SpectralFlow: an integrated platform for spectral data preprocessing and predictive modeling analysis in fruit quality evaluation

Abstract

Near infrared spectroscopy and hyperspectral imaging offer unique analytical advantages and immense application potential in fruit quality evaluation. However, most existing spectral analysis software is limited to simple machine learning models, lacking support for complex hyperparameter tuning and state-of-the-art deep learning architectures. In addition, current software is primarily designed for analyzing one-dimensional spectral data, with limited functionality for extracting features from hyperspectral images. To address these challenges, we developed SpectralFlow software, which integrates interactive spectral data extraction and preprocessing, a built-in model library, custom model training, and dataset management with visualization capabilities. SpectralFlow significantly lowers the technical barrier for spectral analysis by optimizing spectral data analysis and simplifying the model training process. The software's performance was verified through two scenario studies: one involving the early prediction of anthracnose in mangoes, and the other involving the quantitative prediction of dry matter content (DMC) in apples, mangoes, kiwifruit, and pears. All models demonstrated excellent performance, with the mango anthracnose prediction model achieving an accuracy of over 92% and the model for predicting DMC achieving R-squared values of over 0.80. The proposed SpectralFlow proves to be highly effective software for spectral data preprocessing and model training.