Automatizing the large-scale analysis of underwater optical data

The use of autonomous data-driven robotics for acquiring underwater data is making large-scale underwater imaging increasingly popular, and tools to efficiently process and understand demographic changes and spatial dynamics of coral reef communities are strongly needed. If we consider that even traditional acquisition techniques have resulted in the creation of thousands of orthorectified imagery, each capturing hundreds to thousands of coral colonies, we can easily forecast that handling such streams of acquired data is hard to be sustained. Further, the existing manual workflows used to generate highly accurate and precise segmentation for fine-scale colony mapping are time-intensive(~1 hour per m2), creating substantial bottlenecks to downstream analyses. While fully automated semantic segmentation can significantly reduce the amount of processing time, current solutions do not yet meet the level of accuracy attained by expert human operators. Further, to capture colony-specific growth, shrinkage, and death, there is a need to create automated and semi-automated tools to track individual corals through time. Here, we present our experience of using TagLab, a semi-automatic tool for the fast annotation of benthic ortho-mosaics. TagLab integrates an agnostic segmentation tool and a fully automatic semantic segmentation tool, both based on CNNs. Following a human-in-the-loop paradigm, this software provides several instruments for editing the per-pixel predictions, allowing users to attain a high level of accuracy not achievable through standard machine learning methods alone. TagLab automatizes the extraction of demographic information and supports the comparison of multi-temporal surveys offering tools for automatic/supervised tracking of colonies and calculation of growth and shrinkage, as well as mortality and recruitment rates. Evaluation of the efficiency of TagLab revealed reductions of 20 to 50% in annotation time, in comparison to manually derived annotations. Importantly, enabling modification of computer-generated matches allows for explicit consideration of fusion-fission dynamics and tracking of ramet and genet identity. By reducing the time required for ecological post-processing of coral reef imagery, TagLab, will enable researchers to process increasingly large volumes of data without increasing the needed person time, and ultimately facilitate a greater capacity to understand and predict future changes to coral reef ecosystems