An efficient and optimized technique for bio-signals compression using auto-encoder

Sunilkumar K. N., Shivashankar Shivashankar

Abstract


Latest developments in wearable machines permits un-damageable and cheapest way for gathering of medical data such as bio-signals (Electro-Cardio-Gram), Respiration, Blood pressure and so on. Gathering and processing of different biomarkers are considered to provide anticipatory healthcare by personalized software for medical purpose. As the wearable machines are dependent on size of the machine, different conditional resources and are used by batteries, we need to perform small algorithms to robustly control memory and the energy of the device (Machine). The rapid growth of the technology has led to various auto encoders that guarantee the results by extracting feature selection from time, frequency, and time-frequency domain in an efficient way. The main aim was to train hidden layer in a way, which can reconstruct the compressed data almost similar to the input data. While in the previous works, while compressing the data all the features were needed necessarily in the hidden layer to reconstruct it back. But, in our proposed work, a new algorithm BCAE Bio-signals Compression using Auto-Encoder is used, which is a re construction model that performs selection of the important features and compression of the data using auto-encoder. The core idea is to perform combined feature selection and compression for ECG signals at source and to decrease energy consumption while transmitting and thus prolongs the battery lifetime. Fitness monitors for wearables by using new methodologies will differentiate the unwanted features from the necessary once for further classification. Performance results produced with respect to ECG signals of our BCAE computations from viable wearable monitor that is wireless with same of Subject-Adaptive Unsupervised (SURF) ECG compression for devices that are wearable represents the dominance of BCAE when compared with the SURF ECG signal compression for Fitness-Monitors-of-Wearable devices (SURF). 1) Ratio of compression is around 90x ; 2) error of re-construction ranging between 2% - 7% and 3) minimization of power usage of around 2 orders of magnitude based on transmitting the uncompressed signals during maintaining its morphology.

Keywords


Auto-Encoder, Feature Selection, Bio-Medical Signals



DOI: http://doi.org/10.11591/ijece.v11i1.pp%25p
Total views : 0 times


Creative Commons License
This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.

ISSN 2088-8708, e-ISSN 2722-2578