Image Segmentation of Fiducial Marks with Complex Backgrounds Based on the mARU-Net

Area of Science:

• Computer Vision and Image Processing
• Manufacturing Technology
• Electrical Engineering

Background:

• Accurate alignment of printed circuit board (PCB) layers relies on high-precision fiducial mark detection during exposure.
• Complex backgrounds in PCB images pose significant challenges for fiducial mark segmentation and detection accuracy.
• Current detection methods often struggle with the intricate visual data present in PCB manufacturing.

Purpose of the Study:

• To propose an improved image segmentation method, mARU-Net, for fiducial marks with complex backgrounds.
• To enhance the detection accuracy of fiducial marks, thereby improving PCB alignment precision.
• To validate the effectiveness of mARU-Net against existing segmentation techniques in a specialized PCB dataset.

Main Methods:

• Development and implementation of the mARU-Net model for image segmentation of fiducial marks.
• Utilizing a customized dataset of PCB fiducial mark images for training and evaluation.
• Employing the centroid method for circle detection on segmented fiducial mark images.

Main Results:

• The mARU-Net demonstrated superior segmentation accuracy compared to typical methods on customized PCB datasets.
• Segmentation accuracy improved by 3.015% over the original U-Net, with no significant increase in parameters or training time.
• The centroid method achieved high detection efficiency, with deviations within 30 microns, meeting PCB production requirements.

Conclusions:

• The mARU-Net model effectively addresses the challenges of segmenting fiducial marks in complex PCB backgrounds.
• The proposed method significantly enhances fiducial mark detection accuracy, crucial for high-precision PCB manufacturing.
• mARU-Net offers a viable solution for improving the quality and efficiency of PCB production processes.