Non-destructive detection of freshness in crayfish (Procambarus clarkii) based on near-infrared spectroscopy combined with deep learning

The whiteness (W), total volatile basic nitrogen (TVB-N) and total viable count (TVC) are three important indicators to evaluate the freshness of crayfish. This work aimed to develop robust models for non-destructive detecting multiple freshness indicators of crayfish during cold storage based on near-infrared (NIR) spectroscopy. After collecting NIR spectra of crayfish, the W, TVB-N and TVC were measured by traditional methods. Partial least squares regression (PLSR) models along with multiple preprocessing and variable selection methods and one-dimensional convolutional neural network (1D-CNN) models based on raw spectra were constructed. The results demonstrated that the wavelet threshold denoising (WTD) and competitive adaptive reweighted sampling (CARS) could improve the performance of PLSR models. Compared with other models, the 1D-CNN model showed the best performance in predicting TVB-N and TVC, with R-p(2) of 0.9397 and 0.9318, and RPD of 2.8279 and 2.7560, respectively, indicating outstanding advantages of CNN in NIR spectral analysis. To sum up, the overall results suggested that NIR spectroscopy combined with deep learning would be a feasible approach for detecting the freshness of crayfish.