Enhancing silkworm pupae gender classification via surface-specific RGB imaging

Metagenomics profiling of midgut bacterial microbiome of Pectinophora gossypiella (Saunders) fed on double Bt cotton Determining the sex of silkworm pupae is an important task in the field of sericulture. Manually sexing silkworms is extremely time-consuming and prone to error and it is not feasible for large-scale automation. This paper investigates the use of a two-part deep learning approach that uses Surface Net (which identifies whether a surface is ventral or dorsal side) and SexNet (which identifies the sex of a pupae based on a specific surface). From RGB images of Bombyx mori pupae, the proposed approach attained 100% surface detection success. Performance validation was undertaken with a combination of McNemar’s test, two-proportion z tests, and 95% binomial confidence intervals. Results demonstrate that Gen der was classified with higher confidence on the ventral surface (97.3%, 95% CI: 93.3-99.3) as opposed to the dorsal (91.0%, 95% CI: 85.6-94.9) with a confidence of z = 2.35, p = 0.019. With accuracy decreases of 10.7% (ventral) and 14.9% (dorsal) when masked, the occlusion test also verified that the tail region was the most distinguishing feature part. This work establishes the ventral region as the primary sex marker in silkworm p pupae and provides a validated framework for automated, non-invasive pupae sexing in practical sericulture workflows.