Leafy vegetable freshness identification using hyperspectral imaging with deep learning approaches

With more attention being paid to healthy diets, vegetable freshness is a concerning issue for consumers and traders. Rapid and non-destructive evaluation of vegetable freshness plays an important role in vegetable consumption. In this study, hyperspectral imaging was used to identify the freshness of spinach and Chinese cabbage stored at different durations. Classification models were built using the extracted average spectra, including conventional machine learning methods (logistic regression (LR), support vector machine (SVM), and random forest (RF)) and deep learning methods (convolutional neural networks, long-short term memory (LSTM), and CNN combined with LSTM (CNN-LSTM)). Results showed that CNN-LSTM models outperformed the other models for both vegetables, with classification accuracy over 80 % in the training, validation and testing set. GradCAM++ methods showed that great similarity of important wavelengths contributing more to the freshness identification between the vegetables existed. Both fine-tuning and direct prediction using the model built on one vegetable to predict the other vegetable were explored. Good performances were obtained for both situations with the classification accuracy over 80 % for the three sets using all training samples. The overall results illustrated the great potential of hyperspectral imaging with deep learning approaches to identify the freshness of different vegetables. This study provides a basis for further assessment of vegetable freshness.