A total volatile basic nitrogen detection method for yellow croaker based on an electronic nose combined with an...

Area of Science:

• Food Science
• Analytical Chemistry
• Machine Learning

Background:

• Electronic noses offer rapid food quality monitoring but generate complex time-series data.
• Traditional models struggle with non-linear signals, leading to poor feature extraction and prediction accuracy.

Purpose of the Study:

• To develop a novel method for predicting total volatile basic nitrogen (TVB-N) using electronic nose data.
• To improve the accuracy and reliability of fish freshness detection.

Main Methods:

• Proposed an attention improved encoder long short-term memory (AIE-LSTM) hybrid network with dual-stream feature fusion.
• Utilized a modified informer encoder with multi-head attention and a temporal down-sampling layer.
• Integrated raw voltage sequences and physical features using a bidirectional LSTM and multi-level feature fusion.

Main Results:

• The AIE-LSTM model achieved optimal TVB-N prediction performance across nine electronic nose datasets (average R² of 0.960).
• Demonstrated the lowest relative standard deviation for R², root mean square error (RMSE), and mean absolute error (MAE).
• Achieved R² of 0.979, RMSE of 0.988, and MAE of 0.589 in the best-performing batch.

Conclusions:

• The AIE-LSTM model significantly enhances the prediction accuracy of fish TVB-N content.
• Provides a reliable technical solution for rapid, non-destructive, and accurate fish freshness detection.