ISRMyo-I: A database for sEMG-based hand gesture recognition

Citation Author(s):
Yinfeng Fang, Dalin Zhou, Kairu Li, Honghai Liu
Submitted by:
Yinfeng Fang
Last updated:
Thu, 11/08/2018 - 10:34
DOI:
10.21227/H26Q26
Data Format:
License:
341 Views
Categories:
Keywords:
0
0 ratings - Please login to submit your rating.

Abstract 

 

This repository aims to publish a sEMG database for hand gesture recongnition, which is suitable for intra-session, inter-session, inter-day and inter-subject tests. Six subjects were involved in data collection on ten days, and two sessions a day with the interval of half an hour. In each session, one trial (10 secondes) for each geature was conducted. The electrode sleeve did not reweared between two sessions in a day. The utilised sEMG device was customised by the Intelligent System and Biomedical Robotics Group, which was discussed in [1]. 

 

Instructions: 

 

## Hand Gestures

The sEMG signals of thirty hand gestures were colleced, as listed below:

* Label 0: Hand Rest (HR)

* Label 1: Hand Open (HO) 

* Label 2: Hand Closed (HC)

* Label 3: Wrist Flexion (WF) 

* Label 4: Wrist Extension (WE) 

* Label 5: Wrist Pronation (WP) 

* Label 6: Wrist Supination (WS)  

* Label 7: Ulnar Flexion (UF)  

* Label 8: Radial Flexion (RF) 

* Label 9: Fine Pinch (FP) 

* Label 10: Key Pitch (KP)  

* Label 11: Spherical Grasp (SG)  

* Label 12: Cylindrical Grasp (CG) 

 

## Database Organisation

* emgData/RawEMG: the folder that contains the raw recorded signal directly from the device without any processing

* emgData/RawEMG/XXXX: the folder named by subjects' name, the data of which is contained 

* emgData/RawEMG/XXXX/EMGDDS.txt: the data file titled by EMG*dds*.txt, where *dd*, *s* indicate the ID of day and the ID of the session, respectively.

* emgData/train_XXXX: the folder that contained the training samples (first 7 days) after segmentation and prcessing from the raw EMG , where XXXX indicates the subject' name

* emgData/test_XXXX: the folder that contained the testing samples (last 3 days) after segmentation and prcessing, where XXXX indicates the subject's name

* emgData/train_XXXX/CXX_train.txt: the segmented training samples of channel XX. Each row in the file contains 256 scale value, indicating an observation from channel XX.  

* emgData/train_XXXX/X.txt: the extracted TDAR feature (128 dimension) extracted from all segmented data from CXX_train.txt, where 16 channels were collected in a line.

* emgData/train_XXXX/y.txt: the lable of training samples, and each row corresponding to each row in CXX_train.txt and X.txt

* emgData/test_XXXX/CXX_train.txt: the segmented testing samples of channel XX. Each row in the file contains 256 scale value, indicating an observation from channel XX.  

* emgData/test_XXXX/X.txt: the extracted TDAR feature (128 dimension) as testing samples extracted from all segmented data from CXX_train.txt, where 16 channels were collected in a line.

* emgData/test_XXXX/y.txt: the lable of testing samples, and each row corresponding to each row in CXX_train.txt and X.txt

 

## Authors

Dr. Yinfeng Fang 

Prof. Honghai Liu

 

## References

[1] Y. Fang, H. Liu, G. Li, and X. Zhu, “A multichannel surface emg system for hand motion recognition,” Int. J. Humanoid Robot., vol. 12, no. 2, p. 1550011, 2015.