Marine ecosystems provide multiple services to humans, including provisioning services, such as seafood or fossil energy; regulating services, like coastal protection or water purification; cultural services, as tourism or spiritual benefits; and supporting services, like nutrient cycling or habitat provision.

The provided services are endangered by negative impacts that marine ecosystems are suffering due to multiple causes, some examples of which could be overfishing, habitat destruction, or plastic pollution. Therefore, there exists an urgency to develop new protective measures. One highlighted initiative is to develop scientifically and statistically robust monitoring methodologies and tools to control potential risks or assess the effectiveness of protective and recovery initiatives.

Ocean research and management is facing a new era, led by the technological developments in data collection, allowing the collection of vast amounts of data; and deep learning techniques, capable of pro- cessing the data and reducing its processing workload while increasing the spatial and temporal scope of conducted analysis. The marine science community is ready and willing to implement these new tools to a wide range of proposals towards the sustainability of marine ecosystems and its services.

The objective of this thesis is to study the applicability of deep learning solutions, along with com- puter vision, to develop new tools to preserve marine ecosystems and the offered services. Tools have been developed for three different tasks: Posidonia oceanica monitoring, jellyfish quantification and pipeline characterisation. In their development, diverse deep convolutional network model types and architectures have been trained and tested with data gathered from a variety of sources and under different environmen- tal conditions. Additionally, the developed tools have been deployed into diverse platforms and adapted to its features and limitations.

These implementations cover a wide spectrum of scenarios where deep convolutional networks have been applied with good results, automating the data analysis process, expanding the temporal and spa- tial scope of the analysis or surveys, and improving the repeatability of experiments to detect evolution trends. Thus, validating the proposed methodology to implement deep convolutional networks for video processing to preserve marine ecosystem services.