Inquiry about Anomaly Detection in Electrical Infrastructures and Thermal Imaging

[angeelalg]

Inquiry about Anomaly Detection in Electrical Infrastructures and Thermal Imaging

Hello,

I've been going through the "Awesome Industrial Anomaly Detection" repository and I'm impressed with the comprehensive collection of methods and research presented here.

I'm currently researching anomaly detection in electrical infrastructures and also exploring the use of thermal imaging cameras for anomaly detection. I was wondering if anyone is aware of previous works or state-of-the-art methods specifically tailored for:

1. Anomaly detection in electrical infrastructures.
2. Anomaly detection using thermal imaging cameras.

Furthermore, I understand that the methods proposed for RGB cameras in the state of the art might be applicable to thermal images. Additionally, multimodal methods could be use combining visual thermal images with complete temperature matrices.

Any pointers, references, or insights would be greatly appreciated. If there are datasets or benchmarks related to these topics, that would be especially helpful.

Thank you in advance for your assistance and for maintaining this valuable resource!

Best regards, Ángel.

[shirowalker]

Hello Angle, I apologize for not noticing your message earlier. As far as I know, there are not many datasets specifically focused on power infrastructure. You might find what you're looking for in MIAD (https://miad-2022.github.io/). There's also limited research on anomaly detection in thermal images. You might be interested in the following papers: 'Thermal anomaly detection in walls via CNN-based segmentation' (https://www.sciencedirect.com/science/article/pii/S0926580521000789) and 'Anomaly detection in thermal images using deep neural networks' (https://ieeexplore.ieee.org/abstract/document/8296692).

Regarding applying RGB-based methods to thermal images, some methods that don't require pretrained models, such as reconstruction-based approaches, could be referenced. If you wish to use other methods that rely on pretrained models, I suggest using prompts or fine-tuning the entire model based on the existing RGB pretrained model. If you have any further questions, feel free to contact us for discussion again.