2D X-Ray Solder Joint Segmentation Based on K-Means Clustering and Deep Learning

EasyChair Preprint 10226

Abstract

Identifying defective solder joints is crucial in printed circuit board assembly manufacturing, but doing so accurately can be challenging. Image segmentation techniques like the threshold method may not accurately identify defective joints in X-ray images of solder joints. This study aimed to identify a simple segmentation technique used for X-ray images exported from the automatic X-ray inspection machine (AXI) without compromising the accuracy of the results. The proposed approach combines the threshold method and k-means clustering to segment individual pins of the solder joint. We then use the four kinds of padding shade of segmented images from our proposal to train on the YOLOv7, a novel object-detected model. When testing with a test set on X-ray images obtained from identified defective board's solder joint, the model best performs on a replicated border with the gray padding training set.

Keyphrases: Automated X-ray inspection, K-means clustering, Solder joint segmentation

@booklet{EasyChair:10226,
  author    = {Chukiat Boonkorkoer and Phayung Meesad and Maleerat Maliyaem},
  title     = {2D X-Ray Solder Joint Segmentation Based on K-Means Clustering and Deep Learning},
  howpublished = {EasyChair Preprint 10226},
  year      = {EasyChair, 2023}}