Attention Model for Display Panel Defect Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current defect detection methods for display panels, relying on manual inspection or CNN models, are inefficient and prone to misjudgment, leading to high labor costs and potential production delays due to the inability to accurately recognize defects in the manufacturing process.
Innovation Solution
An image recognition method utilizing a first attention model to learn the weight proportion of defect patterns in display panel images, integrating pixel and positional features through dimensionality reduction and attention mechanisms, enabling accurate defect type and position identification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual inspection method is used to detect defects in display panel, then operator can visually observe and recognize defects based on professional experience, but significant labor costs are required and detection efficiency is low
Solution Approach 1:
The patent replaces the manual mechanical inspection process with an automated image recognition system using deep learning models. The system captures images of display panels and uses trained neural networks to automatically detect and classify defects, eliminating the need for manual visual inspection while maintaining high accuracy and significantly improving detection efficiency.
Solution Approach 2:
The image recognition system enables the display panel inspection process to be self-servicing by automatically detecting, classifying, and reporting defects without human intervention. The system processes images, identifies defect patterns, and generates inspection reports autonomously, freeing operators from manual inspection tasks while maintaining consistent quality standards.
2Reliability
If manual inspection method is used to detect defects in display panel, then defects can be recognized based on operator experience, but misjudgment and oversight may occur causing risky defects to be undetected
Solution Approach 1:
The system implements feedback mechanisms where defect detection results are continuously refined through training data from actual production lines. The model learns from historical defect data and inspection outcomes, progressively improving its ability to distinguish between normal variations and actual defects, thereby reducing misjudgments and improving both reliability and precision over time.
Solution Approach 2:
The patent introduces an intermediate image processing stage that captures and analyzes visual information before final defect classification. The system uses image capture devices to obtain panel images, processes them through multiple neural network layers that extract features and identify patterns, and only then makes defect determination. This intermediary processing reduces direct human error while maintaining high detection accuracy.
3Productivity
If CNN model is used to recognize defects in display panel image, then automated detection is achieved, but the model may not accurately capture global context information leading to reduced recognition accuracy
Solution Approach 1:
The patent transitions from traditional two-dimensional CNN processing to three-dimensional spatiotemporal processing by incorporating temporal dimensions into the network architecture. This allows the model to analyze not only spatial features of defects but also temporal patterns across multiple frames or inspection stages, capturing global context information that improves recognition accuracy while maintaining automated efficiency.
Solution Approach 2:
The system segments the defect recognition process into multiple specialized processing stages: image capture, preprocessing, feature extraction, defect detection, and classification. Each stage is handled by dedicated network components that focus on specific aspects of analysis, allowing the system to process global context information effectively while maintaining high detection efficiency through parallel processing.
Data Source
AI summary
The present disclosure relates to an image recognition method and system for a display panel, a training method, and an electronic device and a non-volatile computer-readable storage medium. The image recognition method includes: acquiring an image of a display panel, wherein the image includes gate lines extending in a first direction and data lines extending in a second direction, the gate lines and the data lines intersecting to define a plurality of sub-pixel regions, and the image further includes a defect pattern; and recognizing the defect pattern in the image by using an image recognition model to obtain defect information, wherein the defect information includes at least one of a defect type or a defect position of the defect pattern, the image recognition model comprises a first attention model configured to learn a weight proportion of a feature of the defect pattern in the image.