Menu

AI Algorithms

AI algorithms for complex signals

Ezako develops specialized AI algorithms for detecting, classifying and interpreting complex signals across acoustic, DAS, radio-frequency and time-series data. Our algorithms process large volumes of recordings and pre-label events of interest — from vessel passages and acoustic signatures to biological sources and anomalies. Results are presented to analysts for validation, with every decision recorded and traceable.

Detection and analysis capabilities

  • Signature search

    Search an entire archive for a reference signature, taken from existing labels or defined directly in Upalgo. Results are reviewed on a sample, false detections are excluded, and the validated results are converted into labeling tasks.

  • Vessel passage detection

    Detect and characterize vessel passages, including closest point of approach, speed and confidence. When available, detections can be correlated with AIS data. Sensor position, bathymetry and sound-speed profiles can also be associated with each event.

  • Acoustic source classification

    Identify and classify regions emerging from background noise as biological or anthropogenic sources, including sources and environments not encountered during training.

  • Real-time acoustic detection

    In Upalgo Labeling Sound, detect events directly on the spectrogram during playback and submit them to the analyst for validation.

  • Vessel tracking on fibre

    From vessel signatures recorded along a DAS cable, estimate vessel speed, course and position, with or without AIS data.

  • Similar event retrieval

    Starting from a single labeled event, Upalgo Labeling Timeseries retrieves similar events across the dataset, helping analysts find recurring patterns at scale.

From data to deployed algorithms

  • Custom algorithm development

    When an existing algorithm does not meet the requirement, Ezako develops a specialized model together with the customer's experts — from data preparation and training through validation and deployment in Upalgo.

  • Customer algorithms

    Customer Python scripts can run directly on selected data. Their detections are returned as events for analyst validation and displayed alongside the original signals. Models trained on customer datasets can also be integrated into UpalgoDB.

  • On-premises execution

    All processing runs on the customer's infrastructure, including offline and isolated networks. No data is transmitted to Ezako.

Proven in operational projects

Ezako's algorithms have been developed and evaluated through projects across satellite telemetry, engine testing, underwater sensing and embedded machine learning.

2020Satellite telemetry for CNES, alongside the conventional surveillance methods; ranking of more than 76,000 radio sites for inspection, for ANFR.
2021Abnormal observations in the high-frequency sensor series of engine development tests, with Safran Aircraft Engines.
2022 and 2025Real-time detection of underwater anomalies on autonomous platforms, supported by the DGA.
2023Fault detection by machine learning on a microcontroller, within DeepGreen, the project led by CEA that builds the Aidge platform.

The demonstrations on this site run on public data: hydrophone recordings of NOAA’s SanctSound project, NASA’s C-MAPSS engine simulations (Saxena et al., 2008), and a DAS acquisition in the Trondheimsfjord (Thiem et al., J. Acoust. Soc. Am., 2026, CC BY 4.0).

See what Ezako can do with your data

Describe your signals and your requirements to arrange a demonstration.

Talk to the team