Research on Corrosion Recognition Method of Steel Based on Convolutional Neural Network

Research on Corrosion Recognition Method of Steel Based on Convolutional Neural Network

---

연구 분야: Artificial Intelligence

---

---

학회: 2023 IEEE 6th International Conference on Information Systems and Computer Aided Education (ICISCAE)

---

초록

Steel structure is prone to corrosion even during the normal use. Traditional corrosion identification methods are hampered by subjective judgment and time constraints. Convolutional neural network (CNN) and its variants such as U-Net and residual neural networks (ResNet) show great potential for accurately identifying and segmenting rusty areas in images. In this paper, the application of CNN in the identification and segmentation of corrosion zones in steel structures is studied systematically. Firstly, two case studies are presented, showcasing the effectiveness of CNN in detecting and grading corrosion on diverse objects. In addition, this paper introduces Ensembled CNN (ECNN), which amalgamates multiple CNN models to enhance overall performance and generalization capabilities. The results show that ECNN surpasses individual CNN models, emerging as a promising approach for corrosion identification. The outcomes underscore the feasibility and precision of CNN-based techniques in corrosion recognition, implying their practical applicability in a wide array of scenarios.

---