This is further development of online bine detection algorithm. In this project, a simple two-layer neural network is designed for prediction of bite and non-bite segment over a chewing sequence. The input layer and the hidden layer contain 50 units and 10 units respectively, while a single unit predicts the event at output layer. There are 452 samples splitted into train/test set with 70-30 percentage. Training accuracy of 99.34% and testing accuracy of 91.54% can be achieved after fine-tuning the hyperparameters.
tensorflow 1.15.0, numpy, h5py, pandas, matplotlib
Run TensorFlow_BiteDetection.ipynb to see algorithm's workflow step by step. Hypermarameters can be adjusted to give different results in training phase and testing phase.
Future work includes test in different architectures of deep neural network, including recurrent neural networks (RNN) and long short-term memory models (LSTM)