Segmentation of Ship Propeller Cavitation Area Using Image-Based Anomaly Detection Network

Cavitation is the formation of vapor-filled cavities in the liquid. Cavitation in rotating propeller can cause various engineering problems to ships, such as hull vibration, noise, thrust reduction, and propeller corrosion. To address this cavitation problem in designing ship propellers, it needs to analyze the cavitation phenomenon in real propellers. In this study, we introduce a computer vision method to detect and segment cavitation areas in a rotating propeller. The pro-posed method captures the images of a high-speed rotating propeller in a water tunnel and detect the image area of the cavitation. As the first study in the im-age-based automatic cavitation detection, we employ a deep neural network, an anomaly detection method called DDAD (Anomaly Detection with Conditioned Denoising Diffusion Models). In addition, we also use a STN (Spatial Transformer Networks) network to compensate the image shaking caused by the vibration from the high-speed propeller and water flow. After the STN compensation, normal and abnormal images are automatically aligned to find cavitation areas more accurately. The proposed method is the first to use an anomaly detection network for ship propeller cavitation detection. Experimental results show that the proposed method can automatically detect the cavitation area without any annotation supervision.