Core temperature estimation of food items based on non-contact thermal and high frequency sensor data with an LSTM network

In cooking appliances the long-term dependencies of the cooking climate and its complex interaction with the size and structure of the food makes it hard to estimate the current state of the food item, such as its core temperature. In this paper, the inverse problem of estimating the state is solved by a Long Short-Term Memory (LSTM) network which estimates a full probability density of possible states. A
$k$
-nearest neighbor (kNN) algorithm is presented as a strong and explainable baseline model.