Federico Barbiero
Convolutional Neural Network and Source Separation for bio-signals recognition and classification.
Rel. Gabriella Olmo. Politecnico di Torino, Corso di laurea magistrale in Ingegneria Matematica, 2020
|
PDF (Tesi_di_laurea)
- Tesi
Licenza: Creative Commons Attribution Non-commercial No Derivatives. Download (4MB) | Preview |
Abstract: |
Vivent is a Swiss-based company working on bio-signals and it developed an autonomous multichannel recorder to detect bio-responses of plant. HEIA-FR currently working with them on the development of an autonomous multichannel recorder as well as on determining plants' status by using supervised machine learning. It is proved that electrical signals are fundamental in order to regulate physiological processes such as growth, gas exchange, respiration, transpiration and that bio-electrical activity is modified in response to stress conditions or biological cycles. The idea is to determine the plants' status by inspecting bio-electrical activity. The idea is to use the high performances CNN can reach on image classifications in order to classify signals. At first to enhance classification performances and to better understand the available data a study on source separation has been made. Independent Component Analysis is a statistical tool that allows to separate sources from an observed mixture. Collected data have been processed in order to extract the essential structure of the data which should correspond to some physical causes that were involved in the process. Later in order to classify signals have been converted into images. The ratio is that CNNs (convolutional neural networks) work really well on images and, once trained, are really quick to compute outputs. Two different imaging techniques have been exploited: one using signals spectrograms and one combining Gramian Angular Fields and at Markov Transition Field. Images have been distinguish among stressed and relaxed moments and labelled. At the end a CNN model has been trained and tested and classification performances have been measured in order to evaluate source separation effectiveness. |
---|---|
Relators: | Gabriella Olmo |
Academic year: | 2019/20 |
Publication type: | Electronic |
Number of Pages: | 73 |
Subjects: | |
Corso di laurea: | Corso di laurea magistrale in Ingegneria Matematica |
Classe di laurea: | New organization > Master science > LM-44 - MATHEMATICAL MODELLING FOR ENGINEERING |
Aziende collaboratrici: | UNSPECIFIED |
URI: | http://webthesis.biblio.polito.it/id/eprint/13656 |
Modify record (reserved for operators) |