Feature interpreted convolutional neural networks for real time implementation of respiratory data

Automatic detection of respiratory diseases is important to prevent any sudden death in patients. At present, respiratory diseases are detected by a physician who normally consumes more time to detect. In this work, the respiratory data from the persons are classified into either normal or abnormal using the proposed deep learning architecture. The proposed work consists of two subsequent phases namely training and testing. In training phase of the respiratory classification system (RCS), the respiratory data from both normal and both abnormal cases are individually data augmented in order to eliminate the overfitting issues in deep learning architecture. This data augmented respiratory data from both normal and abnormal case is used to construct data augmented matrix (DAM) which is trained by the proposed feature interpreted convolutional neural networks (FICNN) to produce the trained data. The proposed FICNN work obtained a 99.9% respiratory detection rate (RDR) with 0.05 ms as computational time.