Cover Image

Feature extraction and classification algorithms for assessing muscle condition using multi sensors

Master of Science in Biomedical Electronic Engineering

Saved in:
Bibliographic Details
Main Author: Mohamed Sarillee, Mohamed @ Mohd Kassin
Other Authors: Hariharan, Muthusamy, Dr.
Format: Thesis
Language:English
Published: Universiti Malaysia Perlis (UniMAP) 2017
Subjects:
Multisensor data fusion
Electromyography
Signal processing
Hamstring muscle
Muscle fatigue
Online Access:http://dspace.unimap.edu.my:80/xmlui/handle/123456789/78085
Tags: Add Tag
No Tags, Be the first to tag this record!
id my.unimap-78085
record_format dspace
spelling my.unimap-780852023-03-07T07:12:31Z Feature extraction and classification algorithms for assessing muscle condition using multi sensors Mohamed Sarillee, Mohamed @ Mohd Kassin Hariharan, Muthusamy, Dr. Multisensor data fusion Electromyography Signal processing Hamstring muscle Muscle fatigue Master of Science in Biomedical Electronic Engineering Hamstring muscle strain injury is the most common injuries in sport environment especially in football. Beside sport, dancing is another physical activity that has high risk for hamstring muscle strain injuries. Muscle fatigue is one of the risk factors that cause muscle injury. Muscle fatigue occurs when muscle fail to maintain desired muscle strength. The aim of this work is to assess and distinguish muscle fatigue using 3 different myograms (Electromyogram, Mechanomyogram and Acoustic myogram). To achieve the aim of this project, 3 different myograms were recorded simultaneously from hamstring muscle (Biceps Femoris and Semimembranosus) during isometric contraction with different loads (5, 7.5 and 10kg). 20 healthy male subjects were recruited in this work with aged 22.4 ±2.6 years. The recorded myograms were denoised and segmented. Pearson correlation and linear regression were applied on 9 time and frequency domain features (were extracted from each frame of each myograms) to find the behavior of the myograms and relationship between the myograms. Based on the pearson correlation results there are a strong relationship between AMG and EMG signals and a weak relationship between MMG and EMG signals. However, the relationship of MMG and EMG signals becomes significant when the load is increased. Two types feature sets were used to classify muscle condition and there are conventional features set (CFS) and M-band wavelet transform based feature set (MFS). In CFS, 16 features were extracted from each myograms, which were used in previous studies to assess muscle condition. In MFS, the recorded myograms were decomposed using 4-band wavelet transform and 9 features were extracted from each node of each myograms. Therefore, 108 features were generated for each myograms. The feature sets of EMG, MMG and AMG of CFS and MFS were augmented to form a new feature set. Two stage feature selection was used to reduce feature space in the new feature set. In stage I of features selection, linear locally embedding (LLE) was used. New reductive feature space was used in stage II for feature selection. In that stage, binary particle swarm optimization (bPSO) was used. k-nearest neighbor (k-NN) classifier was used to classify muscle condition (non-fatigue or fatigue) by using individual myogram feature sets, concatenated / augmented feature sets and after 2 stage feature selection feature sets. In the individual assessment, EMG feature sets have obtained high accuracy. However, compare with the concatenated feature sets, this feature sets have given high accuracy. Two stage feature selection was applied to the concatenated feature sets and the features were reduced by 69% and 89% for CFS and MFS in stage 1 respectively. In stage 2 the features were reduced by 7% and 40% for CFS and MFS respectively. The classification result was improved to above 75% and 92% for CFS and MSF respectively. It can be conclude that, the combination of three myogram provides more useful information than a single myogram. 2017 2023-03-07T07:12:31Z 2023-03-07T07:12:31Z Thesis http://dspace.unimap.edu.my:80/xmlui/handle/123456789/78085 en Universiti Malaysia Perlis (UniMAP) Universiti Malaysia Perlis (UniMAP) School of Mechatronic Engineering
institution Universiti Malaysia Perlis
building UniMAP Library
collection Institutional Repository
continent Asia
country Malaysia
content_provider Universiti Malaysia Perlis
content_source UniMAP Library Digital Repository
url_provider http://dspace.unimap.edu.my/
language English
topic Multisensor data fusion
Electromyography
Signal processing
Hamstring muscle
Muscle fatigue
spellingShingle Multisensor data fusion
Electromyography
Signal processing
Hamstring muscle
Muscle fatigue
Mohamed Sarillee, Mohamed @ Mohd Kassin
Feature extraction and classification algorithms for assessing muscle condition using multi sensors
description Master of Science in Biomedical Electronic Engineering
author2 Hariharan, Muthusamy, Dr.
author_facet Hariharan, Muthusamy, Dr.
Mohamed Sarillee, Mohamed @ Mohd Kassin
format Thesis
author Mohamed Sarillee, Mohamed @ Mohd Kassin
author_sort Mohamed Sarillee, Mohamed @ Mohd Kassin
title Feature extraction and classification algorithms for assessing muscle condition using multi sensors
title_short Feature extraction and classification algorithms for assessing muscle condition using multi sensors
title_full Feature extraction and classification algorithms for assessing muscle condition using multi sensors
title_fullStr Feature extraction and classification algorithms for assessing muscle condition using multi sensors
title_full_unstemmed Feature extraction and classification algorithms for assessing muscle condition using multi sensors
title_sort feature extraction and classification algorithms for assessing muscle condition using multi sensors
publisher Universiti Malaysia Perlis (UniMAP)
publishDate 2017
url http://dspace.unimap.edu.my:80/xmlui/handle/123456789/78085
_version_ 1772813129956196352
score 13.209306

Similar Items