Research on Classification of Respiratory Diseases Based on BiGRU and CNN Cascade Neural Network

The mortality rate from respiratory diseases is increasing in our country. To this end, we propose a cascade neural network based on BiGRU and CNN to classify respiratory diseases. We extracted the spectral images of two different pixels by framing, frame shifting and windowing, and then input the cascaded neural network of BiGRU and CNN respectively. Then, the characteristics of the two are spliced to achieve classification. At the same time, we also extract a single pixel and input it into LSTM, GRU and BiGRU neural network for comparative experiments. In order to solve the problem of unbalanced data set class, we slice and extend the data set. Our system classifies six respiratory diseases, and achieves 91.9% ICBHI average accuracy, respectively. The ICBHI 2017 respiratory cycle dataset was used for 10-fold cross-validation of 90% of training data and 10% of test data, and repeated average accuracy.