Marine Survey, Acoustics and Sonar Systems (250587) – Course 2024/25 PDF
Contents
The student will become familiar with the underwater soundscape and some of the technical terms that will be used throughout the course. Introduction to the acoustic data acquisition chain and the way to analyze and describe sounds in time and frequency dimensions Programming foundation to be able to work with Python and to put all students at the same minimal level. Detailed overview of how to measure sound, the EU guidelines for coastal acoustic monitoring, and learning how to interpret Wenz curves and similar graphics. Gaining knowledge to use the most fundamental Python modules that will be used during the course. A brief description of underwater sound propagation with direct applications to noise pollution. All the Python knowledge that has been gained will be applied here to process wave files, correctly calculate various acoustic statistics, plot PSD curves, etc.
Specific Objectives
Understanding the underwater soundscape and the environmental consequences of sound pollution. Understanding of sound acquisition Gaining a basic programming foundation in Python Becoming familiar with sound measurement procedures and terms. Process a wave file Being able to apply the basic sonar equation and estimating transmission ranges Being able to process acoustic data
Dedication
30h Large group + 42h Self StudyIntroduction to machine learning concepts and if possible connect to what students have already learned during previous courses. Python exercises to become familiar with machine learning toolboxes (in particular under SciPy) Understanding how to evaluate model performance through the different common measures, FP/TP rates, ROC, AUC, precision/recall, average precision, etc. Overview of regression modelling as a building of training neural networks. Introduction to ensemble learning and the random forest classifier. This is already a very powerful tool in machine learning and classification. Becoming familiar with the tensorflow toolbox; the focus will be on the use of the CPU to train and execute models, not the GPU. Overview of perceptron classifiers in order to be able to move to DNN Learning to make use of pretrained models that are available on-line and adapting them as needed. Learning to use data augmentation to improve the quality/versatility of the data set. Overview of the building blocks of a CNN and related architectures. The students will receive audio files containing different whale or doplhin species and will try to classify them through the techniques learned throughout the course.
Dedication
22h Laboratory classes + 30h 48m Self StudyHidrophone communications
Dedication
4h Laboratory classes + 5h 36m Self StudyDedication
4h Laboratory classes + 5h 36m Self Study