Improving SSC detection accuracy of hanging-transported strawberries through different correction methods and 1D-CNN

Abstract

The clamp-hanging method for conveying strawberries by their stems offers distinct advantages for online detection of soluble solids content (SSC) by visible/near-infrared (Vis/NIR) spectroscopy. It minimizes fruit damage compared to traditional tray conveyance and facilitates non-destructive sorting in industrial applications. However, variations in hanging height and fruit size can alter detection zones and effective optical path length, while the inherent spatial heterogeneity of SSC distribution collectively compromises measurement accuracy for the whole-fruit. To tackle these obstacles, this study developed a refined clamp-hanging prototype and advanced spectral correction techniques for accurate Vis/NIR spectroscopy-based SSC prediction. A visual imaging module was integrated to monitor fruit size and hanging height in real time. Building on this, single spectral correction methods, including extinction coefficient correction (ECC) and correlation coefficient correction (CCC), were evaluated, with ECC delivering the best performance. A combined ECC-CCC approach further improved accuracy, achieving a determination coefficient of prediction ($R_p^2$) of 0.916 and a root mean square error of prediction (RMSEP) of 0.287°Brix using competitive adaptive reweighted sampling-partial least squares regression (CARS-PLSR), effectively reducing optical path fluctuations. Innovatively, a joint strategy incorporating SSC distribution correction with these spectral correction methods yielded superior results with $R_p^2$ = 0.945 and RMSEP = 0.229°Brix, thereby enhancing overall prediction reliability. Additionally, a one-dimensional convolutional neural network-long short-term memory (1D-CNN-LSTM) model applied directly to raw spectra achieved optimal outcomes with $R_p^2$ = 0.948 and RMSEP = 0.225°Brix, promoting robustness without preprocessing. Collectively, these innovations advance non-destructive, automated SSC detection, outperforming existing methods in accuracy and efficiency for strawberry quality assessment, potentially advancing online internal quality evaluation for small, delicate fruits.