Detection of the origin of wolfberry based on electronic nose and electronic tongue combined with LSTM-AM-M1DCNN

ObjectiveTo achieve rapid detection of the origin of wolfberry.MethodsA rapid discrimination method for the origin of wolfberry was proposed based on an electronic nose and tongue system using a Long Short-Term Memory network-Attention Mechanism-Multi-scale one-Dimensional Convolutional Neural Network (LSTM-AM-M1DCNN) model. First, an electronic nose and tongue were used to detect wolfberries from five different origins. Then, the collected data were fused, and finally, the LSTM-AM-M1DCNN was employed to classify and discriminate the fused data.ResultsCompared with traditional LSTM and CNN methods, the LSTM-AM-M1DCNN effectively extracted deep feature information from the electronic tongue and nose signals. The accuracy, precision, recall, and F1-Score of the test set reached 97.4%, 97.6%, 97.4%, and 0.974, respectively.ConclusionThe use of LSTM-AM-M1DCNN overcomes the limitations of traditional convolutional neural networks that are not fully capable of extracting temporal and spatiotemporal features. It is suitable for processing data collected by the electronic nose and tongue and can effectively and accurately discriminate wolfberries from five different origins.