Trial lecture - time and place
11th of March, 10:15 AM, Kristen Nygaards sal (5370), Ole-Johan Dahls hus
“Machine learning for classification of acoustic data”
Conferral summary
Tor Inge Lønmo har undersøkt hvordan forbedre kartleggende ekkolodd via bedre signalbehandling. Han har funnet metoder som kan gi et mer nøyaktig bilde av havbunnen og vannmassene over gjennom bedre utnyttelse av eksisterende data.
Main research findings
Accurate knowledge of the seabed is of vital importance for many human endeavors. Applications range from safe navigation to climate change models. Swath sonars are a key tool for ecient and high-resolution mapping of the seabed. This thesis aims to improve the quality of swath sonars by improving the beamformer, a key part of current signal processing. We explore two methods: Adaptive beamforming and autocalibration.
Adaptive beamforming improves the beamforming process by adapting the beamforming to the received signal. We investigate the adaptive Capon and Low Complexity Adaptive (LCA) beamformers on simulated and field data. We find that they improve resolution, edge definition, and sidelobe level in the water column, and also give more accurate amplitude detections. Capon has higher performance, while LCA is faster and more robust. The effect of adaptive beamforming on phase detections is mixed.
Autocalibration estimates calibration errors without external reference sources. We estimate amplitude and phase errors per element using data from normal operation via the Generalized Interferometric Array Response. This greatly lowers sidelobes on simulated data. However, the effect is smaller and less consistent on field data. The difference seems to be caused by an insufficient calibration model. Autocalibration may also further improve the adaptive beamformers.
Contact information to Department: Mozhdeh Sheibani Harat