About Machine Learning in Echoview
Echoview offers machine learning (ML) capabilities to expedite data processing on echograms.
Echoivew's ML operators utilize inference models that are trained using supervised learning to detect specific hydroacoustic features in your echograms.
Machine learning implementation in Echoview
The Trained model bottom exclusion (experimental) operator creates a bottom exclusion virtual line on your input echogram. This is also available as a Line picking algorithm for editable lines. Both of these utilize the Trained model bitmap (experimental) operator to detect the bottom.
The Trained model bitmap (experimental) operator uses an inference model to generate a 10m wide band commencing from the first sample immediately below the bottom that it detects in your echogram. Samples within this band have true Boolean values in the output bitmap.
Future versions of Echoview will keep expanding on ML capability.
Experimental
These ML operators are an experimental feature to Echoview and can produce varying results.
The accuracy of these operators depends on how closely your input echogram resembles the Machine learning training datasets.
Acknowledgment
The development of the ML in Echoview was facilitated by training data shared by CSIRO Oceans & Atmosphere, Hobart, Australia.
Machine learning operator performance
Model inference is computationally intensive and you can expect quicker results on devices with faster CPUs.
Here is a comparison of the time taken to execute an ML operator on the same dataset but with different hardware.
|
CPU |
10th Gen Intel(R) Core(TM) i9-10900 CPU @ 2.8Ghz (20 CPUs) |
962 seconds |
|
11th Gen Intel(R) Core(TM) i9-11900KF @ 3.50GHz (16 CPUs) |
823 seconds |
|
|
GPU |
NVIDIA Quadro P1000 |
513 seconds |
|
NVIDIA GeForce RTX 3090 |
93 seconds |
GPU acceleration
Echoview uses TensorFlowTM which is optimized for NvidiaTM graphics processing units (GPUs) enabled with the CUDA Deep Neural Network (cuDNN).
Preliminary tests indicate that GPU acceleration offers significant performance benefits with the ML implementation in Echoview (see Machine learning operator performance).
Refer to the Installing the CUDA Deep Neural Network library page for instructions on setting up cuDNN on your system.