Noisy Electronic Tongue Signal Prediction for Tea Quality Estimation Using Autoencoder

Assessing the quality of tea is crucial in the tea industry. While traditional methods relied on skilled tea tasters, modern times have seen a rise in the use of electronic tongues (ET) for automated quality assessment. Yet, interpreting ET signals can be complex due to their many aspects, like dimensionality, variability, and vulnerability to noise. However, the problem of recognizing ET signals in the presence of noise has not been explored in existing studies. In our research, we tackled the challenge of predicting tea quality from noisy ET signals. We explored how an auto-encoder model responds on noisy ET signal in a more compact way. The encoded patterns from the autoencoder act as the distinct features of the ET signal. Through experiments on noisy ET signals with different input pulse waveforms, we found that autoencoder-generated features are very reliable. They can achieve a high level of accuracy, even when the signal to noise ratio (SNR) is 20 dB. This indicates that the proposed method can efficiently distinguish ET signals even when there’s substantial noise present.