CN-121599974-B - Image recognition system for detecting surface defects of zinc cathode plate

CN121599974BCN 121599974 BCN121599974 BCN 121599974BCN-121599974-B

Abstract

The invention discloses an image recognition system for detecting surface defects of a zinc cathode plate, and relates to the technical field of industrial vision, comprising the steps of obtaining the zinc cathode plate, performing clamping and positioning on the zinc cathode plate, and generating plate body posture information and to-be-detected area information; the method comprises the steps of driving a prearranged three-dimensional line laser scanning camera set based on plate posture information and to-be-detected area information, scanning to generate a multi-view three-dimensional original point cloud, constructing a unified point cloud model based on the multi-view three-dimensional original point cloud, obtaining partition template data of a zinc cathode plate, performing posture fitting on the partition template data and the unified point cloud model to generate a template matching posture, dividing the unified point cloud model into a plurality of area point cloud subsets based on the template matching posture, binding a defect judging result set with the zinc cathode plate, generating traceable detection records, performing qualification judgment and outputting a detection report. The invention realizes closed loop binding and traceable output of the detection result and the plate body identification.

Inventors

HE JUN
LI JIE
PENG TIEJUN

Assignees

中际山河科技有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260128

Claims (7)

1. An image recognition system for detecting surface defects of a zinc cathode plate is characterized by comprising, The clamping and positioning module is used for acquiring a zinc cathode plate, performing clamping and positioning on the zinc cathode plate and generating plate body posture information and to-be-detected area information; the original point cloud module is used for driving the prearranged three-dimensional line laser scanning camera set based on the plate posture information and the to-be-detected area information, and scanning the prearranged three-dimensional line laser scanning camera set to generate a multi-view three-dimensional original point cloud; The gesture fitting module is used for constructing a unified point cloud model based on the multi-view three-dimensional original point cloud, obtaining partition template data of the zinc cathode plate, performing gesture fitting on the partition template data and the unified point cloud model to generate a template matching gesture, and dividing the unified point cloud model into a plurality of regional point cloud subsets based on the template matching gesture; The method comprises dividing the unified point cloud model into a plurality of regional point cloud subsets based on template matching gesture, specifically comprising the following steps, Mapping the boundary of the structural partition in the partition template data to a unified point cloud model through a template matching gesture to generate a boundary range set of each structural partition; Partitioning and cutting the unified point cloud model based on boundary range sets of all structural partitions to generate a plurality of regional point cloud subsets, wherein the regional point cloud subsets comprise a zinc stripping opening regional point cloud subset, a plate surface regional point cloud subset, a lifting lug regional point cloud subset, a cross beam regional point cloud subset and an edge covering adhesive tape regional point cloud subset; The judging result module is used for respectively carrying out defect quantitative analysis on the plurality of regional point cloud subsets to generate a defect quantitative feature set, and comparing the defect quantitative feature set with preset judging conditions to generate a defect judging result set; the defect quantification feature set is generated by the following steps, Respectively executing reference geometric fitting on the plurality of regional point cloud subsets to generate reference geometric elements of the plurality of regions; extracting quantization characteristics of reference geometric elements of a plurality of areas to generate a defect quantization characteristic set; the method comprises extracting quantized features of reference geometric elements of multiple regions to generate defect quantized feature set, specifically comprising the following steps, Performing defect connected domain extraction on the point cloud subset of the zinc stripping opening area to generate damage connected information; performing convex clustering on the point cloud subsets of the plate surface area to generate convex cluster information; respectively carrying out contour comparison analysis on the lifting lug area point cloud subsets and the cross beam area point cloud subsets to generate lifting lug contour deviation information and cross beam contour deviation information; performing interruption detection on the point cloud subset of the edge-covering adhesive tape area to generate edge continuity information; Packaging the damage communication information, the bulge cluster information, the lifting lug contour deviation information, the cross beam contour deviation information and the edge continuity information to generate a defect quantification feature set; and the detection report module is used for binding the defect judging result set with the zinc cathode plate, generating a traceable detection record, performing qualification judgment and outputting a detection report.
2. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the generating of plate body posture information and to-be-detected area information comprises the following steps of, A zinc cathode plate is obtained, the zinc cathode plate is placed in a detection station, limiting clamping is applied, and a stable positioning state is output; reading limit position parameters of the zinc cathode plate in a stable positioning state, comparing the limit position parameters with a preset positioning reference threshold value, and outputting plate body posture information; And executing gesture transformation on the plate body gesture information, determining the plate body boundary position, executing window cutting processing on the plate body boundary position, and generating the region information to be detected.
3. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the generating of the multi-view three-dimensional origin cloud comprises the following steps of, Performing alignment calibration on plate attitude information to generate camera alignment calibration parameters, performing window mapping on the region information to be detected to generate scanning window parameters; summarizing the camera alignment calibration parameters and the scanning window parameters to generate scanning control parameters; Driving a prearranged three-dimensional line laser scanning camera set through scanning control parameters, and synchronously scanning the information of the region to be detected to generate multi-view scanning data; and carrying out point cloud reconstruction on the multi-view scanning data to generate three-dimensional point sets of all views, and collecting the three-dimensional point sets of all views to generate multi-view three-dimensional original point clouds.
4. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the construction of the unified point cloud model comprises the following specific steps of, Performing coordinate unification on the multi-view three-dimensional original point cloud to generate a point cloud set with unified coordinates; performing multi-view alignment and splicing fusion on the point cloud set with the unified coordinates to generate a complete plate point cloud; and eliminating noise points and outliers of the complete plate point cloud, and constructing a unified point cloud model.
5. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the generating a template matching gesture comprises the steps of, Obtaining partition template data of the zinc cathode plate, and extracting features of the partition template data to generate template reference features; Extracting plate body reference features from the unified point cloud model, performing feature point registration on the template reference features and the plate body reference features, and outputting an initial matching relationship; and adjusting and aligning the relative gesture of the partition template data and the unified point cloud model by using the initial matching relation to generate a template matching gesture.
6. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the generating of the defect determination result set comprises the following steps, Comparing each region characteristic in the defect quantification characteristic set with a preset judgment condition item by item to generate a plurality of judgment marks; a plurality of judgment marks are collected to generate a defect judgment result set.
7. The image recognition system for detecting surface defects of a zinc cathode plate according to claim 1, wherein the qualification is performed, a detection report is output, the method comprises the following steps, Performing item coding on the defect judging result set to generate judging mark summarizing information; mapping the judging mark summary information into a judging mark sequence, executing integrity check on the judging mark sequence, determining a disqualification conclusion when a rejection type judging mark exists in the judging mark sequence, determining a qualification conclusion when the rejection type judging mark does not exist and the integrity check passes, and determining a rechecking conclusion when the integrity check does not pass; Any one detection conclusion is determined from the unqualified conclusion, the qualified conclusion and the rechecking conclusion, and the detection conclusion and the traceable detection record are combined and sorted, so that a detection report is output.

Description

Image recognition system for detecting surface defects of zinc cathode plate Technical Field The invention relates to the technical field of industrial vision, in particular to an image recognition system for detecting surface defects of a zinc cathode plate. Background In zinc electrolysis production, a zinc cathode plate is used as a deposition carrier and a turnover plate, the geometric integrity and the surface state of the zinc cathode plate directly influence zinc stripping efficiency, cathode stripping quality and impurity entrainment risk of a follow-up casting link, the defect detection of the zinc cathode plate is surrounded, related technologies gradually evolve from early manual visual inspection and gauge sampling inspection to an automatic scheme based on machine vision and on-line detection, and two-dimensional imaging, threshold segmentation, edge detection, texture feature analysis and other methods are used for identifying obvious appearance anomalies such as residual zinc adhesion, scratch, pollution spots and the like. However, the prior art still has the following defects that the zinc surface has weak reflection intensity and texture and local pollution and oxidation exist, two-dimensional imaging is easy to be subjected to illumination drift, specular reflection and shielding interference, so that the defect boundary is unstable, the false leakage detection and the detection coexist, a zinc cathode plate is provided with a zinc stripping opening, a lifting lug, a cross beam, a bag Bian Jiaotiao, a plate surface and other structural areas, and the rough ROI is often used for processing, so that the judging caliber of the same defect in different structural areas is not uniform. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides an image recognition system for detecting the surface defects of the zinc cathode plate, which solves the problem that the judgment caliber of the same defect in different structural areas is not uniform. In order to solve the technical problems, the invention provides the following technical scheme: In a first aspect, the present invention provides an image recognition system for detecting surface defects of a zinc cathode plate, comprising, The clamping and positioning module is used for acquiring a zinc cathode plate, performing clamping and positioning on the zinc cathode plate and generating plate body posture information and to-be-detected area information; the original point cloud module is used for driving the prearranged three-dimensional line laser scanning camera set based on the plate posture information and the to-be-detected area information, and scanning the prearranged three-dimensional line laser scanning camera set to generate a multi-view three-dimensional original point cloud; The gesture fitting module is used for constructing a unified point cloud model based on the multi-view three-dimensional original point cloud, obtaining partition template data of the zinc cathode plate, performing gesture fitting on the partition template data and the unified point cloud model to generate a template matching gesture, and dividing the unified point cloud model into a plurality of regional point cloud subsets based on the template matching gesture; The judging result module is used for respectively carrying out defect quantitative analysis on the plurality of regional point cloud subsets to generate a defect quantitative feature set, and comparing the defect quantitative feature set with preset judging conditions to generate a defect judging result set; and the detection report module is used for binding the defect judging result set with the zinc cathode plate, generating a traceable detection record, performing qualification judgment and outputting a detection report. As an optimal scheme of the image recognition system for detecting the surface defects of the zinc cathode plate, the invention generates plate body posture information and to-be-detected area information, and comprises the following specific steps of, A zinc cathode plate is obtained, the zinc cathode plate is placed in a detection station, limiting clamping is applied, and a stable positioning state is output; reading limit position parameters of the zinc cathode plate in a stable positioning state, comparing the limit position parameters with a preset positioning reference threshold value, and outputting plate body posture information; And executing gesture transformation on the plate body gesture information, determining the plate body boundary position, executing window cutting processing on the plate body boundary position, and generating the region information to be detected. As a preferable scheme of the image recognition system for detecting the surface defects of the zinc cathode plate, the method for generating the multi-view three-dimensional origin point cloud compr