Printed matter quality detection method based on artificial intelligence

Abstract

The invention relates to the field of printed matter quality detection, in particular to a printed matter quality detection method based on artificial intelligence. Comprising the steps of collecting image data, constructing and enhancing a data set, correcting an image by adopting a feature point detection and matching algorithm, extracting an image feature vector through a pre-trained deep learning model VGG16, inputting the feature vector into a support vector machine classifier for quality classification, and giving a quality problem description detection report for unqualified printed matters. Wherein the image enhancement comprises zooming, contrast adjustment and color conversion; the feature point detection and matching algorithm can eliminate image distortion; the VGG16 model calculates feature mapping and extracts feature vectors through a convolution layer and a pooling layer, and finally standardization processing is carried out. And the support vector machine classifier outputs a classification result according to the decision function and performs quality feedback. The method improves the quality detection speed and efficiency of the printed matter.

Description

Technical Field

[0001] The present invention relates to the field of printed matter quality detection, and in particular to a printed matter quality detection method based on artificial intelligence. Background Art

[0002] In the modern printing industry, the quality inspection of printed products is of vital importance. The existing methods for inspecting the quality of printed products mainly rely on manual visual inspection, which is not only inefficient but also easily affected by human factors, resulting in inaccurate inspection results. With the continuous development of science and technology, artificial intelligence technology has been widely used in various fields, bringing new opportunities for printed product quality inspection. Therefore, in order to improve the efficiency and accuracy of printed product quality inspection, there is an urgent need for an artificial intelligence-based printed product quality inspection method that can automatically and quickly detect various defects in printed products and give accurate quality assessment results. This method can not only reduce labor costs and improve production efficiency, but also ensure the quality stability of printed products and meet the market demand for high-quality printed products. Summary of the invention

[0003] In view of the technical problems existing in the above-mentioned printing technology, the present invention proposes a method which is simple, easy to operate and can effectively judge whether the quality of printed products is qualified.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is a printed matter quality detection method based on artificial intelligence, comprising the following steps:

[0005] S1. First, collect the image data of the printed product, build the initial image data set, and perform data enhancement on the data set;

[0006] S2. Secondly, feature point detection and matching algorithm are used to perform image correction to eliminate image distortion and improve the quality of image data samples;

[0007] S3, then extracting the feature vector of the image through the deep learning model VGG16, wherein the deep learning model VGG16 is pre-trained with a large amount of printed image data;

[0008] S4, then input the extracted feature vector into the support vector machine classifier to classify the quality of the printed matter and determine whether it is qualified;

[0009] S5. Finally, for unqualified printed products, a corresponding quality problem description test report is given;

[0010] The process of extracting the feature vector of the image by the deep learning model VGG16 in step S3 is:

[0011] S31. First, select the VGG16 network as the feature extraction model. VGG16 means a convolutional network with 16 layers of depth.

[0012] S32, secondly, adjusting the historical printed image to a size that meets the VGG16 input requirements, and training the model;

[0013] S33, then forward propagation is performed, the preprocessed image is input into the VGG16 model, and the feature map is calculated through the convolution layer and pooling layer of the model;

[0014] S34, the feature vector is extracted below. The output of the convolution layer and the pooling layer is used to flatten the feature map into a one-dimensional vector to obtain the feature vector of the image. The expansion operation is flatten (I pool )=[I pool [0],I pool [1],...,I pool [n]] Among them, I pool It is the feature map after pooling, and the flattening operation converts it into a one-dimensional feature vector;

[0015] S35. Finally, post-process the feature vector and standardize the feature vector to improve the classification performance.

[0016] Preferably, the step S1 uses image enhancement including scaling, contrast adjustment and color transformation operations; the scaling rule is to perform scaling operations on the image, including zooming in or out, with a scaling ratio of 0.8-1.2 times; the contrast adjustment is to increase or decrease the contrast of the image to enhance the image; the color transformation is to adjust the hue, saturation and brightness of the image.

[0017] Preferably, the specific operation of the feature point detection and matching algorithm used in step S2 is:

[0018] S21. First, use the scale-invariant feature transformation principle to detect feature points in the image: perform Gaussian blur of different scales on the image: L(x,y,σ)=G(x,y,σ)*I(x,y), where L(x,y,σ) is the blurred image, G(x,y,σ) is the Gaussian kernel, I(x,y) is the original image, * represents convolution calculation, calculate the difference between Gaussian blurred images of different scales: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where k is the scale space factor, detect extreme points on the differential Gaussian pyramid as feature points;

[0019] S22, then calculate the gradient direction and amplitude of the neighborhood of the feature point for the detected feature point:

[0020]

[0021] Where m(x,y) is the gradient amplitude, θ(x,y) is the gradient direction, and the gradient direction and amplitude are combined into a feature vector;

[0022] S23, then matching the feature points in the two images, and calculating the Euclidean distance between the two feature point descriptors;

[0023] S24. Then use the random sampling consistency algorithm to remove the wrongly matched feature point pairs, randomly select the minimum number of feature point pairs, calculate the transformation model: Hx = x', where H is the homography matrix, x, x' are the matched feature point pairs, continue to calculate the errors of all feature point pairs, and find the point pairs that meet the model: Repeat the above steps to find the model with the smallest error;

[0024] S25. Finally, image correction is performed, the original image is transformed into a corrected image using a homography matrix, and interpolation calculation is performed on the transformed pixels to generate a corrected image.

[0025] Preferably, the step S4 inputs the extracted feature vector into a support vector machine classifier, and the specific operation of classifying the quality of the printed product is:

[0026] S41. First, given a new feature vector x, the decision function of the support vector machine classifier is f(x)=sign(w T x+b), where w is the weight vector, b is the bias term, and the sign function sign is used to output the final classification result;

[0027] S42, then use the cross-validation method to adjust the parameters and select the best parameter combination;

[0028] S43. Finally, a prediction is performed to output the classification results as qualified or unqualified, and quality feedback is provided based on the classification results.

[0029] Preferably, in step S5, for unqualified books with color difference, scratches and stains, the detected defective areas should be marked, the locations of the problems should be marked, and a test report should be generated.

[0030] Compared with the prior art, the advantages and positive effects of the present invention are:

[0031] 1. It realizes automated detection, greatly improves the detection speed, saves manpower and time costs, and improves production efficiency.

[0032] 2. Through data augmentation operations, the data set is enriched, enabling the model to learn more features of printed products in different situations, enhancing the generalization ability of the model and the accuracy of the prediction results.

[0033] 3. Feature point detection and matching algorithms can effectively eliminate image distortion, improve image quality, and more accurately identify and locate defects and problems in printed products.

[0034] 4. The combination of deep learning model VGG16 and support vector machine classifier improves the classification performance and can more accurately judge whether the quality of printed materials is qualified. DETAILED DESCRIPTION

[0035] In order to more clearly understand the above-mentioned purpose, features and advantages of the present invention, the present invention is further described below in conjunction with embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0036] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways than those described herein. Therefore, the present invention is not limited to the specific embodiments of the following disclosure.

[0037] Embodiment, the existing printed product detection method is easily affected by human factors, resulting in inaccurate detection results. With the emergence of artificial intelligence, new opportunities have been brought for printed product quality detection. The present invention proposes a printed product quality detection method based on artificial intelligence. First, the image data of the printed product is collected, an initial image data set is constructed, and the data set is enhanced. Through data enhancement, more diverse image data can be generated, the data set can be enriched, and the model can learn more printed product features under different circumstances, thereby enhancing the accuracy of the prediction results. The image enhancement includes scaling, contrast adjustment and color transformation operations; the scaling rule is to perform scaling operations on the image, including zooming in or out, and the scaling ratio is 0.8-1.2 times; the contrast adjustment is to increase or decrease the contrast of the image to enhance the image; the color transformation is to adjust the hue of the image, change the saturation, and adjust the brightness.

[0038] Considering that the image after data enhancement may have image quality problems such as graphic distortion, feature point detection and matching algorithm is used for image correction. This precise image correction helps to improve the accuracy of print quality detection and can more accurately identify and locate defects and problems in prints.

[0039] The specific operation of the image correction is to first detect feature points in the image using the principle of scale-invariant feature transformation: Gaussian blur of different scales is performed on the image L(x,y,σ)=G(x,y,σ)*I(x,y), where L(x,y,σ) is the blurred image, G(x,y,σ) is the Gaussian kernel, I(x,y) is the original image, * represents convolution calculation, and the difference is calculated between Gaussian blurred images of different scales: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where k is the scale space factor, and extreme points are detected on the differential Gaussian pyramid as feature points. Then, the gradient direction and amplitude of the neighborhood of the feature point are calculated for the detected feature point:

[0040]

[0041] Where m(x,y) is the gradient amplitude, θ(x,y) is the gradient direction, and the gradient direction and amplitude are combined into a feature vector; secondly, the feature points in the two images are matched, and the Euclidean distance between the two feature point descriptors is calculated; then the random sampling consistency algorithm is used to remove the mismatched feature point pairs, and the minimum number of feature point pairs are randomly selected to calculate the transformation model: Hx = x′, where H is the homography matrix, x, x′ are the matched feature point pairs, and the errors of all feature point pairs are calculated to find the point pairs that meet the model: Repeat the above steps to find the model with the smallest error; then perform image correction, apply the homography matrix to transform the original image into the corrected image, and perform interpolation calculation on the transformed pixels to generate the corrected image.

[0042] Further, after training a large amount of printed image data, the deep learning model VGG16 is used to extract the feature vector of the image. Extracting the feature vector of the image is a key step in classifying and judging whether the printed image is qualified. First, the VGG16 network is selected as the feature extraction model. VGG16 means a convolutional network with 16 layers of depth; secondly, the historical printed image is adjusted to a size that meets the input requirements of VGG16, and the model is trained; then forward propagation is performed, and the pre-processed image is input into the VGG16 model, and the feature map is calculated through the convolution layer and pooling layer of the model; the feature vector is extracted next, and the output of the convolution layer and the pooling layer is used to flatten the feature map into a one-dimensional vector to obtain the feature vector of the image. The expansion operation is flatten(I pool )=[I pool [0],I pool [1],...,I pool [n]] Among them, I poolIt is the feature map after pooling, and the flattening operation converts it into a one-dimensional feature vector; finally, the feature vector is post-processed and standardized to improve the classification performance.

[0043] Finally, these feature vectors are input into the support vector machine classifier, and the classifier determines whether the quality of the printed product belongs to the qualified or unqualified category based on the feature vector. First, given a new feature vector x, the decision function of the support vector machine classifier is f(x) = sign(w T x+b), where w is the weight vector, b is the bias term, and the symbol function sign is used to output the final classification result; then the cross-validation method is used to adjust the parameters and select the best parameter combination; then the prediction output classification results are qualified or unqualified, and quality feedback is given based on the classification results.

[0044] The above description is only a preferred embodiment of the present invention and does not limit the present invention in other forms. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention still falls within the protection scope of the technical solution of the present invention.

Claims

1. A printed product quality detection method based on artificial intelligence, characterized in that: The following steps are involved: S1. First, collect the image data of the printed product, build the initial image data set, and perform data enhancement on the data set; S2. Secondly, feature point detection and matching algorithm are used to perform image correction to eliminate image distortion and improve the quality of image data samples; S3, then extracting the feature vector of the image through the deep learning model VGG16, wherein the deep learning model VGG16 is pre-trained with a large amount of printed image data; S4, then input the extracted feature vector into the support vector machine classifier to classify the quality of the printed matter and determine whether it is qualified; S5. Finally, for unqualified printed products, a corresponding quality problem description test report is given; The process of extracting the feature vector of the image by the deep learning model VGG16 in step S3 is: S31. First, select the VGG16 network as the feature extraction model. VGG16 means a convolutional network with 16 layers of depth. S32, secondly, adjusting the historical printed image to a size that meets the VGG16 input requirements, and training the model; S33, then forward propagation is performed, the preprocessed image is input into the VGG16 model, and the feature map is calculated through the convolution layer and pooling layer of the model; S34, the feature vector is extracted below. The output of the convolution layer and the pooling layer is used to flatten the feature map into a one-dimensional vector to obtain the feature vector of the image. The expansion operation is flatten (I pool )=[I pool [0],I pool [1],...,I pool [n]] Among them, I pool It is the feature map after pooling, and the flattening operation converts it into a one-dimensional feature vector; S35. Finally, post-process the feature vector and standardize the feature vector to improve the classification performance.

2. The artificial intelligence-based printed matter quality detection method according to claim 1, characterized in that: The step S1 uses image enhancement including scaling, contrast adjustment and color transformation operations; the scaling rule is to perform scaling operations on the image, including zooming in or out, with a scaling ratio of 0.8-1.2 times; the contrast adjustment is to increase or decrease the contrast of the image to enhance the image; the color transformation is to adjust the hue, saturation and brightness of the image.

3. The artificial intelligence-based printed matter quality detection method according to claim 1, characterized in that: The specific operation of the feature point detection and matching algorithm in step S2 is as follows: S21. First, use the scale-invariant feature transformation principle to detect feature points in the image: perform Gaussian blur of different scales on the image: L(x,y,σ)=G(x,y,σ)*I(x,y), where L(x,y,σ) is the blurred image, G(x,y,σ) is the Gaussian kernel, I(x,y) is the original image, * represents convolution calculation, calculate the difference between Gaussian blurred images of different scales: D(x,y,σ)=L(x,y,kσ)-L(x,y,σ), where k is the scale space factor, detect extreme points on the differential Gaussian pyramid as feature points; S22, then calculate the gradient direction and amplitude of the neighborhood of the feature point for the detected feature point: Where m(x,y) is the gradient amplitude, θ(x,y) is the gradient direction, and the gradient direction and amplitude are combined into a feature vector; S23, then matching the feature points in the two images, and calculating the Euclidean distance between the two feature point descriptors; S24. Then use the random sampling consistency algorithm to remove the wrongly matched feature point pairs, randomly select the minimum number of feature point pairs, calculate the transformation model: Hx = x', where H is the homography matrix, x, x' are the matched feature point pairs, continue to calculate the errors of all feature point pairs, and find the point pairs that meet the model: Repeat the above steps to find the model with the smallest error; S25. Finally, image correction is performed, the original image is transformed into a corrected image using a homography matrix, and interpolation calculation is performed on the transformed pixels to generate a corrected image.

4. The method for detecting quality of printed matter based on artificial intelligence according to claim 1, characterized in that: The step S4 inputs the extracted feature vector into the support vector machine classifier, and the specific operation of classifying the quality of the printed product is as follows: S41. First, given a new feature vector x, the decision function of the support vector machine classifier is f(x)=sign(w T x+b), where w is the weight vector, b is the bias term, and the sign function sign is used to output the final classification result; S42, then use the cross-validation method to adjust the parameters and select the best parameter combination; S43. Finally, a prediction is performed to output the classification results as qualified or unqualified, and quality feedback is provided based on the classification results.

5. The method for detecting quality of printed matter based on artificial intelligence according to claim 1, characterized in that: In step S5, for unqualified books with color difference, scratches and stains, the detected defective areas should be marked, the locations of the problems should be marked, and a test report should be generated.

Images