| Autor: |
Shi-Jian Liu, Fatana Jafari, Rui Xu, Jean Bosco Ndungutse, Gilbert Shyirambere |
| Rok vydání: |
2021 |
| Předmět: |
|
| Zdroj: |
Proceedings of the International Conference on Advanced Intelligent Systems and Informatics 2021 ISBN: 9783030897000 |
| DOI: |
10.1007/978-3-030-89701-7_8 |
| Popis: |
This research uses a deep learning algorithm to detect insulator strings automatically. Besides, all available disks (or caps) in the insulator can be successfully detected, which makes the missing cap defect easily perceived by human eyes. To fulfill that, aerial images of power equipment are collected and manually annotated. Segmented insulators are blended into background images to produce synthetic images for data augmentation. The system architecture is fine-tuned based on the mask regions with a convolutional neural network. The mAP of our method is equal to 94%, and 90% for insulator and disk respectively, which outperform the YOLOv4-based approach. The experimental results depict the robustness and accurateness of the proposed approach. |
| Databáze: |
OpenAIRE |
| Externí odkaz: |
|
