CN-122023421-A - Electronic engraving system based on machine learning and engraving method thereof

CN122023421ACN 122023421 ACN122023421 ACN 122023421ACN-122023421-A

Abstract

The invention discloses an electronic engraving system based on machine learning and an engraving method thereof, aiming at overcoming the defects of low efficiency, poor parameter adjustment precision and the like of the traditional electronic engraving test engraving, wherein the system comprises modules of image extraction, net pit analysis, characteristic calculation and the like, the net pit characteristics are identified and parameters are calculated through a random forest classifier, after the net pit comparison module outputs a difference result, a complex engraving selection module accurately attributes difference types, a reference net pit is selected, a layered progressive strategy is adopted to execute second test engraving, the first-level shaping net pit shape and position are adopted, the second-level optimizing depth profile is adopted, the third-level edge and texture modification are carried out, and each level is subjected to real-time feedback adjustment; according to the invention, the trial engraving is optimized in a targeted manner, redundant engraving actions are reduced, the parameter adjustment accuracy and the engraving quality stability are improved, and the trial engraving cost is reduced.

Inventors

FANG PING
FANG WENDA
XU XIANGYANG
FANG HUA
ZHOU BIN
GAO SHAN

Assignees

杭州科迅印刷设备有限公司
杭州正乾机电有限公司

Dates

Publication Date: 20260512
Application Date: 20260414

Claims (10)

1. An electronic engraving system based on machine learning, characterized by comprising: the image extraction module is used for acquiring an image acquired by the net pit observer during test carving as an analysis image; The network hole analysis module is used for extracting the middle ridge line of the image block from the analysis image, dividing the area of the image block by the middle ridge line, analyzing the position relation between the gray scale characteristics of each area and the middle ridge line to obtain gray scale gradient characteristics, extracting the edge contour of the image block through edge detection, calculating contour characteristics, judging whether the gray scale gradient characteristics and the contour characteristics are network holes or not through a random forest classifier, and if yes, positioning the network holes; the feature calculation module is used for calculating coordinates and sizes of the net holes according to the outline features; the overlapping calculation module is used for calculating the overlapping width and the overlapping height of the two net holes according to the coordinates and the sizes of the two adjacent net holes; The net hole comparison module is used for indexing standard net hole image data in a preset net hole library according to the positioned net hole type, respectively comparing the gray gradient characteristics, net hole size, overlapping width and overlapping height of two adjacent net holes of the test carving net hole with the standard net hole image data, and outputting a re-carving instruction if any one of the net holes has difference; The parameter adjusting module is preset with engraving parameters, the engraving parameters comprise a high photoelectric current value, a dark current value and an oscillation current amplitude, and when a complex engraving instruction is received, the engraving parameters are adjusted.
2. The machine learning based electronic engraving system of claim 1, wherein said net hole analysis module includes a gray feature analysis strategy, said gray feature analysis strategy includes extracting a middle ridge line in two directions of an aspect of an image block, dividing the image block into four areas of an upper left, an upper right, a lower left and a lower right with an intersection of said middle ridge line as an origin, said gray feature distribution feature includes an x-direction gradient ratio defined as a rate ratio at which gray scales of pixels in the two areas of the upper and lower left near the middle ridge line in the lateral direction change from the center to the edge, a y-direction gradient ratio defined as a rate ratio at which gray scales of pixels in the two areas of the upper and lower left near the middle ridge line in the longitudinal direction change from the center to the edge, and a radial gradient ratio characterizing a gray scale decay rate on a path extending outward from the origin along a radial angle.
3. The machine learning based electronic engraving system of claim 2, wherein said mesh analysis module includes a contour feature analysis strategy, said contour feature analysis strategy includes calculating compactness, ellipse eccentricity, vertex distance divergence and eccentricity as contour features according to edge contours, said compactness is obtained by dividing square of contour perimeter by inner area of contour, said ellipse eccentricity is obtained by fitting minimum circumscribed ellipse and calculating ratio of long half axis to short half axis, said vertex distance divergence is obtained by calculating corresponding vertex coordinates of contour, said eccentricity is obtained by calculating half length of major axis of ellipse and half length of minor axis of ellipse.
4. The electronic engraving system based on machine learning of claim 3, wherein a random forest model is arranged in the random forest classifier, the random forest model output comprises a net hole center coordinate, an equivalent diameter, a depth estimated value and an adjacent net hole overlapping area proportion, the random forest classifier comprises a feature importance assessment unit, the feature importance assessment unit counts the contribution degree of each input feature to the splitting gain in the random forest model training process, and key feature subsets with the contribution degree higher than a preset threshold value to geometric parameter prediction are dynamically screened out.
5. The machine learning-based electronic engraving system according to any one of claims 1 to 4, further comprising a complex engraving selection module, wherein the complex engraving selection module performs difference attribution according to the comparison difference output by the net pit comparison module in the first time of trial engraving, attribution results comprise a degree of overlapping inconsistency, a gray gradient feature inconsistency or a shape feature inconsistency, and then determines a reference net pit and an engraving position of the second time of trial engraving according to the attribution results, performs the second time of trial engraving through a hierarchical progressive strategy, acquires an analysis image under the current level of trial engraving in real time to extract corresponding feature parameters, performs comparison with standard net pit image data, and determines whether to perform the next level engraving according to the comparison results.
6. The machine learning based electronic engraving system as set forth in claim 5, wherein the complex engraving selection module includes a reference net pit selection submodule, wherein the reference net pit selection submodule includes a step of screening net pits with highest gray gradient characteristics and contour characteristics matched with standard net pits from net pits of the first trial engraving as reference net pits of the second trial engraving when the result is that the overlapping degree is not matched, a step of planning engraving positions of the second trial engraving in adjacent areas according to the coordinate positions of the reference net pits, and a step of randomly selecting one net pit from the net pits of the first trial engraving as a reference net pit when the result is that the gray gradient characteristics or the shape characteristics are not matched, and a step of planning engraving positions of single net pit in peripheral interval areas of the reference net pit.
7. The machine learning based electronic engraving system of claim 6, wherein said hierarchical progressive strategy comprises: Determining carving motion parameters of the second trial carving net holes according to coordinates of the reference net holes or the reference net holes at preset carving angles and carving intervals, and calling preset carving parameters to carve net hole contours of the first level; the second level, according to gray gradient characteristic deviation after the first trial engraving, adjust the high current value and dark current value, take the combination of the adjusted high current value, the adjusted dark current value and the initial preset oscillation current amplitude as the engraving parameters of the second level, and carry on the net pit depth engraving of the second level; And the third level, according to the comparison of the contour features of the net holes engraved by the second level and the contour features of the standard net holes, obtaining deviation quantity adjustment oscillation current amplitude values, combining the high photoelectric current value and the dark current value which are adjusted by the second level with the oscillation current amplitude values which are adjusted by the third level to serve as engraving parameters of the third level, and performing net hole contour engraving of the third level.
8. The electronic engraving system based on machine learning of claim 7, wherein said multiple engraving selection module includes a correction selection sub-module, said correction selection sub-module includes a step of obtaining a contour feature detail of a deviation area to judge a contour deviation type when a comparison result shows that a deviation exists between a contour of a net pit after the first-level engraving is completed and a standard net pit contour, said contour deviation type includes a contour displacement, a contour distortion and a contour defect, a first-level engraving for planning a next trial engraving net pit is selected according to said contour deviation type, and a correction engraving motion parameter instruction is outputted.
9. The electronic engraving system based on machine learning of claim 8, wherein said correction selection sub-module further comprises means for obtaining gray scale feature details of the area of deviation extracted by the mesh analysis module to determine a gray scale deviation type when the comparison result shows that the mesh depth after the second-level engraving is completed deviates from the standard mesh depth, said gray scale deviation type includes a transition unbalance and uneven distribution, and selecting the first-level engraving for planning the next trial engraving mesh according to said gray scale deviation type, and outputting a correction engraving parameter instruction.
10. The electronic engraving method based on machine learning is characterized by comprising the following steps of: Acquiring an image acquired by a net hole observer during test carving as an analysis image; Extracting a middle ridge line of an image block from the analysis image, dividing the area of the image block by the middle ridge line, analyzing the position relation between the gray scale characteristics of each area and the middle ridge line to obtain gray scale gradient characteristics, extracting the edge contour of the image block through edge detection, calculating contour characteristics, judging whether the gray scale gradient characteristics and the contour characteristics are net holes or not through a random forest classifier, and if yes, positioning the net holes; calculating coordinates and sizes of the net holes according to the contour features; calculating the overlapping width and the overlapping height of two net holes according to the coordinates and the sizes of two adjacent net holes; Indexing standard net hole image data in a preset net hole library according to the positioned net hole type, respectively comparing the gray gradient characteristics, net hole size, overlapping width and overlapping height of two adjacent net holes of the pilot carving net hole with the standard net hole image data, and outputting a re-carving instruction if any one of the two net holes has difference; Carving parameters are preset, the carving parameters comprise a high photoelectric current value, a dark current value and an oscillation current amplitude, and when a compound carving instruction is received, the carving parameters are adjusted.

Description

Electronic engraving system based on machine learning and engraving method thereof Technical Field The invention relates to the technical field of printing gravure electronic engraving, in particular to an electronic engraving system based on machine learning and an engraving method thereof. Background The electronic engraving system is a professional printing intaglio processing manufacturing system, the electronic engraving system in the prior art needs repeated trial engraving before engraving operation, formal engraving cannot be started until ideal engraving parameters are obtained, the system cannot monitor engraving holes in real time in the engraving process, the engraving parameters cannot be adjusted in real time, the equipment is in an open-loop state, the automation degree and the working efficiency are low, the processing quality is unstable, reworking is easy to cause, and the intellectualization of the electronic engraving system is an urgent requirement of the intaglio manufacturing industry. The method comprises the steps of identifying and analyzing carving network hole images, automatically adjusting carving parameters, namely realizing an intelligent necessary technical route of an electronic carving system, wherein the aim of identifying the carving network hole images is to accurately obtain the central positions, the sizes of all network holes, the mutual position relation among the network holes and the like, and adjusting the carving parameters on the basis, so that the number of trial carving times is reduced, the carving process is controlled, the basic working principle of electronic carving is to carve on an electroplated copper layer with the surface roughness of Ra0.05, grinding channels and dirty points exist due to uneven reflection of the surface of the copper layer, the carving image obtained by a network hole observer not only has miscellaneous points, but also has irregular morphology of the network holes, and even has the phenomena of missing and incomplete network holes, and the traditional image processing is adopted, such as threshold segmentation and edge detection method, unstable gray level change, difficult accurate setting of threshold value, easy noise interference of edge detection, poor identification effect and low precision; when the traditional machine learning method is used for algorithms such as a support vector machine, the characteristic engineering relies on manual design, the comprehensive characteristics of the network points are difficult to extract effectively, the recognition accuracy is low, and the accurate calculation of the geometric parameters cannot be completed at the same time. After searching, a carving process of a pattern with the publication number of CN107364270A is disclosed, and the process of detecting the depth of a concave cavity after electronic carving is described in the patent, but the detection link still stays at a manual or semi-automatic measurement level, an image recognition module is not integrated, a machine learning model is not adopted to carry out robust analysis on complex network cavity images with serious noise interference, self-adaptive adjustment of carving parameters cannot be supported, and core problems of multiple test carving times, unstable carving quality and the like cannot be solved. In summary, the degree of intellectualization of the electronic engraving system in the prior art is low, and the image recognition method in the prior art is difficult to meet the requirement of intellectualization of the electronic engraving system. Disclosure of Invention Aiming at the defects of the prior art, the invention aims to provide an electronic engraving system based on machine learning and an engraving method thereof, and aims to solve the problems that the traditional trial engraving difference cause judgment is general, the core contradiction of inconsistent overlapping degree, gray gradient characteristics and contour characteristics cannot be accurately distinguished, and the parameter adjustment blindness is high. In order to achieve the above purpose, the present invention provides the following technical solutions: An electronic engraving system based on machine learning, comprising: the image extraction module is used for acquiring an image acquired by the net pit observer during test carving as an analysis image; The network hole analysis module is used for extracting the middle ridge line of the image block from the analysis image, dividing the area of the image block by the middle ridge line, analyzing the position relation between the gray scale characteristics of each area and the middle ridge line to obtain gray scale gradient characteristics, extracting the edge contour of the image block through edge detection, calculating contour characteristics, judging whether the gray scale gradient characteristics and the contour characteristics are network holes or not through a random forest classif