Machine Learning Applied to the Underwater Radar-Encoded Laser System

In this work we demonstrate the use of machine learning to enhance imagery collected by the underwater radar-encoded laser imaging system. Laser-based sensors offer the potential for high-resolution, three-dimensional imaging in the underwater environment. However, these capabilities become degraded in turbid water environments due to scattering. This work presents experimental results applying a denoising autoencoder to imagery collected in our lab test tank. We experiment with both shallow and deep network architectures at a variety of water conditions. The use of machine learning allows us to suppress both backscatter and forward scatter. In particular, by applying the denoising autoencoder we are able to acquire imagery at 6.9 attenuation lengths, representing a 25% improvement over our baseline processing scheme.