Although modern machine learning has the potential to greatly speed up the interpretation of imagery, the varied nature of the seabed and limited availability of expert annotations form barriers to its widespread use in seafloor mapping applications. This motivates research into unsupervised methods that function without large databases of human annotations. This paper develops an unsupervised feature learning method for georeferenced seafloor visual imagery that considers patterns both within the footprint of a single image frame and broader scale spatial characteristics. Features within images are learnt using an autoencoder developed based on the AlexNet deep convolutional neural network. Features larger than each image frame are learnt using a novel loss function that regularises autoencoder training using the Kullback–Leibler divergence function to loosely assume that images captured within a close distance of each other look more similar than those that are far away. The method is used to semantically interpret images taken by an autonomous underwater vehicle at the Southern Hydrates Ridge, an active gas hydrate field and site of a seafloor cabled observatory at a depth of 780 m. The method's performance when applied to clustering and content‐based image retrieval is assessed against a ground truth consisting of more than 18,000 human annotations. The study shows that the location based loss function increases the rate of information retrieval by a factor of two for seafloor mapping applications. The effects of physics‐based colour correction and image rescaling are also investigated, showing that the improved consistency of spatial information achieved by rescaling is beneficial for recognising artificial objects such as cables and infrastructures, but is less effective for natural objects that have greater dimensional variability.

Manganese crusts (Mn-crusts) are a type of mineral deposit that exists on the surface of seamounts and guyots at depths of >800 m. We have developed a method to efficiently map their distribution using data collected by autonomous underwater vehicles and remotely operated vehicles. Volumetric measurements of Mn-crusts are made using a high-frequency subsurface sonar and a 3-D visual mapping instrument mounted on these vehicles. We developed an algorithm to estimate Mn-crust distribution by combining continuous subsurface thickness measurements with the exposed surface area identified in 3-D maps. This is applied to data collected from three expeditions at Takuyo Daigo seamount at depths of ∼1400 m. The transects add to ∼11 km in length with 12 510 m2 mapped. The results show that 52% of the surveyed area is covered by Mn-crusts with a mean thickness of 69.6 mm. The mean Mn-crust occurrence is 69.6 kg/m2 with a maximum of 204 kg/m2 in the mapped region. The results are consistent with estimates made from samples retrieved from the area, showing more detailed distribution patterns and having significantly lower uncertainty bounds for regional-scale Mn-crust inventory estimation.

We describe a novel semi-supervised learning method that reduces the labelling effort needed to train convolutional neural networks (CNNs) when processing georeferenced imagery. This allows deep learning CNNs to be trained on a per-dataset basis, which is useful in domains where there is limited learning transferability across datasets. The method identifies representative subsets of images from an unlabelled dataset based on the latent representation of a location guided autoencoder. We assess the method's sensitivities to design options using four different ground-truthed datasets of georeferenced environmental monitoring images, where these include various scenes in aerial and seafloor imagery. Efficiency gains are achieved for all the aerial and seafloor image datasets analysed in our experiments, demonstrating the benefit of the method across application domains. Compared to CNNs of the same architecture trained using conventional transfer and active learning, the method achieves equivalent accuracy with an order of magnitude fewer annotations, and 85 % of the accuracy of CNNs trained conventionally with approximately 10,000 human annotations using just 40 prioritised annotations. The biggest gains in efficiency are seen in datasets with unbalanced class distributions and rare classes that have a relatively small number of observations.

Camera equipped Autonomous Underwater Vehicles (AUVs) are now routinely used in seafloor surveys. Obtaining effective representations from the images they collect can enable perception-aware robotic exploration such as information-gain-guided path planning and target-driven visual navigation. This letter develops a novel self-supervised representation learning method for seafloor images collected by AUVs. The method allows deep-learning convolutional autoencoders to leverage multiple sources of metadata to regularise their learning, prioritising features observed in images that can be correlated with patterns in their metadata. The impact of the proposed regularisation is examined on a dataset consisting of more than 30 k colour seafloor images gathered by an AUV off the coast of Tasmania. The metadata used to regularise learning in this dataset consists of the horizontal location and depth of the observed seafloor. The results show that including metadata in self-supervised representation learning can increase image classification accuracy by up to 15% and never degrades learning performance. We show how effective representation learning can be applied to achieve class balanced representative image identification for summarised understanding of imbalanced class distributions in an unsupervised way.

Current estimations of seabed methane release into the ocean (0.4–48 Tg yr⁻¹) are based on short‐term observations and implicitly assume that fluxes are constant over time. However, the intensity of gas seepage varies significantly throughout a seep lifetime. We used instruments operated by the Ocean Observatories Initiative's Regional Cabled Array to monitor variations of gas emissions over the entire Southern Hydrate Ridge summit. We show that bubble plumes emanate from distinct and persistent vents. Multiple plumes can occur within each vent and the location of their outlets may shift progressively. Active bubble plumes vary temporally in number and intensity, even within single vents. Gas emission fluctuations are partly periodic and linked to the local tide. However, short‐term variability and high ebullition events unrelated to tidal cycles are also commonly observed. Our data indicate that small‐scale processes beneath or at the sediment surface are responsible for the short‐term variability of the venting activity that is otherwise modulated by tides. Furthermore, a decrease of venting at one vent may coincide with an increase in plume activity at other vents. Our results depict a spatially and temporally dynamic seep environment, the variability of which cannot be fully characterized without systematic and comprehensive monitoring of the entire area. These results indicate that flux estimations may be largely overestimated or underestimated depending on the time, duration, and place of observation. Although sudden ebullition bursts are hardly predictable, we argue that tidal cycles must be taken into consideration when estimating gas fluxes.

Measuring the distribution, characteristics and dynamics of marine microscale plankton and other particulate matter is essential to understand the vertical flux of elements in the marine environment. Digital holographic microscopy is a powerful approach for measuring these and studying their 3-D trajectories in a relatively large observation volume. This article demonstrates a compact, in-line digital holographic microscope that allows large-volume and high-resolution recording of marine particles through combining a continuous wave laser and a short exposure CMOS camera with efficient global shutters. A resolution of better than 10 μm is demonstrated in air and the minimum distinguishable size of targets recorded in water is approximately 20 μm. The maximum volumetric throughput of the setup is 1904 mL/s. The microscope can take motion blur free holograms of particles moving at up to 490 mm/s in theory, and has been tested in the ∼200-mm/s flowing water. The orientation of the measured volume improves the ability of digital holography in profiling sinking rates and active vertical migration. The system was tested onboard a research vessel to record a range of live plankton and other particles. The motion of some samples, including the sinking motion and swimming motion, was analyzed using custom developed image processing software. The experimental results show that the combination of high resolution and a large volume over which motion of sparse-distribution particles can be tracked, can improve the ability to differentiate between different types of marine particle and identify behaviors of live plankton.