Simultaneous identification of sex and species of silkworm pupae using hyperspectral imaging with PCA-t-SNE dimensionality reduction method

The breeding of silkworms is a crucial step in silkworm rearing. To breed high quality hybrid silkworm species, it is necessary to obtain silkworm pupae of different sex and species with high purity. This study explores the feasibility of simultaneously identifying the sex and species of silkworm pupae using hyperspectral imaging (HSI) technology. Firstly, a total of 400 hyperspectral images of 4 species of silkworm pupae were acquired using the hyperspectral imaging system, and the spectral information within the wavelength range of 400.89–1002.19 nm was extracted. The original spectral data was preprocessed using Savitzky-Golay (SG) smoothing combined with multivariate scattering correction (MSC), and a combined dimensionality reduction method of PCA-t-SNE was constructed to reduce the dimensionality of spectral data, aiming to retain the global and local features of the original spectra as much as possible. Subsequently, multiple classifiers were utilized to achieve the simultaneous identification of the sex and species of silkworm pupae. Among them, classification accuracies of the test set, utilizing PCA-t-SNE combined dimensionality reduction on the WHO-SVM and CNN models, achieve 93.75% and 95%, respectively. The results indicate that the combined dimensionality reduction method of PCA-t-SNE can extract representative spectral information more effectively and improve the classification accuracy. Simultaneously, the CNN model with strong learning ability has great potential in the classification of silkworm pupae. In conclusion, HSI technology can be effectively utilized for the identification of silkworm pupae to accomplish high-precision and complex classification tasks.