Steam leakage in industrial plant pipelines under high-temperature and high-pressure conditions leads to serious risks to human safety. Therefore, rapid and accurate detection is essential . Conventional AE sensor-based detection methods are effective for real-time monitoring but become impractical for wide-area coverage because of their complex setups and high costs. Recently, vision-based approaches have emerged as practical solutions that offer wide-area non-contact coverage. However, their detection accuracy often degrades under varying illumination and leak conditions, leading to false positives or missed detections. To overcome these limitations, we propose a micro-leakage detection method using active infrared optical gas imaging with deep learning. The proposed method extracts pixels with high temporal variability from sequential frames, defines candidate leakage regions through image segmentation and histogram transformation, and identifies leakage using an autoencoder-based deep learning model. The effectiveness of the proposed method was validated through experiments conducted on a high-temperature and high-pressure steam leakage testbed.
Park et al. (Sun,) studied this question.