A Study on a Method for Detecting Surface Defects in Optical Modules Based on Information Entropy Feature Extraction

Area of Science:

• Optoelectronics
• Image Processing
• Artificial Intelligence Hardware

Background:

• Optical modules are critical for AI computing and digital communications, with rapidly growing demand.
• Current surface defect detection methods struggle with noise interference and low-contrast defects, leading to high false negative rates.
• Accurate and robust defect detection is essential for the mass production of optical modules.

Purpose of the Study:

• To develop a traditional image processing scheme for optical module surface defect detection that enhances accuracy and noise resistance.
• To address the limitations of existing methods in handling environmental noise and detecting low-contrast, minute defects.
• To provide a cost-effective and efficient solution suitable for large-scale mass production.

Main Methods:

• A novel detection scheme (WOMC) integrating information entropy constraints into traditional image processing.
• Noise suppression via entropy-weighted Hough transform during calibration.
• Contour localization using Canny edge detection and local entropy filtering.
• Defect recognition through Hu similarity matching and entropy difference verification.

Main Results:

• Achieved an average improvement of 35.77% in image clarity compared to traditional methods.
• Demonstrated a 2.25-fold increase in detection rate under Gaussian and salt-and-pepper noise.
• Reported high performance metrics: 96.67% accuracy, 97.32% recall, 3.31% false positive rate, and a 96.99% comprehensive performance score.
• Successfully detected low-contrast and minute defects with improved robustness.

Conclusions:

• The proposed WOMC method offers superior performance in optical module surface defect detection, particularly in noisy environments.
• This approach provides a viable, low-computing-cost alternative to deep learning models for mass production.
• The method is highly accurate, reliable, and suitable for the stringent requirements of industrial-scale manufacturing.