Intelligent control system of inkjet printer based on computer vision

Through computer vision technology, 3D reconstruction is carried out using quaternion rotation and linear interpolation, combining single-view segmentation and multi-view feature interaction, the network structure is optimized, and the problem of insufficient adaptability of the injector to multi-angle surfaces and network structures is solved. , improves the accuracy of spray painting abnormality detection and the effect of intelligent control of spray painting machine.

CN118840619BActive Publication Date: 2024-11-22ANHUI LIYU COMPUTER EQUIP MFG CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411324411.6
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-23
Publication Date
2024-11-22
Estimated Expiration
2044-09-23

AI Technical Summary

Technical Problem

The existing sprayer control system has weak adaptability to multi-angle and complex spray surfaces, resulting in low efficiency in identifying spray surface abnormalities, unable to distinguish between normal and abnormal spray areas, high leakage detection rate, and network structure cannot adapt to the accuracy of spray painting abnormal detection. and stability, resulting in poor control effect of the spray painting process.

Method used

Using an intelligent control system based on computer vision, 3D reconstruction and linear interpolation are used to generate new viewpoints through quaternion rotation, improving the model's adaptability to multi-angle and complex surfaces; single-view segmentation model training and multi-view feature interaction are carried out to form More expressive feature representation; optimize the network structure through secondary segmentation and adaptive adjustment to adapt to the abnormal detection task; use nested linear transformation and nonlinear activation functions for deep feature representation, enhance the system to paint the surface Abnormal recognition ability.

Benefits of technology

The accuracy of abnormal detection and control strategies during the injection painting process are improved, the system's ability to identify abnormalities on the injection painting surface is enhanced, the intelligent control effect of the injection painting machine is improved, the leakage detection rate is reduced and the control effect is improved.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118840619B_ABST
    Figure CN118840619B_ABST
Patent Text Reader

Abstract

The present invention discloses an intelligent control system for inkjet printers based on computer vision, including a data acquisition module, an image enhancement module, an inkjet printer anomaly detection model building module and an inkjet printer intelligent control module. The present invention belongs to the field of intelligent control technology, and specifically refers to an intelligent control system for inkjet printers based on computer vision. The scheme performs 3D reconstruction based on quaternion rotation, generates new viewpoints based on linear interpolation, and improves the model's adaptability to multi-angle and complex inkjet printing surfaces; performs single-view segmentation model training and multi-view feature interaction to form a more expressive feature representation, thereby improving the accuracy of anomaly detection, and further optimizing the control strategy during the inkjet printing process; adds nodes to each layer and adaptively adjusts according to the residual reduction; performs deep feature representation through nested linear transformations and nonlinear activation functions, and enhances the system's ability to identify abnormalities on the inkjet printing surface, thereby improving the system's intelligent control of the inkjet printer.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the field of intelligent control technology, and in particular to an intelligent control system for an inkjet printer based on computer vision. Background Art

[0002] The inkjet printing anomaly detection system is a comprehensive system for managing and optimizing inkjet printer operations. It realizes intelligent control of the inkjet printing process by detecting anomalies in the inkjet printed images, improving the quality and efficiency of inkjet printing, and reducing errors and anomalies. However, the general control system of inkjet printers has a weak adaptability to multi-angle and complex inkjet printing surfaces, which leads to low efficiency in identifying abnormalities on the inkjet printing surface, and cannot distinguish between normal and abnormal inkjet printing areas, resulting in a high missed detection rate. The network structure of the general control system of inkjet printers cannot adapt to the accuracy and stability of inkjet printing anomaly detection, and the feature mapping has a weak ability to identify abnormalities on the inkjet printing surface, which leads to poor control of the inkjet printing process. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides an intelligent control system for inkjet printers based on computer vision. In view of the fact that the general inkjet printer control system has weak adaptability to multi-angle and complex inkjet printing surfaces, which leads to low efficiency in identifying abnormalities on the inkjet printing surface, inability to distinguish between normal and abnormal inkjet printing areas, and high missed detection rate, the present invention performs 3D reconstruction based on quaternion rotation, captures the details of the inkjet printing surface more accurately through multiple viewpoints, generates new viewpoints based on linear interpolation, and improves the adaptability of the model to multi-angle and complex inkjet printing surfaces; generates a rotation matrix and projects the 3D image to a 2D image for accurate positioning and identification of abnormalities on the inkjet printing surface; performs single-view segmentation model training and multi-view feature segmentation The features interact with each other to form a more expressive feature representation, thereby improving the accuracy of anomaly detection and optimizing the control strategy in the printing process; the network structure of the general inkjet printer control system cannot adapt to the accuracy and stability of inkjet printing anomaly detection, and the feature mapping has weak recognition ability for inkjet surface anomalies, which leads to poor control effect of the inkjet printing process. This scheme performs fine-grained analysis through secondary segmentation, adds nodes to each layer and adaptively adjusts according to the reduction of residuals to optimize the network structure to adapt to the inkjet printing anomaly detection task; deep feature representation is performed through nested linear transformation and nonlinear activation function to enhance the system's recognition ability for inkjet surface anomalies, thereby improving the system's intelligent control of inkjet printers.

[0004] The technical solution adopted by the present invention is as follows: the computer vision-based intelligent control system of the inkjet printer provided by the present invention includes a data acquisition module, an image enhancement module, an inkjet abnormality detection model establishment module and an inkjet printer intelligent control module;

[0005] The data acquisition module acquires historical inkjet printed surface image data;

[0006] The image enhancement module performs 3D reconstruction based on quaternion rotation, generates new viewpoints based on linear interpolation, generates a rotation matrix and projects 3D to a 2D image, performs single-view segmentation model training and performs multi-view feature interaction, and finally realizes image enhancement;

[0007] The printing anomaly detection model building module performs fine-grained analysis through secondary segmentation, adds nodes to each layer and adaptively adjusts according to the residual reduction, represents deep features through nested linear transformation and nonlinear activation function, and performs pseudo-inverse solution to model weights, and finally realizes the establishment of the printing anomaly detection model;

[0008] The inkjet printer intelligent control module realizes intelligent control of the inkjet printer based on the established inkjet printer abnormality detection model.

[0009] Furthermore, in the data acquisition module, enhanced image annotation is performed on the historical inkjet surface image data, and the enhanced image annotation categories are foreground and background, which are only used for image enhancement.

[0010] Furthermore, the image enhancement module specifically includes the following contents:

[0011] Image reconstruction; 3D reconstruction based on quaternion rotation formula, the formula used is as follows:

[0012] ;

[0013] ;

[0014] In the formula, q n It is a quaternion, used to represent rotation; is the rotation angle; is the unit vector of the rotation axis, , and They are Components along the x, y, and z axes; n is the time index;

[0015] Generate new viewpoints based on linear interpolation, using the following formula:

[0016] ;

[0017] ;

[0018] In the formula, and are the quaternion representations of the known viewpoints at the nth moment and the n+1th moment respectively; and are the translation vectors of the known viewpoints at the nth moment and the n+1th moment respectively; is the parameter of linear interpolation; and They are the quaternion representation and translation vector of the intermediate viewpoint obtained by linear interpolation;

[0019] Generate a new rotation matrix, expressed as: ; Project 3D into 2D image for guidance of printer control, expressed as ;in, is the rotation matrix; q0, q1, q2, q3 are the components of the quaternion; and is the coordinate of the 3D point projected onto the 2D image plane; s x and y is the scaling factor; t x and t y is the translation amount; x1 and y1 are the components of the 3D coordinates on the x-axis and y-axis;

[0020] Single view segmentation model training; Use augmented data to train the ResUNet segmentation model, and generate pseudo labels through a semi-supervised learning strategy to enhance the training data; Construct a loss function, expressed as: Where, L tt is the total loss function; is the segmentation probability of the i-th pixel; is the true label of the i-th pixel; γ is the adjustment factor; is a smoothing term;

[0021] Multi-view feature interaction; calculate feature similarity, expressed as: ; Get the similarity feature map, expressed as: ; Calculate the optimized features, expressed as: ;in, is the feature similarity, g x and g y is the eigenvector; is the mold length; is the similarity value; g1 is the feature of the reference image collected by the inkjet printer; is the eigenvalue; p and q are the position indexes in the feature map; It is an updated feature; is the interactive feature; N is the total number of features; the enhanced image data set is obtained, and the image data set is annotated, and the annotation types are normal printing and abnormal printing.

[0022] Furthermore, the printing anomaly detection model building module is to build an adaptive neural network model, which specifically includes the following contents:

[0023] Secondary segmentation; the segmentation formula for each input image is expressed as ; Where Q is the number of regions after segmentation, H is the input image height, Ph is the height of the segmentation window, stride is the step size, W is the input image width, and Pw is the width of the segmentation window;

[0024] Adaptive network; add nodes to each layer and check the reduction of residual. If the residual reduction is less than , then stop modeling this layer and build the next layer; the residual reduction value calculation formula is: ; The residual calculation is expressed as ;in, is the mean residual reduction, and are the residual values ​​of the nth layer network and the residual values ​​of the first ns layers, respectively, and s is the number of nodes added in the current layer; is the residual of the k+1th layer, and are the error vectors of the kth layer and the k+1th layer respectively, is the feature response vector of the k+1th layer; r is the weight factor;

[0025] Feature mapping calculation: Use nested linear transformations and nonlinear activation functions to build feature representations layer by layer, as shown below:

[0026] ; In the formula, is the nth layer for input x i2 feature representation; k is the layer index, is the number of nodes; is the weight coefficient; is the sigmoid activation function; is the feature weight vector; is the input vector; is the additional weight vector; is the bias term; is the initial feature representation;

[0027] Model optimization: After configuring the nodes, perform pseudo-inverse solution to the model weights, obtain the optimal model configuration, and output the prediction results; the pseudo-inverse solution is expressed as: ;in, is the optimal model weight vector solved; is the weight vector of the model; H is the input feature matrix; T is the target matrix; It is the Moore-Penrose pseudo-inverse; is the square of the Frobenius norm.

[0028] Furthermore, the inkjet printer intelligent control module is based on the established inkjet printing anomaly detection model, collects inkjet printing surface image data in real time, and inputs it into the inkjet printing anomaly detection model after image enhancement. If the inkjet printing anomaly detection model outputs abnormal inkjet printing, early warning processing is performed to realize intelligent control of the inkjet printer.

[0029] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0030] (1) In view of the fact that general inkjet printer control systems have weak adaptability to multi-angle and complex printed surfaces, which leads to low efficiency in identifying printed surface anomalies, inability to distinguish between normal and abnormal printed areas, and high missed detection rate, this scheme performs 3D reconstruction based on quaternion rotation, captures the details of the printed surface more accurately through multiple viewpoints, generates new viewpoints based on linear interpolation, and improves the model's adaptability to multi-angle and complex printed surfaces; generates a rotation matrix and projects the 3D image onto a 2D image for accurate positioning and identification of printed surface anomalies; trains a single-view segmentation model and performs multi-view feature interaction to form a more expressive feature representation, thereby improving the accuracy of anomaly detection and optimizing the control strategy during the printing process.

[0031] (2) In view of the problem that the network structure of general inkjet printer control systems cannot adapt to the accuracy and stability of inkjet anomaly detection, and the feature mapping has weak recognition ability for inkjet surface anomalies, which leads to poor control of the inkjet process, this scheme performs fine-grained analysis through secondary segmentation, optimizes the network structure to adapt to the inkjet anomaly detection task by adding nodes to each layer and adaptively adjusting according to the reduction of residuals; deep feature representation is performed through nested linear transformation and nonlinear activation function, which enhances the system's recognition ability for inkjet surface anomalies, thereby improving the system's intelligent control of inkjet printers. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 A schematic diagram of the flow chart of the intelligent control system of the inkjet printer based on computer vision provided by the present invention;

[0033] Figure 2 It is a flowchart of the image enhancement module;

[0034] Figure 3 Schematic diagram of the process of building a module for the inkjet anomaly detection model.

[0035] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0037] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0038] Example 1, see Figure 1 The computer vision-based intelligent control system for inkjet printers provided by the present invention includes a data acquisition module, an image enhancement module, an inkjet abnormality detection model building module and an inkjet printer intelligent control module;

[0039] The data acquisition module acquires historical printing surface image data; and sends the data to the image enhancement module;

[0040] The image enhancement module receives the data sent by the data acquisition module; performs 3D reconstruction based on quaternion rotation, generates new viewpoints based on linear interpolation, generates a rotation matrix and projects the 3D image to a 2D image, performs single-view segmentation model training and performs multi-view feature interaction, and finally realizes image enhancement; and sends the data to the printing anomaly detection model establishment module;

[0041] The inkjet printing anomaly detection model establishment module receives the data sent by the image enhancement module; performs fine-grained analysis through secondary segmentation, adds nodes to each layer and adaptively adjusts according to the residual reduction situation, performs deep feature representation through nested linear transformation and nonlinear activation function, and performs pseudo-inverse solution of model weights, and finally realizes the establishment of the inkjet printing anomaly detection model; and sends the data to the inkjet printer intelligent control module;

[0042] The inkjet printer intelligent control module receives data sent by the inkjet printing anomaly detection model establishment module; and realizes intelligent control of the inkjet printer based on the established inkjet printing anomaly detection model.

[0043] Example 2, see Figure 1 This embodiment is based on the above embodiment. In the data acquisition module, the historical printing surface image data is enhanced and annotated. The enhanced image annotation categories are foreground and background, which are only used for image enhancement.

[0044] Example 3, see Figure 1 and Figure 2 This embodiment is based on the above embodiment, and the image enhancement module specifically includes the following contents:

[0045] Image reconstruction; 3D reconstruction based on quaternion rotation formula, the formula used is as follows:

[0046] ;

[0047] ;

[0048] In the formula, q n It is a quaternion, used to represent rotation; is the rotation angle; is the unit vector of the rotation axis, , and They are Components along the x, y, and z axes; n is the time index;

[0049] Generate new viewpoints based on linear interpolation, using the following formula:

[0050] ;

[0051] ;

[0052] In the formula, and are the quaternion representations of the known viewpoints at the nth moment and the n+1th moment respectively; and are the translation vectors of the known viewpoints at the nth moment and the n+1th moment respectively; is the parameter of linear interpolation; and They are the quaternion representation and translation vector of the intermediate viewpoint obtained by linear interpolation;

[0053] Generate a new rotation matrix, expressed as: ; Project 3D into 2D image for guidance of printer control, expressed as ;in, is the rotation matrix; q0, q1, q2, q3 are the components of the quaternion; and is the coordinate of the 3D point projected onto the 2D image plane; s x and y is the scaling factor; t x and t y is the translation amount; x1 and y1 are the components of the 3D coordinates on the x-axis and y-axis;

[0054] Single view segmentation model training; Use augmented data to train the ResUNet segmentation model, and generate pseudo labels through a semi-supervised learning strategy to enhance the training data; Construct a loss function, expressed as: Where, L tt is the total loss function; is the segmentation probability of the i-th pixel; is the true label of the i-th pixel; γ is the adjustment factor; is a smoothing term;

[0055] Multi-view feature interaction; calculate feature similarity, expressed as: ; Get the similarity feature map, expressed as: ; Calculate the optimized features, expressed as: ;in, is the feature similarity, g x and g y is the eigenvector; is the mold length; is the similarity value; g1 is the feature of the reference image collected by the inkjet printer; is the eigenvalue; p and q are the position indexes in the feature map; It is an updated feature; is the interactive feature; N is the total number of features; the enhanced image data set is obtained, and the image data set is annotated, and the annotation types are normal printing and abnormal printing.

[0056] By performing the above operations, the control system of general inkjet printers has weak adaptability to multi-angle and complex inkjet printing surfaces, which leads to low efficiency in identifying abnormalities on the inkjet printing surface, inability to distinguish between normal and abnormal inkjet printing areas, and high missed detection rate. This solution performs 3D reconstruction based on quaternion rotation, captures the details of the inkjet printing surface more accurately through multiple viewpoints, generates new viewpoints based on linear interpolation, and improves the model's adaptability to multi-angle and complex inkjet printing surfaces; generates a rotation matrix and projects the 3D image to a 2D image for accurate positioning and identification of abnormalities on the inkjet printing surface; trains a single-view segmentation model and performs multi-view feature interaction to form a more expressive feature representation, thereby improving the accuracy of anomaly detection and optimizing the control strategy in the inkjet printing process.

[0057] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. The printing abnormality detection model building module is to build an adaptive neural network model, which specifically includes the following contents:

[0058] Secondary segmentation; the segmentation formula for each input image is expressed as ; Where Q is the number of regions after segmentation, H is the input image height, Ph is the height of the segmentation window, stride is the step size, W is the input image width, and Pw is the width of the segmentation window;

[0059] Adaptive network; add nodes to each layer and check the reduction of residual. If the residual reduction is less than , then stop modeling this layer and build the next layer; the residual reduction value calculation formula is: ; The residual calculation is expressed as ;in, is the mean residual reduction, and are the residual values ​​of the nth layer network and the residual values ​​of the first ns layers, respectively, and s is the number of nodes added in the current layer; is the residual of the k+1th layer, and are the error vectors of the kth layer and the k+1th layer respectively, is the feature response vector of the k+1th layer; r is the weight factor;

[0060] Feature mapping calculation: Use nested linear transformations and nonlinear activation functions to build feature representations layer by layer, as shown below:

[0061] ; In the formula, is the nth layer for input x i2 feature representation; k is the layer index, is the number of nodes; is the weight coefficient; is the sigmoid activation function; is the feature weight vector; is the input vector; is the additional weight vector; is the bias term; is the initial feature representation;

[0062] Model optimization: After configuring the nodes, perform pseudo-inverse solution to the model weights, obtain the optimal model configuration, and output the prediction results; the pseudo-inverse solution is expressed as: ;in, is the optimal model weight vector solved; is the weight vector of the model; H is the input feature matrix; T is the target matrix; It is the Moore-Penrose pseudo-inverse; is the square of the Frobenius norm.

[0063] By performing the above operations, the network structure of the general inkjet printer control system cannot adapt to the accuracy and stability of inkjet anomaly detection, and the feature mapping has weak recognition ability for inkjet surface anomalies, which leads to poor control of the inkjet process. This solution performs fine-grained analysis through secondary segmentation, adds nodes to each layer and adaptively adjusts according to the reduction of residuals to optimize the network structure to adapt to the inkjet anomaly detection task; deep feature representation is performed through nested linear transformation and nonlinear activation function to enhance the system's recognition ability for inkjet surface anomalies, thereby improving the system's intelligent control of the inkjet printer.

[0064] Example 5, see Figure 1 This embodiment is based on the above embodiment. The intelligent control module of the inkjet printer is based on the established inkjet printing abnormality detection model. It collects the image data of the printing surface in real time and inputs it into the inkjet printing abnormality detection model after image enhancement. If the inkjet printing abnormality detection model outputs abnormal printing, early warning processing is performed to realize intelligent control of the inkjet printer.

[0065] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0066] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0067] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. Intelligent control system of inkjet printer based on computer vision, characterized by: The system includes a data acquisition module, an image enhancement module, a printing anomaly detection model building module and a printing machine intelligent control module; The data acquisition module acquires historical inkjet printed surface image data; The image enhancement module performs 3D reconstruction based on quaternion rotation, generates new viewpoints based on linear interpolation, generates a rotation matrix and projects 3D to a 2D image, performs single-view segmentation model training and performs multi-view feature interaction, and finally realizes image enhancement; The printing anomaly detection model building module performs fine-grained analysis through secondary segmentation, adds nodes to each layer and adaptively adjusts according to the residual reduction, represents deep features through nested linear transformation and nonlinear activation function, and performs pseudo-inverse solution to model weights, and finally realizes the establishment of the printing anomaly detection model; The inkjet printer intelligent control module realizes intelligent control of the inkjet printer based on the established inkjet printer abnormality detection model; The image enhancement module includes multi-view feature interaction; Calculate feature similarity, expressed as: ; The similarity feature map is obtained, which is expressed as: ; Calculate the optimized features, expressed as: ;in, is the feature similarity, g x and g y is the eigenvector; is the mold length; is the similarity value; g1 is the feature of the reference image collected by the inkjet printer; is the eigenvalue; p and q are the position indexes in the feature map; It is an updated feature; is the interaction feature; N is the total number of features; The enhanced image data set is obtained, and the image data set is annotated, and the annotation types are normal printing and abnormal printing.

2. The computer vision-based intelligent control system for inkjet printer according to claim 1, characterized in that: The image enhancement module specifically includes the following contents: Image reconstruction; 3D reconstruction based on quaternion rotation formula, the formula used is as follows: ; ; In the formula, q n It is a quaternion, used to represent rotation; is the rotation angle; is the unit vector of the rotation axis, , and They are Components along the x, y, and z axes; n is the time index; Generate new viewpoints based on linear interpolation, using the following formula: ; ; In the formula, and are the quaternion representations of the known viewpoints at the nth moment and the n+1th moment respectively; and are the translation vectors of the known viewpoints at the nth moment and the n+1th moment respectively; is the parameter of linear interpolation; and They are the quaternion representation and translation vector of the intermediate viewpoint obtained by linear interpolation; Generate a new rotation matrix, expressed as: ; Project 3D into 2D image for guidance of printer control, expressed as ;in, is the rotation matrix; q0, q1, q2, q3 are the components of the quaternion; and is the coordinate of the 3D point projected onto the 2D image plane; s x and y is the scaling factor; t x and t y is the translation amount; x1 and y1 are the components of the 3D coordinates on the x-axis and y-axis; Single view segmentation model training; Use augmented data to train the ResUNet segmentation model, and generate pseudo labels through a semi-supervised learning strategy to enhance the training data; Construct a loss function, expressed as: Where, L tt is the total loss function; is the segmentation probability of the i-th pixel; is the true label of the i-th pixel; γ is the adjustment factor; is a smoothing term; Multi-view feature interaction.

3. The computer vision-based intelligent control system for inkjet printer according to claim 1, characterized in that: The printing anomaly detection model building module is to build an adaptive neural network model, which specifically includes the following contents: Secondary segmentation; the segmentation formula for each input image is expressed as ; Where Q is the number of regions after segmentation, H is the input image height, Ph is the height of the segmentation window, stride is the step size, W is the input image width, and Pw is the width of the segmentation window; Adaptive network; add nodes to each layer and check the reduction of residual. If the residual reduction is less than , then stop modeling this layer and build the next layer; the residual reduction value calculation formula is: ; The residual calculation is expressed as ;in, is the mean residual reduction, and are the residual values ​​of the nth layer network and the residual values ​​of the first ns layers, respectively, and s is the number of nodes added in the current layer; is the residual of the k+1th layer, and are the error vectors of the kth layer and the k+1th layer respectively, is the feature response vector of the k+1th layer; r is the weight factor; Feature mapping calculation: Use nested linear transformations and nonlinear activation functions to build feature representations layer by layer, as shown below: ; In the formula, is the nth layer for input x i2 feature representation; k is the layer index, is the number of nodes; is the weight coefficient; is the sigmoid activation function; is the feature weight vector; is the input vector; is the additional weight vector; is the bias term; is the initial feature representation; Model optimization.

4. The computer vision-based intelligent control system for inkjet printers according to claim 3 is characterized in that: The model optimization is to perform pseudo-inverse solution on the model weights after configuring the nodes, obtain the optimal model configuration, and output the prediction results; the pseudo-inverse solution is expressed as: ;in, is the optimal model weight vector solved; is the weight vector of the model; H is the input feature matrix; T is the target matrix; It is the Moore-Penrose pseudo-inverse; is the square of the Frobenius norm.

5. The computer vision-based intelligent control system for inkjet printer according to claim 1, characterized in that: In the data acquisition module, the historical spray-painted surface image data is enhanced and annotated. The enhanced image annotation categories are foreground and background, which are only used for image enhancement.

6. The computer vision-based intelligent control system for inkjet printers according to claim 1 is characterized in that: The inkjet printer intelligent control module is based on the established inkjet printing anomaly detection model, collects inkjet printing surface image data in real time, and inputs it into the inkjet printing anomaly detection model after image enhancement. If the inkjet printing anomaly detection model outputs abnormal inkjet printing, early warning processing is performed to realize intelligent control of the inkjet printer.

Citation Information

Cited By

  • RPA input method for identifying cultural creative spray drawing product yield

    CN122887394A

  • A method and device for identifying and compensating for spray point offset for a UV plotter

    CN122900385A