LAUSR

The detection of bubbles in images of transmitter and receiver printed circuit boards (TR-PCBs) is a difficult task due to the need for precise subregional detection, fast processing speed, and high accuracy. The uneven illumination of the images further complicates the process of bubble segmentation. To address these issues, a TR-PCB bubble segmentation framework, the efficient channel and spatial attention (ECSA)-efficient feature-fused autoencoder (EFA)-dynamic threshold (DT) framework, is proposed, which has two stages. In the first stage, an ECSA module is designed to analyze complex background features by employing the complementary advantages of the channel and spatial position features. Then, an EFA is proposed for region segmentation in TR-PCB images. In the second stage, a DT approach based on maximizing the interclass variance and a sliding window is proposed to eliminate severe shadows and issues with uneven illumination in TR-PCB images. The DT approach segments the bubbles in each region and calculates the bubble rate more accurately than previous methods. The experimental results show that the proposed framework achieves superior performance in several quantitative metrics and is suitable for industrial applications, obtaining an average Dice score of 87.5% and an intersection over union (IoU) score of 77.8% on the region segmentation test set, and a 0.312 misclassification error (ME) value on the bubble segmentation test set.