Aimed at the challenge of high precision detection for filling flow during the filling process of viscous food, a real-time filling flow detection method based on deep learning was proposed. The collected flow-related process variables were serialized and normalized to be converted into data that could be handled by the supervised learning network firstly. Then, the neural network of long short-term memory based on the attention mechanism (LSTM-Attention) was trained and generalized by using the Adaptive moment estimation (Adam) optimization algorithm, thus the model for detecting filling flow of thick liquid food was established. Lastly, the filling flow value detected by this method was compared with the actual flow value, and in the meanwhile the performance of the model in the filling flow detection was evaluated with the mean square error function (MSE). Moreover, compared with the Recurrent Neural Network (RNN) and common long short-term memory (LSTM) method, the experimental results demonstrated that the validity and the high precision of the proposed model were verified, and the real-time tracking effect of the filling flow rate data was better.