SURFACE DEFECT DETECTION OF GLASS COVERSLIPS BASED ON DEEP LEARNING: HYBRID CNN TRANSFORMATION METHOD

Read full text Summarise with VIRAT
Authors: Zhenshou Yang, Qunsheng Yi
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Glass substrates, such as coverslips and electronic product glass, are extensively used in laboratory, biomedical, and consumer electronics applications, where surface integrity is essential for ensuring functional reliability and experimental accuracy. Performance can be greatly impacted by even little Surface Defects (SDs), such as stains, cracks, and scratches. In high-throughput settings, traditional manual inspection is subjective, time-consuming, and inefficient at identifying minute flaws. In order to overcome these obstacles, this study suggests a brand-new hybrid Deep Learning (DL) framework for automated glass SD detection and categorization called Attention-Boosted Convolution with Discrete Wavelet Transform (ATT-BC-DWT). The method integrates Convolutional Neural Networks (CNNs) to extract hierarchical spatial features, DWT to enhance multi-resolution defect representation and suppress noise, and an attention mechanism to emphasize defect-relevant regions while reducing background interference. Image normalization and Gaussian filtering are employed to improve contrast consistency and noise robustness. The coordinated merging of spatial, frequency-domain, and attention-based learning, which allows for exact localization and accurate recognition of defects of various sizes and forms, is what makes the proposed approach novel. In comparison to traditional CNN and detection models, experimental assessment utilizing Python-based DL tools shows better performance, with a 99.23% recall, enhanced precision, F1-score, and fewer False Positives (FPs). The results confirm the robustness and effectiveness of the ATT-BC-DWT framework for real-time inspection, supporting seamless integration into automated laboratory and industrial quality control systems.

Fields of Study & Subjects

Industrial Vision Systems and Defect Detection Infrastructure Maintenance and Monitoring Non-Destructive Testing Techniques

Publication details

Journal / source
Surface Review and Letters
Volume
33
Issue
06
Publication date
26 March 2026
Language
EN