FishID

Automated analysis of underwater video footage

FishID is an online platform that automates the analysis of underwater video footage.

Users upload videos to the FishID platform for analysis using an optimised deep learning algorithm. They can inspect results overlaid on videos, and export data in CSV format for data manipulation.

FishID is sophisticated software designed for ease of use in providing actionable data from aquatic ecosystem monitoring. The Global Wetlands Project is committed to informing and empowering coastal conservation using tools such as FishID for projects like Moreton Bay Live

Applications of FishID

Target Species

When select species are of interest, FishID can detect and count the target species while correctly ignoring all other species as well as background structure. See example that remain accurate across very different water clarity and locations.

doi: https://doi.org/10.3389/fmars.2020.00429
doi: https://doi.org/10.1101/2020.05.19.105056

Fish Communities

When monitoring aims to measure the entire suite of fish present, FishID can detect a wide range of species trained for particular locations or habitats.
Community data is produced from monitoring with unbaited cameras, ROVs, and BRUVS (baited remote underwater video stations).

Behaviour

Specific behaviours of fish and other animals can be automatically detected. Animal behaviour is reliably captured on cameras, more so than when divers are present in the water. Examples of behaviour recorded include grazing activity by herbivorous fish. FishID can detect, count, and record duration of behaviours in underwater videos.

Fish Movement

Measurement of small-scale animal movements underwater is difficult using traditional methods. FishID applies the latest tracking algorithms to automate the monitoring and measurement of marine animal movement.

Movement direction and speed are useful for answering questions about animal health, ecological dynamics and other aspects of management and conservation.

Acoustics

Acoustic technologies can provide clear underwater imagery when regular video is limited by low light (turbidity, night time).
FishID removes the bottleneck of manual data analysis.

Other Species

FishID can be trained to identify a very wide range of animals to suit monitoring objectives.

It can also be used for a collection of marine species including reptiles and mammals and even birds.

FishID app

We’ve developed an intuitive user interface that contains all of the tools needed to create and use FishID models:

  • dataset management and version control
  • video uploading tools
  • tools for annotating datasets including bounding-box and segmentation masks
  • suggested annotations using FishID models for a faster annotation workflow
  • multiple users and real-time collaboration
  • model training, validation and evaluation
  • evaluate trained models against test datasets and explore performance results
  • run predictions with trained models to detect, count and monitor fish species in underwater videos

Partners