A flexographic printing pressure prediction method based on deep belief network

Abstract

The invention discloses a flexographic printing pressure prediction method based on a deep belief network. The method comprises the following steps: performing a binarization operation on a color separation manuscript image to obtain a binarized image, performing an inversion operation on the binarized image, dividing the inverted binarized image into grids, calculating the area of ​​images and texts in the grid area, and obtaining an area matrix; performing a gradient calculation on the color separation manuscript image to obtain a gradient image, dividing the gradient image into grids, calculating the maximum gradient value in each grid, and obtaining a maximum gradient value matrix; performing a grayscale co-occurrence matrix calculation on the color separation manuscript image to obtain a grayscale co-occurrence matrix; then juxtaposing the area matrix, the maximum gradient value matrix, and the grayscale co-occurrence matrix horizontally to obtain a feature matrix of the color separation manuscript image; finally, establishing a network prediction model with the feature matrix as input data and the actual printing pressure value as output data of the network prediction model, and building a deep belief network pressure prediction model, thereby realizing effective prediction of flexographic printing pressure.

Description

Technical Field

[0001] The invention belongs to the technical field of flexographic printing, and in particular relates to a flexographic printing pressure prediction method based on a deep belief network. Background Art

[0002] The fourth industrial revolution is coming, and the new generation of information technology will have a profound impact on the traditional manufacturing industry. Intelligent production will drive changes in the production model, that is, from the previous single, large-scale, and scale-effective traditional product manufacturing to a variety of products that meet the personalized needs of consumers, thereby creating a flexible and intelligent production model enabled by technology, integrating technologies such as artificial intelligence and machine learning with traditional production, so that enterprises can greatly reduce labor input and significantly improve production efficiency.

[0003] Using artificial intelligence and machine learning methods to predict the optimal printing pressure value for flexographic printing can replace the traditional flexographic printing pressure prediction method, save manpower and material resources, and greatly improve the company's production efficiency. Summary of the invention

[0004] The purpose of the present invention is to provide a flexographic printing pressure prediction method based on a deep belief network, which can predict the optimal printing pressure of flexographic printing by a deep learning method without the aid of a dedicated flexographic printing plate pressure measurement device.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A flexographic printing pressure prediction method based on a deep belief network is implemented in the following steps:

[0007] Step 1: Binarize the color separation original image required for flexographic printing to obtain a binary image;

[0008] Step 2: The binary image obtained in step 1 is inverted to obtain an inverted binary image;

[0009] Step 3: Divide the inverted binary image obtained in step 2 into 28×28 grids, calculate the area of ​​the image and text in each grid area, and obtain the area matrix;

[0010] Step 4: Perform gradient calculation on the color separation original image required for flexographic printing to obtain a gradient image;

[0011] Step 5: Divide the gradient image obtained in step 4 into 28×28 grids, calculate the maximum gradient value in each grid, and obtain the maximum gradient value matrix;

[0012] Step 6: Calculate the grayscale co-occurrence matrix of the color separation original image required for flexographic printing, and use the graycomatrix function provided by MATLAB to calculate and obtain the grayscale co-occurrence matrix;

[0013] Step 7: Arrange the three groups of vectors, namely, the area matrix, the maximum gradient value matrix, and the gray-level co-occurrence matrix, obtained in steps 3, 5, and 6, in parallel horizontally to obtain a characteristic matrix of the color separation original image;

[0014] Step 8: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation original image on the flexographic printing machine. At the same time, use the feature matrix of the color separation original image obtained in step 7 as the input data of the deep confidence network model; use the actual optimal printing pressure value as the output data of the deep confidence network model, establish a deep confidence network pressure prediction model, optimize the model parameters, and obtain the optimal deep confidence network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep confidence network pressure prediction model.

[0015] Furthermore, the step 3 is specifically implemented according to the following method:

[0016] Step 3.1: Divide the inverted binary image obtained in step 2 into a regular grid of 28*28 for calculating the area of ​​the image and text in the grid area. The number of grid divisions has an important influence on the experimental results. After optimization, the grid is finally divided into 28×28;

[0017] Step 3.2: After processing in step 3.1, mark the number of connected regions in the grid, then count the number of pixels in each connected region, and finally add the number of pixels in each connected region to obtain the area of ​​the image and text in the grid;

[0018] Step 3.3: Calculate the area of ​​the image and text in each grid, and finally obtain a 28*28 area matrix.

[0019] Furthermore, the step 4 is specifically implemented according to the following method:

[0020] Step 4.1: Use the commonly used Robert operator horizontal gradient template to perform convolution calculation on the color separation original image to obtain a horizontal gradient image;

[0021] Step 4.2: Use the commonly used Robert operator longitudinal gradient template to perform convolution calculation on the color separation original image to obtain a longitudinal gradient image;

[0022] Step 4.3: After steps 4.1 and 4.2, calculate the modulus of the gradient to obtain a gradient image. The modulus of the gradient is usually referred to as the gradient. Therefore, the calculation of the gradient is:

[0023]

[0024] In the formula, Gx is the transverse gradient, Gy is the longitudinal gradient, and G is the modulus of the gradient, that is, the gradient; after calculating the modulus of the gradient, the gradient image can be obtained.

[0025] Furthermore, the step 8 is specifically implemented according to the following method:

[0026] Step 8.1: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation manuscript image on the flexographic printing machine, and use the feature matrix of the color separation manuscript image obtained in step 7 as the input data of the deep belief network pressure prediction model, and the actual printing pressure value as the output data of the deep belief network pressure prediction model, and divide the data set consisting of the input data and the output data into a training set and a test set;

[0027] Step 8.2: Input the data in the training set into the deep belief network pressure prediction model structure, train the deep belief network, and obtain the optimal deep belief network pressure prediction model after normalization, network model evaluation and parameter optimization;

[0028] The specific process is:

[0029] (1) Normalize the feature matrix. Use the linear normalization method to normalize the feature matrix to the interval [0,1]. The expression is as follows:

[0030]

[0031] In the formula, X and X' represent the original data and the normalized data, respectively, and Xmin and Xmax represent the minimum and maximum values ​​of the original data, respectively;

[0032] (2) The evaluation parameters of the deep belief network model used are: mean absolute percentage error MAPE, linear regression correlation coefficient R and cumulative prediction accuracy function CS(θ):

[0033] Among them, MAPE is the expected value of the square of the difference between the predicted value and the actual value, which reflects the overall deviation of the model's predicted value relative to the actual value. The smaller the value of MAPE, the better the accuracy of the prediction model.

[0034] The regression coefficient R reflects the correlation between the predicted value and the actual value. The closer the R value is to 1, the better the correlation between the experimental data described by the prediction model and the actual value.

[0035] CS(θ) is the percentage of samples whose absolute error between the predicted value and the actual value is less than θ in the total number of samples. Its function expression is:

[0036]

[0037] Where Ne ≤θ is the number of data with prediction error less than θ, and N is the number of prediction samples;

[0038] (3) After parameter optimization, the structure of the deep belief network is finally determined. The network adopts a four-layer structure, the neuron structure is 2352-1568-784-500, the network learning rate is 0.005, the network momentum is 0.8, the number of iterations is 100, and the number of random samples in each iteration is 5. The deep belief network assigns the learned weights to the artificial network by unfolding, and the pressure prediction is realized through the artificial neural network. The activation function of the artificial neural network is the sigm function, the activation function of the network output layer is the linear function, the network learning rate is 0.005, the network momentum is 0.9, the number of iterations is 100, and the number of random samples in each iteration is 5;

[0039] Step 8.3: After the deep belief network training is completed, a nonlinear mapping relationship between the feature matrix and the actual printing pressure value is established;

[0040] Step 8.4: Input the test set data into the trained deep belief network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep belief network pressure prediction model.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] The present invention proposes a flexographic printing pressure prediction method based on a deep belief network, which obtains the image and text information feature matrix by extracting three image and text information features of the color separation original, namely the image and text area, the maximum gradient value, and the grayscale co-occurrence matrix; and proposes a flexographic printing pressure prediction model based on a deep belief network, which realizes the effective prediction of the optimal printing pressure of flexographic printing. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 It is an algorithm flow chart of a flexographic printing pressure prediction method based on a deep belief network of the present invention. DETAILED DESCRIPTION

[0044] The present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0045] A flexographic printing pressure prediction method based on deep belief network, such as Figure 1 As shown, the specific implementation steps are as follows:

[0046] Step 1: Binarize the color separation original image required for flexographic printing to obtain a binary image;

[0047] Step 2: The binary image obtained in step 1 is inverted to obtain an inverted binary image;

[0048] Step 3: Divide the inverted binary image obtained in step 2 into 28×28 grids, calculate the area of ​​the image and text in each grid area, and obtain an area matrix; specifically, follow the following steps:

[0049] Step 3.1: Divide the inverted binary image obtained in step 2 into a regular grid of 28*28 for calculating the area of ​​the image and text in the grid area. The number of grid divisions has an important influence on the experimental results. After optimization, the grid is finally divided into 28×28;

[0050] Step 3.2: After processing in step 3.1, mark the number of connected regions in the grid, then count the number of pixels in each connected region, and finally add the number of pixels in each connected region to obtain the area of ​​the image and text in the grid;

[0051] Step 3.3: Calculate the area of ​​the image and text in each grid, and finally obtain a 28*28 area matrix;

[0052] Step 4: Perform gradient calculation on the color separation original image required for flexographic printing to obtain a gradient image; the specific implementation method is as follows:

[0053] Step 4.1: Use the commonly used Robert operator horizontal gradient template to calculate the color separation original image to obtain a horizontal gradient image;

[0054] Step 4.2: Use the commonly used Robert operator longitudinal gradient template to calculate the color separation original image to obtain a longitudinal gradient image;

[0055] Step 4.3: After steps 4.1 and 4.2, calculate the modulus of the gradient to obtain a gradient image. The modulus of the gradient is usually referred to as the gradient. Therefore, the calculation of the gradient is:

[0056]

[0057] In the formula, Gx is the transverse gradient, Gy is the longitudinal gradient, and G is the modulus of the gradient, that is, the gradient. After calculating the modulus of the gradient, the gradient image can be obtained;

[0058] Step 5: Divide the gradient image obtained in step 4 into a regular grid of 28*28, calculate the maximum gradient value in the grid area, and obtain a 28*28 maximum gradient value matrix;

[0059] Step 6: Calculate the grayscale co-occurrence matrix of the color separation original image required for flexographic printing, and use the graycomatrix function provided by MATLAB to calculate and obtain the grayscale co-occurrence matrix;

[0060] Step 7: Arrange the three groups of vectors, namely, the area matrix, the maximum gradient value matrix, and the gray-level co-occurrence matrix, obtained in steps 3, 5, and 6, in parallel horizontally to obtain a characteristic matrix of the color separation original image;

[0061] Step 8: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation manuscript image on the flexographic printing machine. At the same time, use the feature matrix of the color separation manuscript image obtained in step 7 as the input data of the deep belief network model; use the actual optimal printing pressure value as the output data of the deep belief network model, establish a deep belief network pressure prediction model, optimize the model parameters, and obtain the optimal deep belief network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep belief network pressure prediction model, which is specifically implemented according to the following method:

[0062] Step 8.1: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation manuscript image on the flexographic printing machine, and use the feature matrix of the color separation manuscript image obtained in step 7 as the input data of the deep belief network pressure prediction model, and the actual printing pressure value as the output data of the deep belief network pressure prediction model, and divide the data set consisting of the input data and the output data into a training set and a test set;

[0063] Step 8.2: Input the data in the training set into the deep belief network pressure prediction model structure, train the deep belief network, and obtain the optimal deep belief network pressure prediction model after normalization, network model evaluation and parameter optimization;

[0064] The specific process is:

[0065] (1) Normalize the feature matrix. Use the linear normalization method to normalize the feature matrix to the interval [0,1]. The expression is as follows:

[0066]

[0067] In the formula, X and X' represent the original data and the normalized data, respectively, and Xmin and Xmax represent the minimum and maximum values ​​of the original data, respectively;

[0068] (2) The evaluation parameters of the deep belief network model used are: mean absolute percentage error MAPE, linear regression correlation coefficient R and cumulative prediction accuracy function CS(θ):

[0069] Among them, MAPE is the expected value of the square of the difference between the predicted value and the actual value, which reflects the overall deviation of the model's predicted value relative to the actual value. The smaller the value of MAPE, the better the accuracy of the prediction model.

[0070] The regression coefficient R reflects the correlation between the predicted value and the actual value. The closer the R value is to 1, the better the correlation between the experimental data described by the prediction model and the actual value.

[0071] CS(θ) is the percentage of samples whose absolute error between the predicted value and the actual value is less than θ in the total number of samples. Its function expression is:

[0072]

[0073] Where N e ≤θ is the number of data with prediction error less than θ, and N is the number of prediction samples;

[0074] (3) After parameter optimization, the deep belief network structure is finally determined. The network adopts a four-layer structure, the neuron structure is 2352-1568-784-500, the network learning rate is 0.005, the network momentum is 0.8, the number of iterations is 100, and the number of random samples for each iteration is 5. The deep belief network assigns the learned weights to the artificial network by unfolding, and realizes pressure prediction through the artificial neural network. The activation function used by the artificial neural network is the sigm function, the activation function of the network output layer is the linear function, the network learning rate is 0.005, the network momentum is 0.9, the number of iterations is 100, and the number of random samples for each iteration is 5;

[0075] Step 8.3: After the deep belief network training is completed, a nonlinear mapping relationship between the feature matrix and the actual printing pressure value is established;

[0076] Step 8.4: Input the test set data into the trained deep belief network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep belief network pressure prediction model.

[0077] The above-mentioned embodiments only express the specific implementation of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of the invention patent. It should be pointed out that for ordinary technicians in this field, several modifications and improvements can be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be based on the attached claims.

Claims

1. A flexographic printing pressure prediction method based on deep belief network, characterized in that: Follow the steps below to implement it: Step 1: Binarize the color separation original image required for flexographic printing to obtain a binary image; Step 2: The binary image obtained in step 1 is inverted to obtain an inverted binary image; Step 3: Divide the inverted binary image obtained in step 2 into 28×28 grids, calculate the area of ​​the image and text in each grid area, and obtain the area matrix; Step 4: Perform gradient calculation on the color separation original image required for flexographic printing to obtain a gradient image; Step 5: Divide the gradient image obtained in step 4 into 28×28 grids, calculate the maximum gradient value in each grid, and obtain the maximum gradient value matrix; Step 6: Calculate the grayscale co-occurrence matrix of the color separation original image required for flexographic printing, and use the graycomatrix function provided by MATLAB to calculate and obtain the grayscale co-occurrence matrix; Step 7: Arrange the three groups of vectors, namely, the area matrix, the maximum gradient value matrix, and the gray-level co-occurrence matrix, obtained in steps 3, 5, and 6, in parallel horizontally to obtain a characteristic matrix of the color separation original image; Step 8: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation original image on the flexographic printing press, and at the same time, use the feature matrix of the color separation original image obtained in step 7 as the input data of the deep belief network model; The actual optimal printing pressure value is used as the output data of the deep confidence network model. A deep confidence network pressure prediction model is established. The model parameters are optimized to obtain the optimal deep confidence network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep confidence network pressure prediction model.

2. The method for predicting flexographic printing pressure based on deep belief network according to claim 1, characterized in that: The step 3 is specifically implemented according to the following method: Step 3.1: Divide the inverted binary image obtained in step 2 into a regular grid of 28*28 for calculating the area of ​​the image and text in the grid area. The number of grid divisions has an important influence on the experimental results. After optimization, the grid is finally divided into 28×28; Step 3.2: After processing in step 3.1, mark the number of connected regions in the grid, then count the number of pixels in each connected region, and finally add the number of pixels in each connected region to obtain the area of ​​the image and text in the grid; Step 3.3: Calculate the area of ​​the image and text in each grid, and finally obtain a 28*28 area matrix.

3. The method for predicting flexographic printing pressure based on deep belief network according to claim 1, characterized in that: The step 4 is specifically implemented according to the following method: Step 4.1: Use the commonly used Robert operator horizontal gradient template to perform convolution calculation on the color separation original image to obtain a horizontal gradient image; Step 4.2: Use the commonly used Robert operator longitudinal gradient template to perform convolution calculation on the color separation original image to obtain a longitudinal gradient image; Step 4.3: After steps 4.1 and 4.2, calculate the modulus of the gradient to obtain a gradient image. The modulus of the gradient is usually referred to as the gradient. Therefore, the calculation of the gradient is: In the formula, Gx is the transverse gradient, Gy is the longitudinal gradient, and G is the modulus of the gradient, that is, the gradient; after calculating the modulus of the gradient, the gradient image can be obtained.

4. The method for predicting flexographic printing pressure based on deep belief network according to claim 1, characterized in that: The step 8 is specifically implemented according to the following method: Step 8.1: Collect the actual optimal printing pressure value of the printing plate corresponding to the color separation manuscript image on the flexographic printing machine, and use the feature matrix of the color separation manuscript image obtained in step 7 as the input data of the deep belief network pressure prediction model, and the actual printing pressure value as the output data of the deep belief network pressure prediction model, and divide the data set consisting of the input data and the output data into a training set and a test set; Step 8.2: Input the data in the training set into the deep belief network pressure prediction model structure, train the deep belief network, and obtain the optimal deep belief network pressure prediction model after normalization, network model evaluation and parameter optimization; The specific process is: (1) Normalize the feature matrix. Use the linear normalization method to normalize the feature matrix to the interval [0,1]. The expression is as follows: In the formula, X and X' represent the original data and the normalized data, respectively, and Xmin and Xmax represent the minimum and maximum values ​​of the original data, respectively; (2) The evaluation parameters of the deep belief network model used are: mean absolute percentage error MAPE, linear regression correlation coefficient R and cumulative prediction accuracy function CS(θ): Among them, MAPE is the expected value of the square of the difference between the predicted value and the actual value, which reflects the overall deviation of the model's predicted value relative to the actual value. The smaller the value of MAPE, the better the accuracy of the prediction model. The regression coefficient R reflects the correlation between the predicted value and the actual value. The closer the R value is to 1, the better the correlation between the experimental data described by the prediction model and the actual value. CS(θ) is the percentage of samples whose absolute error between the predicted value and the actual value is less than θ in the total number of samples. Its function expression is: Where N e ≤θ is the number of data with prediction error less than θ, and N is the number of prediction samples; (3) After parameter optimization, the structure of the deep belief network is finally determined. The network adopts a four-layer structure, the neuron structure is 2352-1568-784-500, the network learning rate is 0.005, the network momentum is 0.8, the number of iterations is 100, and the number of random samples in each iteration is 5. The deep belief network assigns the learned weights to the artificial network by unfolding, and the pressure prediction is realized through the artificial neural network. The activation function of the artificial neural network is the sigm function, the activation function of the network output layer is the linear function, the network learning rate is 0.005, the network momentum is 0.9, the number of iterations is 100, and the number of random samples in each iteration is 5; Step 8.3: After the deep belief network training is completed, a nonlinear mapping relationship between the feature matrix and the actual printing pressure value is established; Step 8.4: Input the test set data into the trained deep belief network pressure prediction model. The optimal printing pressure value can be predicted by the optimal deep belief network pressure prediction model.

Images