Image Recognition With Region-Level Defect Reliability
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image recognition systems struggle to accurately determine the correctness of individual defect predictions when multiple defects appear in an image, as they calculate features for the entirety of the image, failing to separately identify and evaluate each defect.
Innovation Solution
The system calculates the importance of features for each target shape and type, comparing this importance with statistical data to determine the correctness of recognition results for each defect, using a convolutional neural network and error backpropagation to assess feature influence.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If features are calculated for the entirety of one input image, then the classification result reliability can be determined, but individual defects cannot be separately identified and evaluated
Solution Approach 1:
The patent divides the image into multiple candidate regions based on prediction results, where each region corresponds to a potential defect. By calculating features and determining reliability separately for each candidate region rather than for the entire image, the system achieves both overall classification reliability and individual defect identification precision simultaneously.
2Device complexity
If an aggregate of multiple defects is identified by a single feature, then processing is simplified, but correctness of prediction results for each individual defect cannot be determined
Solution Approach 1:
The patent segments the defect detection process by creating separate candidate regions for each predicted defect. Each region undergoes independent feature calculation and reliability determination, allowing individual defect evaluation without requiring complex aggregate processing, thus maintaining simplicity while achieving precise individual defect analysis.
Solution Approach 2:
The patent applies local quality by calculating features and determining reliability specifically for each candidate region rather than uniformly across the entire image. This localized approach allows each defect to be evaluated with appropriate attention, improving individual defect prediction correctness without significantly increasing overall processing complexity.
Data Source
AI summary
According to the present invention, an image recognition system calculates importance of a feature for each target shape recognized in an image and for each type of feature, and determines correctness of a recognition result by comparing the importance with a statistic for each type of feature, for each target shape.