Optimizing Automated Optical Inspection: An Adaptive Fusion and Semi-Supervised Self-Learning Approach for Elevated...

Area of Science:

• Computer Vision
• Machine Learning
• Industrial Automation

Background:

• Automatic Optical Inspection (AOI) systems require high object detection accuracy for defect identification.
• Traditional methods often depend on extensive annotated datasets, which are costly and time-consuming to acquire.
• Limited labeled data poses a significant challenge for developing robust AOI models.

Purpose of the Study:

• To develop innovative strategies for enhancing object detection accuracy in AOI.
• To reduce the reliance on large annotated datasets for training defect detection models.
• To propose and evaluate a novel semi-supervised learning method for AOI.

Main Methods:

• Development of a defect detection model using a dataset of 3579 images across 32 categories.
• Implementation of data augmentation and annotation refinement techniques.
• Proposal of the Adaptive Fused Semi-Supervised Self-Learning (AFSL) method, incorporating a Bounding Box Assigner, Adaptive Training Scheduler, and Data Allocator.

Main Results:

• The AFSL method improved mean average precision (mAP) from 43.5% to 57.1% on the COCO dataset.
• A 2.6% improvement in mAP was observed on the specific AOI dataset.
• The method demonstrated enhanced performance on AOI datasets with minimal labeled data.

Conclusions:

• The AFSL method effectively enhances object detection accuracy in AOI.
• Semi-supervised learning approaches, like AFSL, are viable for overcoming data limitations in AOI.
• The proposed method offers a more efficient and precise solution for defect detection in industrial applications.