Automated Corrosion Detection Using YOLOv8s: A Deep Learning Pipeline for Real-Time Industrial Inspection
2026 International Conference on Digital Transformation, Innovation & Sustainable Development (DTISD), ´Adan, Yemen, 20 - 22 Temmuz 2026, ss.1-9, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/dtisd69661.2026.11668310
- Basıldığı Şehir: ´Adan
- Basıldığı Ülke: Yemen
- Sayfa Sayıları: ss.1-9
- Kocaeli Üniversitesi Adresli: Evet
Özet
Corrosion is a critical threat to metallic infrastructure integrity, imposing substantial economic and safety consequences globally. Existing automated detection approaches are constrained by limited generalization, high computational cost, or lack of real-time applicability. This paper presents a reproducible end-to-end deep learning pipeline for automated corrosion detection using the YOLOv8s object detection framework. A dataset of 506 real-world corrosion images was subjected to systematic exploratory data analysis (EDA), revealing characteristic warm chromatic bias consistent with iron oxide signatures. Offline augmentation via the Albumentations library expanded the training corpus from 354 to 1,380 images, enhancing robustness against photometric and geometric variability. A supplementary diagnostic feature extraction stage—employing ResNet-50 embeddings and handcrafted descriptors—characterized dataset structure and informed augmentation design; this stage is explicitly not part of the training pipeline. The YOLOv8s model was trained on an NVIDIA RTX 3070 GPU using the AdamW optimizer with early stopping, converging at epoch 94. Evaluation on the heldout test set yielded a precision of 0.986, recall of 0.985, mAP@0.5 of 0.993, and mAP@0.5:0.95 of 0.981, with real-time inference at approximately 108 FPS. These results demonstrate that a carefully designed data pipeline combined with a lightweight single-stage detector achieves near-perfect corrosion detection performance suitable for practical industrial deployment.