In 2017, three PhD students from CURES completed their thesis work and presented the results to a grand jury.They are Joseph Oyobe Okassa, Guy Merlin Ngounou and Colince Welba.
1) Joseph Oyobe Okassa's PhD thesis on the « Contribution to the enhancement of compressor algorithms of electromyographic signals by wavelet»is a contribution to the compression of surface electromyographic signals using orthogonal transformations and particularly wavelets.
Joseph proceeded in several steps:
- Presentation of the electromyographic signals;
- Study of the directional transformations which allowed us to retain wavelets as main working tool;
- Implementation of algorithms and multiresolution analysis with Stéphane Mallat's pyramid;
- Modification and implementation of modified versions of these algorithms and comparisons based on quantitative and qualitative compression criteria.
This research work carried out under the supervision of Prof. Pierre Elé, was presented before the grand jury on June 23, 2017 and resulted in five scientific publications.
2) The PhD student of CURES, Guy Merlin Ngounou, has been working since 2011 on the "Contribution to the design and realization of a screening equipment for neonatal deafness". He completed his scientific work in 2016 and presented his PhD results in front of the grand jury on September 12, 2017
3) In his PhD thesis work, Colince Welba was interested in "Exploitation of vector quantization and the combination of several transforms for the compression of ElectroMyoGraphic (EMG) signals".
His work aims at developing new EMG signal compression methods based on the combination of several transforms, the vector quantization and the superposition of several encoders.This study, under the supervision of Prof. Pierre Elé and Prof. Pascal Ntsama, was presented to the grand jury on November 03, 2017.
It appears that discrete wavelet transform and discrete cosine transform can be retained. It is also the case for SPIHT (set partitioning in hierarchical trees coding) and arithmetic coders. Several compression approaches have been developed and tested on two databanks of EMG signals. The first databank is made of EMG signals generated by our research team at 12bits per sample. The second databank on the other hand is made of EMG signals generated at 16 bits per sample by a Brazilian research team. Among the compression approaches that were developed, only five showed improvements over the literature. And among them, the best method is the MDWPT + TCD compression method followed by the innovative Modified Discret Wavelet Packet Transform (MDWPT) method. We recommend them for archiving and transmitting EMG signals. However, the other approaches can also be used. The results are very encouraging compared to those found in the literature and lead to many perspectives.