CSR-YOLO: A Deep Learning Approach for Optical Lens Defect Detection in Microscopic Vision
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 5010208-5010208
DOI: 10.1109/tim.2026.3693457 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Xiande Bao und 2 weiteren Autor:innen. Die Arbeit nennt 25 Literaturangaben. Gefördert von: Scientific Research Innovation Capability Support Project for Young Faculty, Natural Science Foundation of Guangdong Province.
Abstract (Originalsprache)
Accurate and efficient measurement of optical lens surface defects is challenging due to their subtle appearance, irregular morphology, and scale variability. This article presents CSR-YOLO, a lightweight detection framework designed for optical lens surface defect inspection in industrial measurement systems. The framework employs a channel–spatial self-correction attention fusion block (CSSCAFusionBlock), which integrates channel attention, spatial attention, and a global context-based self-correction mechanism to adaptively recalibrate defect features and improve measurement robustness across multiple scales. A lightweight Shuffle Upsample (ShuffleUp) module is further incorporated, combining depthwise convolution, channel shuffle, and pointwise convolution to efficiently enhance feature resolution while preserving fine-grained defect details with low computational overhead. In addition, the WIoU loss function is utilized to refine bounding box regression by dynamically emphasizing hard samples, thereby improving localization accuracy for irregular defects. Experimental results demonstrate that CSR-YOLO achieves reliable inspection performance on an optical lens surface defect dataset, with an mAP50 of 0.924, an mAP50-95 of 0.677, and a processing speed of 163 FPS, satisfying the requirements of real-time industrial optical lens surface inspection and measurement.
Förderung
- Scientific Research Innovation Capability Support Project for Young Faculty
- Natural Science Foundation of Guangdong Province
Zitieren
APA
Bao, X., Takamasu, K., & Chen, M. (2026). CSR-YOLO: A Deep Learning Approach for Optical Lens Defect Detection in Microscopic Vision. IEEE Transactions on Instrumentation and Measurement, 75, 5010208-5010208. https://doi.org/10.1109/tim.2026.3693457
Vancouver
Bao X, Takamasu K, Chen M. CSR-YOLO: A Deep Learning Approach for Optical Lens Defect Detection in Microscopic Vision. IEEE Transactions on Instrumentation and Measurement. 2026;75:5010208-5010208. doi:10.1109/tim.2026.3693457
BibTeX
@article{Bao2026,
author = {Bao, Xiande and Takamasu, Kiyoshi and Chen, Meiyun},
title = {CSR-YOLO: A Deep Learning Approach for Optical Lens Defect Detection in Microscopic Vision},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {5010208--5010208},
doi = {10.1109/tim.2026.3693457}
}
EUVIMED