Authors |
|
|||||||
|
||||||||
Supporting Institution |
: | |||||||
|
||||||||
Project Number |
: | |||||||
|
||||||||
Thanks |
: |
Cover Download | Context Page Download |
Ahmet ÇINAR1 , Şule ŞENLER YILDIRIM2
It is important to perform basic rehabilitation movements such as walking and sitting, which are planned individually, for the follow-up of the patient undergoing physical rehabilitation treatment and to examine the development of the patient over time. In this paper, movement classification based on hybrid deep learning algorithm is proposed to determine if the exercises given by the doctor are performed by the patient accurately. Six basic movements are classified as standing, sitting, laying, walking, walking upstairs, walking downstairs by means of the proposed Alexnet-SVM (Alexnet-Support Vector Machine) hybrid model. During the training and testing stages of the system, 1492 movement signals obtained from smartphone (Samsung Galaxy S II) on the waist are used. For the classification of movement signals by hybrid Alexnet-SVM, at first, the spectrogram images of the signals are obtained by means of Short-time Fourier transform. In order to use the obtained spectrographic images directly in Alexnet-SVM architecture, crop and resize preprocessing are applied. To show the superiority of the proposed Alexnet-SVM, the results are compared with the results of Alexnet, Resnet18 and Resnet18-SVM. The performance parameters of the proposed Hybrid Alexnet-SVM classifier are calculated using accuracy and F1-scores, resulting in 87.67% and 93.4%, respectively. The results obtained facilitate the follow-up of the patients who have lost their mobility, whether they are doing the movement correctly and it is possible to determine whether the correct treatment is applied or not.
Keywords
Deep learning,
Movement Human Actions,
Classification,
Alexnet,
SVM ,
Authors |
|
|||||||
|
||||||||
Supporting Institution |
: | |||||||
|
||||||||
Project Number |
: | |||||||
|
||||||||
Thanks |
: |