Towards Semi Real-Time Environmental Monitoring Using Shadowgraphy

Environmental monitoring for detecting habitat changes as well as for mining impact assessment has become a crucial topic in recent years. This requires large-scale spatial and temporal data collection for analysis. At JAMSTEC, we have been working on the development of shadowgraph-based imaging systems to address this issue. Shadowgraph-based imaging systems consume low amounts of power and are not affected by external lighting conditions, making them ideal for monitoring. Also, since shadowgraphs use collimated light, it is easy to extract particle sizes along with particle counts. We have developed multiple inhouse shadowgraph systems with different viewing areas and resolutions for different monitoring requirements. A data processing pipeline has been made which can analyze the shadowgraph videos post-cruise and analyze plankton distributions and abundances as well as provide other information such as particle distributions, size distributions and detect events such as plumes and mixing layers using pattern matching. Since the shadowgraph videos can contain a large variety of animals, we are also using human-in-the-loop machine learning for training target-specific models based on the target assessment. The system has been installed on various underwater platforms, such as a long-term seafloor lander as well as CTD-rosette systems. We are now developing an extension to the system that can provide near real-time feedback to the host platform based on the survey results. We aim to provide alerts when particular events and observations are detected, which can be customized based on the monitoring targets. In the future, we are planning to integrate these shadowgraphs with other underwater platforms such as AUVs and seafloor stations.