Machine Learning Training Datasets

Echoview is commencing support for machine learning (ML) capability in some operators.

We used supervised learning to train the inference model for these operators. The echograms comprising the training datasets contain examples of a specific hydroacoustic feature (e.g., the bottom) identified by humans. Echoview then detects these features in your echograms with the resulting inference model.

This page describes the properties of the datasets that we utilized for training. When using Echoview's ML operators, your results will vary based on how closely your data resembles the datasets which trained the operator.

Bottom exclusion

The following is a description of the data used to train the model for the "Bottom exclusion" Feature to detect setting in the Trained model bitmap (experimental) operator, and shared by the Trained model bottom exclusion (experimental) operator and Line pick algorithm.

Dataset details

  • Simrad ES60 and ES70 38 kHz Sv data
  • Pings with schools of fish in close proximity to the bottom
  • Contiguous pings with bottom echoes (i.e., no gaps in the echogram)
  • Horizontal bottom orientation

Acknowledgment

The training data was provided by CSIRO Oceans & Atmosphere, Hobart, Australia.

See also

About Machine Learning in Echoview