HV5051UC-M camera that digital camera wherein adopts Imax Corp. of Daheng to produce, the face battle array CMOS color image sensor of this camera to be resolution be 5,000,000 pixels, it carries out information interaction by USB interface and computing machine.Floor color characteristic acquisition system is utilized to be captured in the image on each sample floor under same light line strength, image is selected the region comprehensively can reflecting floor color, calculate the mean value of this area image color component R, G, B respectively, this mean value is the color feature value on sample floor, and computing formula is as follows:

$\stackrel{\‾}{R}=\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}R(i,j)/(m*n)$

$\stackrel{\‾}{G}=\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}G(i,j)/(m*n)$

$\stackrel{\‾}{B}=\underset{i=0}{\overset{m-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{n-1}{\mathrm{\Σ}}}B(i,j)/(m*n)$

In formula, R (i, j) is the red color component value of the i-th+1 row, j+1 row pixel in image; G (i, j) is the green component values of the i-th+1 row, j+1 row pixel in image; B (i, j) is the blue color component value of the i-th+1 row, j+1 row pixel in image, and m, n are respectively the row and column of pixel in image.

The color feature value of collecting sample ground laminin plate after the same method.The color feature value of plain with it for sample floor plate is carried out mathematic interpolation respectively, obtains the color characteristic difference on sample floor.Integrating step 1, sets up the database of sample floor color characteristic difference and paint formulation.

Step 3), BP network model is set up:

As shown in Figure 2, set up three layers of BP neural network model, floor color characteristic difference DELTA R, Δ G and Δ B is input as according to it, export as floor paint formula, the i.e. ratio of main paint, red look essence, black look essence and yellow look essence, the neuron number of input layer, hidden layer, output layer is respectively 3,4,4, and the excitation function of hidden layer neuron is S type function, and the neuronic excitation function of output layer is linear function.

Step 4), BP network model is trained:

Data in step 2 are input in three layers of BP neural network, network is trained, after network error meets the demands, obtain the BP neural network trained.Concrete training step is as follows:

Step 4.1), the weight w of initialization input layer and hidden layer _ij, hidden layer and output layer weight w _jk, hidden layer threshold value a _j, output layer threshold value b _k, learning rate η and computational accuracy ε, wherein, i=1,2,3; J=1,2,3,4; K=1,2,3,4;

Step 4.2), according to input vector x _i, w _ijand a _j, calculate hidden layer neuron and export y _j:

$y_j=f_1({\mathrm{\Σ}}_{i=1}^3{x}_i{w}_{\mathrm{ij}}-a_j)$

In formula, f ₁for the excitation function of hidden layer neuron;

Step 4.3), output layer exports and calculates: export y according to hidden layer _j, w _jkand b _k, calculate BP neural network and export o _k:

$o_k=f_2({\mathrm{\Σ}}_{j=1}^4{y}_j{w}_{\mathrm{jk}}-b_k)$

In formula, f ₂for the neuronic excitation function of output layer;

Step 4.4), export o according to network _kwith desired output d _k, calculate network error e _k:

e _k＝d _k-o _k；

Step 4.5), calculate the error E of whole sample _p:

$E_P=\frac{1}{2}{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4{e}_{\mathrm{pk}}^2$

In formula, number of training is P, p=1,2,3 ... P, for the training error of single sample p;

Step 4.6), according to error E _padjustment network weight w _ijand w _jk:

$w_{\mathrm{ij}}=w_{\mathrm{ij}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}}){f_2}^{\′}\left(S_k\right)w_{\mathrm{jk}}{f_1}^{\′}\left(S_j\right)x_i$

$w_{\mathrm{jk}}=w_{\mathrm{jk}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}}){f_2}^{\′}\left(S_k\right)y_j$

In formula, d _pkfor the desired output of sample p, o _pkfor the network of sample p exports, S _jfor the neuronic input of jth in hidden layer, S _kfor the neuronic input of kth in output layer;

Step 4.7), judge whether network error is less than or equal to default error threshold, if network error is greater than default error threshold, then repeat step 4.2) to step 4.6), until network error is less than or equal to default error threshold.

Step 5), extract needing the paint floor color sample feature difference of color matching: according to the color characteristic data acquisition method described in step 2, gather the color feature value of paint floor sample and plain plate thereof respectively.Color component in paint floor sample plate plain with it is carried out mathematic interpolation respectively, obtains the color characteristic difference of this paint floor sample with .

Step 6), floor sample paint formulation calculates: the color characteristic difference of paint floor sample is input to the BP neural network after training, can obtains the paint formulation of this floor sample.

Claims (5)

1., based on the solid wooden floor board paint Automatic color matching method of BP neural network, it is characterized in that, comprise following steps:

Step 1), set up some floor paint formulas according to the formation of main paint and look essence in solid wooden floor board paint and proportional range thereof, and each floor paint formula corresponding produces the sample that each paints solid wooden floor board;

Step 2), calculate the color characteristic difference of each paint solid wooden floor board sample relatively laminin plate, set up the database of sample floor color characteristic difference and paint formulation;

Step 3), set up BP network model, make it be input as floor color characteristic difference, export as floor paint formula;

Step 4), the data in the database of sample floor color characteristic difference and paint formulation are inputted in BP network model, BP network model is trained, until the network error of BP network model is less than or equal to default error threshold;

Step 5), extract the color characteristic difference of the paint floor sample needing color matching;

Step 6), will the color characteristic difference input BP neural network of the paint floor sample of color matching be needed, draw the paint formulation of its correspondence.

2. the solid wooden floor board based on BP neural network according to claim 1 paint Automatic color matching method, is characterized in that, described step 1) detailed step as follows:

Step 1.1), in main paint and red, black, yellow three kinds of look essences two kinds of look essence ratio-dependents condition under, by another look essence ratio in the scope of 0 to 4%, the mode increased progressively with 0.3% from 0 forms each paint formulation;

Step 1.2), for each paint formulation, Manufacture of Floor priming paint and finish paint respectively;

Step 1.3), the process sequence ground electroplate primer made according to each paint formulation and finish paint painted below according to first priming paint over the ground laminin plate paints;

Step 1.4), drying process is carried out to each paint floor, obtains the sample of each paint solid wooden floor board.

3. the solid wooden floor board based on BP neural network according to claim 1 paint Automatic color matching method, is characterized in that, described step 2) detailed step as follows:

Step 2.1), under same light line strength, the sample of laminin plate and each paint solid wooden floor board is taken over the ground;

Step 2.2), the image of each shooting is selected the region comprehensively can reflecting floor color, and calculates the mean value of its color of image components R, G, B respectively, to obtain the color feature value of ground laminin plate and each paint solid wooden floor board sample;

Step 2.3), the color feature value of each paint solid wooden floor board sample and the color feature value of ground laminin plate are carried out mathematic interpolation respectively, obtains the color characteristic difference of each paint solid wooden floor board sample relatively laminin plate;

Step 2.4), according to each paint solid wooden floor board sample relatively the color characteristic difference of laminin plate and its correspondence paint formulation, set up the database of sample floor color characteristic difference and paint formulation.

4. the paint of the solid wooden floor board based on BP neural network Automatic color matching method according to claim 1, it is characterized in that, described step 3) in BP network model be three layers of BP network model, the neuron number of its input layer, hidden layer, output layer is respectively 3,4,4, the excitation function of hidden layer neuron is S type function, and the neuronic excitation function of output layer is linear function.

5. the solid wooden floor board based on BP neural network according to claim 4 paint Automatic color matching method, is characterized in that, described step 4) detailed step as follows:

Step 4.1), the weight w of initialization input layer and hidden layer _ij, hidden layer and output layer weight w _jk, hidden layer threshold value a _j, output layer threshold value b _k, learning rate η and computational accuracy ε, wherein, i=1,2,3; J=1,2,3,4; K=1,2,3,4;

Step 4.2), according to input vector x _i, w _ijand a _j, calculate hidden layer neuron and export y _j:

$y_j=f_1({\mathrm{\Σ}}_{i=1}^3{x}_i{w}_{\mathrm{ij}}-a_j)$

In formula, f ₁for the excitation function of hidden layer neuron;

Step 4.3), output layer exports and calculates: export y according to hidden layer _j, w _jkand b _k, calculate BP neural network and export o _k:

$o_k=f_2({\mathrm{\Σ}}_{j=1}^4{y}_j{w}_{\mathrm{jk}}-b_k)$

In formula, f ₂for the neuronic excitation function of output layer;

Step 4.4), export o according to network _kwith desired output d _k, calculate network error e _k:

e _k＝d _k-o _k；

Step 4.5), calculate the error E of whole sample _p:

$E_P=\frac{1}{2}{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4{e}_{\mathrm{pk}}^2$

In formula, number of training is P, p=1,2,3 ... P, for the training error of single sample p;

Step 4.6), according to error E _padjustment network weight w _ijand w _jk:

$w_{\mathrm{ij}}=w_{\mathrm{ij}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}})f_2^{\′}\left(S_k\right)w_{\mathrm{jk}}{f}_1^{\′}\left(S_j\right)x_i$

$w_{\mathrm{jk}}=w_{\mathrm{jk}}+{\mathrm{\Σ}}_{p=1}^P{\mathrm{\Σ}}_{k=1}^4\mathrm{\η}(d_{\mathrm{pk}}-o_{\mathrm{pk}})f_2^{\′}\left(S_k\right)y_j$

In formula, d _pkfor the desired output of sample p, o _pkfor the network of sample p exports, S _jfor the neuronic input of jth in hidden layer, S _kfor the neuronic input of kth in output layer;

Step 4.7), judge whether network error is less than or equal to default error threshold, if network error is greater than default error threshold, then repeat step 4.2) to step 4.6), until network error is less than or equal to default error threshold.