Genetic Algorithm Threshold Optimization for Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing detection technologies rely on subjective or empirical threshold settings, leading to inaccuracies in detection results due to overkill or missed detections, and existing methods like grid search struggle to find precise thresholds efficiently.
Innovation Solution
A detection method using genetic algorithms to determine non-inferior thresholds based on objective functions of overkill and missed detection rates, optimizing thresholds for multiple detection items to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If subjective or empirical threshold settings are used, then the detection process is simple, but the detection accuracy deteriorates due to overkill or missed detections
Solution Approach 1:
The patent transforms the threshold setting from a static subjective parameter to a dynamic optimized parameter. By using genetic algorithms to iteratively adjust threshold values based on objective functions (overkill rate and missed detection rate), the system finds optimal threshold combinations that minimize detection errors while maintaining operational simplicity.
Solution Approach 2:
The patent implements feedback mechanisms where detection results are continuously evaluated against objective functions. The overkill rate and missed detection rate serve as feedback signals that guide the genetic algorithm to adjust thresholds in subsequent iterations, creating a closed-loop optimization system that improves accuracy through iterative refinement.
2Measurement precision
If traditional grid search methods are used to find optimal thresholds, then the detection accuracy improves, but the computational efficiency deteriorates
Solution Approach 1:
The patent replaces the static grid search approach with a dynamic genetic algorithm optimization process. Instead of exhaustively checking all possible threshold combinations, the genetic algorithm dynamically adapts the search strategy by evolving threshold values through selection, crossover, and mutation operations, converging toward optimal solutions more efficiently.
Solution Approach 2:
The patent substitutes the mechanical grid search method with a bio-inspired genetic algorithm. This replacement transitions from a deterministic brute-force approach to a probabilistic evolutionary approach, where thresholds evolve through simulated natural selection, significantly reducing computational burden while maintaining or improving detection accuracy.
3Ease of operation
If a single threshold is used for all detection items, then the system is simple to operate, but the detection accuracy deteriorates due to varying importance of different defects
Solution Approach 1:
The patent applies local quality by assigning different threshold values to different detection items based on their specific characteristics and importance. Instead of using a uniform threshold, the system optimizes each threshold individually through the genetic algorithm, allowing critical defects to have stricter thresholds while less critical ones have more lenient thresholds, thereby improving overall detection accuracy.
Solution Approach 2:
The patent segments the threshold setting process into multiple independent threshold parameters, one for each detection item. This segmentation allows the system to handle different defect types with appropriate specificity, optimizing each threshold separately while maintaining manageable system complexity through automated optimization.
Data Source
Figure 1~2
Figure 3~4
Figure 5~6
AI summary
Embodiments of the present application provide a detection method, a detection device, and a storage medium. The detection method comprises: obtaining a to-be-detected image; obtaining a plurality of confidence levels corresponding to a plurality of detection items according to the to-be-detected image; and determining a detection result of the to-be-detected image according to the plurality of confidence levels and a plurality of detection thresholds, where the plurality of detection thresholds are corresponding to the plurality of detection items and are a non-inferior solution of objective functions of an overkill rate and a missed detection rate, the overkill rate is the ratio of qualified ones detected as defective, and the missed detection rate is the ratio of defective ones detected as qualified. The technical solution of the present application may improve the accuracy of a detection result.