Low-Cost Wearable Sensor System And BI-LSTM-HMM For Effective Monitoring and Detection of Food Intake Events

From the outset, we created the iHearken, a pair of non-intrusive headphones that can record sounds of food chewing and then use those recordings to categorize the different kinds of food intake that patients experience. This will help dietitians and health care providers keep tabs on their patients’ food use. Combining convolutional as well as recurrent neural networks with bidirectional long short-term memory (BiLSTM) is the hybrid model we presented.In a separate study, we integrated a method for chew event detection from chewing noise—a hybrid strategy for chew sound event recognition based on a plus with bidirectional long short-term memory- Hidden Markov Model (BiLSTM-HMM)—into our existing methodology. For this classification with several labels task, we use a cutting-edge chew recognizing RNN and HMM model. Since this method only needs a little quantity of experimental data and doesn’t need analytical features, it finds easy application to several problems. Concerns driven by biology may benefit from an emphasis on transitions as it is crucial to understand the causes of the occurrence of certain dynamic states. High-level behaviour with multiple low-level dynamics states may also benefit from it. After comparing other techniques, it is clear that the suggested BiLSTMHMM framework outperforms them all with an accuracy of $96.43 \%$. Bi-LSTM then shows impressive results, but long short-term memory (LSTM) and convolutional neural networks (CNN) are not as accurate. As a result, it is evident that the BiLSTM-HMM framework provides the most accurate solution compared to the others.