Appearance Inspection Using One-Class Learning and Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing appearance inspection methods for objects like tires lack the accuracy needed to detect defects effectively, leading to potential defects being overlooked.
Innovation Solution
The method employs a combination of one-class learning and object detection algorithms to sort images of inspected objects as either non-defective or defective, with features of defects being extracted and compared to improve detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional appearance inspection methods are used, then the inspection process is simple, but the defect detection accuracy is insufficient
Solution Approach 1:
The inspection process is divided into two sequential stages: first sorting (using one-class learning to identify potential defects) and second sorting (using object detection to confirm and classify defects). This segmentation allows each stage to focus on specific tasks, improving overall detection accuracy while maintaining manageable system complexity through modular algorithm deployment
Solution Approach 2:
The first sorting step using one-class learning performs preliminary filtering by identifying images that deviate from the normal class. This preliminary action reduces the number of images requiring detailed defect analysis in the second sorting step, thereby improving overall inspection accuracy while optimizing computational resources
2Measurement precision
If a single sorting method is used, then the system is simpler to operate, but the defect classification accuracy is limited
Solution Approach 1:
The system merges two different machine learning approaches (one-class learning and object detection) into a unified dual-sorting inspection system. This combination leverages the strengths of both methods: one-class learning for efficient anomaly detection and object detection for precise defect classification, thereby achieving high classification accuracy while maintaining operational simplicity through automated sequential processing
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An appearance inspection method and the like capable of detecting an appearance defect with high accuracy are provided. The appearance inspection method includes: an imaging step S1 of imaging a plurality of to-be-inspected objects and acquiring images thereof; a first sorting step S2 of sorting the images as non-defective product estimation images within a predetermined standard and defective product estimation images outside the standard by using an algorithm of one class learning; and a second sorting step S3 of detecting a defective product from the to-be-inspected objects by extracting features of defects from the defective product estimation images using an algorithm of object detection, and comparing the images with the features.