Echogram segmentation with pymovies_3d involves 3 steps:
- Fine scale echo-integration (EI) of raw acoustic data. EI cells are typically 3m long by 1m high (3m long Elementary Distance Sampling Units, EDSUs)
- Dimension reduction on small EDSUs batches, to retain only useful spectral (and spatial) information. This step is performed using (modified) Local Linear Embedding (LLE).
- EI cells clustering (Kmeans), based on raw frequency spectra, or LLE components.
- Acoustic noise or seabed samples can be optionnally filtered out, based on the clusters mean frequency spectra
Step 2 is optionnal but highly recommended when processing broadband acoustic data.
Echogram segmentation can be performed using this script.
In Spyder, the IPython console's graphic backend must be set to "Automatic" to allow interacting with graphical devices (Preferences/IPython/Graphics/Graphic backend).
- path_evt: path to a text file with the survey metadata. This file is used to filter out unrelevant acoustic recordings.
- path_config: path to MOVIES3D configuration folder to be used for performing EI.
- path_hac_survey: path to .HAC files to be echo-integrated
Key parameters include:
- ind_decoup: list of the indices of acoustic sequences to be processed (sequences order defined in path_evt).
- indice_freq_max: index of maximum frequency to process in classification