Snapshot Hyperspectral Imaging for Underwater Object Segmentation

Due to increased numbers of offshore structures and subsea cables, there is a high demand for underwater maintenance and monitoring. Common options to meet this demand are sonar mapping and imaging. Sonar mapping provides a reliable way for object detection with a high penetration depth, but it is not suitable for tasks that require a detailed insight within the material composition and color of the object. Imaging can provide in-depth, comprehensive information on material properties and external features. This makes it reasonable to investigate its use for object segmentation. Hyperspectral imaging is a subset of imaging which proved to be more effective for airborne object segmentation compared to RGB imaging. This stems from the fact that hyperspectral imaging contains a higher number of spectral bands, justifying the investigation of its applicability in underwater environments. However, underwater imaging faces major challenges such as a variable data quality which is strongly affected by water turbidity, color distortion and a narrow wavelength transmission window. Most of the prior studies conducted on underwater object segmentation relied on RGB images, such as the work carried out by AAU Energy on object segmentation relying on synthetic data [1]. The applicability of hyperspectral reliant object segmentation underwater is yet to be conclusively defined, however, the promising results obtained in airborne conditions are an encouraging prospect. The contribution of this paper is to investigate the applicability of hyperspectral data for underwater object segmentation. In particular, a segmentation algorithm, evaluated in an artificial environment, was researched.