A discrimination model for dead cocoons using near-infrared transmission spectra analyses

Silk from cocoons is widely used as a functional raw material for alleviating skin diseases, improving memory, preventing diabetes, and as antioxidants. Dead cocoons are screened to utilize silk, resulting in poor marketability and cross-contamination of other cocoons. Considering that the screening of cocoons is performed manually to date, there is a high demand for automating this process. Spectroscopic analyses are actively used for non-destructive analysis methods. In this study, a discrimination model was developed to determine whether a cocoon is dead or alive using the near-infrared transmission spectra. A total of 367 normal cocoons and 152 dead cocoons were used in our experiment, and the cocoons were cut to determine whether they were dead or alive. The near-infrared transmission spectra were obtained for each cocoon for the wavelength band 900–1700 nm. Nine pre-processing methods were applied to eliminate spectral noise; pre-processing related to normalization demonstrated significant difference compared to other pre-processing methods. Furthermore, principal component analysis (PCA) was used to visualize the distribution of spectra, and discrimination models were developed using the wavelength range of the highest transmittance intensity and partial least squares-discriminant analysis (PLS-DA). PCA successfully discriminated between normal and dead cocoons. The discrimination model according to the wavelength range of the highest transmittance intensity demonstrated a discriminant accuracy of 94.80 %. Using PLS-DA with range normalization pre-processing spectra, the discriminant accuracies were 94.56 % and 94.64 % for calibration and validation, respectively. Thus, the proposed method can help discriminate between normal and dead cocoons non-destructively using the near infrared transmission spectra.