A machine vision-based online detection method and system for food packaging sealing

By improving the MaskR-CNN network and adopting an adaptive threshold update mechanism, the problems of false detection and missed detection in food packaging sealing detection are solved, achieving high-precision online detection and stable operation.

CN122415587APending Publication Date: 2026-07-17WUHAN HUANTAI PACKAGING & PRINTING CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
WUHAN HUANTAI PACKAGING & PRINTING CO LTD
Filing Date
2026-05-27
Publication Date
2026-07-17

AI Technical Summary

Technical Problem

Existing food packaging sealing test methods suffer from problems such as offline detection, unstable edge extraction, insufficient defect identification, and difficulty in adaptive thresholds, resulting in high false positive and false negative rates, making it difficult to meet the online detection needs of continuous production lines.

Method used

An improved MaskR-CNN network is adopted, which replaces some standard convolutions in the deep residual backbone feature extraction network with deformable convolutions, introduces a dual attention module and an edge-aware sub-branch, and combines Dice loss and edge-sensitive weighted cross-entropy loss for training. The decision threshold is adaptively updated when the illumination changes, and gray-scale stretching preprocessing is combined to reduce noise interference.

Benefits of technology

It improves the accuracy and adaptability of food packaging sealing test, reduces the false positive and false negative rates, and ensures the stability and reliability of the online testing system.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122415587A_ABST
    Figure CN122415587A_ABST
Patent Text Reader

Abstract

The present application belongs to the technical field of online monitoring, and particularly relates to a food packaging sealing property online detection method and system based on machine vision, which comprises the following steps: replacing the main part standard convolution with deformable convolution; adding a double-path attention module to the feature pyramid transverse connection, and calculating the channel weight from the standard deviation and the mean value; training by combining the Dice loss and the edge-sensitive weighted cross-entropy loss; using gray stretch for pretreatment; outputting the edge and the main mask by the network, and performing weighted fusion according to the edge pixel proportion to extract the sealing contour and calculate the curvature change rate and the width uniformity; using the initial threshold value when the number of qualified samples does not reach the preset value, and using the threshold value equal to the mean value of the historical qualified samples plus 2 times the standard deviation when the number of qualified samples reaches the preset value; and the product is qualified when both the two indexes are lower than the threshold value. The present application can enhance edge perception and adaptive threshold, and reduce the false detection and missed detection rate.
Need to check novelty before this filing date? Find Prior Art