Efficient Pretraining Model based on Multi-Scale Local Visual Field Feature Reconstruction for PCB CT Image Element Segmentation

TL;DR

This work tackles PCB CT image element segmentation using self-supervised pretraining. It introduces EMLR-seg, a teacher-guided Mask Image Modeling framework augmented with a multi-scale local visual field extraction (MVE) module, followed by a streamlined 4-block decoder and UperNet-based finetuning. The approach yields 88.6% mIoU on a PCB CT dataset, with substantial training-time reductions (≈29.6 hours) compared with strong baselines, and ablations demonstrate the effectiveness of local-field focus, masking strategy, and modest decoder depth. This method offers a practical path to high-accuracy, data-efficient PCB defect segmentation in nondestructive testing settings, with implications for faster model convergence and robustness to image noise and artifacts.

Abstract

Element segmentation is a key step in nondestructive testing of Printed Circuit Boards (PCB) based on Computed Tomography (CT) technology. In recent years, the rapid development of self-supervised pretraining technology can obtain general image features without labeled samples, and then use a small amount of labeled samples to solve downstream tasks, which has a good potential in PCB element segmentation. At present, Masked Image Modeling (MIM) pretraining model has been initially applied in PCB CT image element segmentation. However, due to the small and regular size of PCB elements such as vias, wires, and pads, the global visual field has redundancy for a single element reconstruction, which may damage the performance of the model. Based on this issue, we propose an efficient pretraining model based on multi-scale local visual field feature reconstruction for PCB CT image element segmentation (EMLR-seg). In this model, the teacher-guided MIM pretraining model is introduced into PCB CT image element segmentation for the first time, and a multi-scale local visual field extraction (MVE) module is proposed to reduce redundancy by focusing on local visual fields. At the same time, a simple 4-Transformer-blocks decoder is used. Experiments show that EMLR-seg can achieve 88.6% mIoU on the PCB CT image dataset we proposed, which exceeds 1.2% by the baseline model, and the training time is reduced by 29.6 hours, a reduction of 17.4% under the same experimental condition, which reflects the advantage of EMLR-seg in terms of performance and efficiency.

Figures (7)

• Figure 1: Our EMLR-seg architecture. The student network consists of an encoder, a MVE module and a decoder. The teacher network consists of a MVE module and an encoder.
• Figure 2: Multi-scale local visual field extraction module in the student network. Firstly, the encoder's output and the mask tokens are combined and arranged in the order of the original image. Then the local visual field features are extracted respectively as reconstruction units.
• Figure 3: The finetuning model. Where B, C, H, W respectively represent the batch size, the channel, image height and width.
• Figure 4: Examples of PCB CT images and labels. The blue pixels are vias, the green ones are pads, and the purple ones are wires.
• Figure 5: Visualization results of element segmentation. The three columns are the original image, dBOT and EMLR-seg segmentation results in turn.