Skip to content
Open access

Deep learning based edge segmentation and classification for plasterboard defect detection

Oct 2026 · Scientific Reports · Vol 16 · 0 citations · 60 references

Abstract

This study presents a two-stage framework for binary plasterboard quality inspection that combines edge-region-of-interest (ROI) segmentation with lightweight classification. A production-line dataset of 5,557 images was acquired using an operator-triggered single-board-computer imaging system. On the 916-image manually masked subset, TransAttUNet produced higher ROI-overlap and boundary metrics than U-Net. The proposed MobileNetV2-MobileViT classifier contains 3.91 million parameters and was evaluated on a held-out set of 838 images, achieving 97.37% overall accuracy and 96.99% balanced accuracy. It achieved the highest accuracy among the evaluated transfer-learning baselines under the shared test protocol. Raspberry Pi 4 was used for image acquisition only; on-device inference, latency, memory, and power were not evaluated. Accordingly, the system should be interpreted as a compact, deployment-oriented framework that requires device-level validation before claims of real-time operation can be made.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.