Paper Details

In sericulture, the gender classification of silkworm pupae is particularly important since it has an impact on the
amount and the quality of silk seeds produced.This paper presents a hybrid deep learning framework that combines CNN-based
feature extraction with handcrafted morphological features for gender classification of ventral silkworm pupae images. The
model uses highly specific morphological features of the pupal body, such as the presence of a vertical line in female pupae and
the dot-like structure in male pupae in posterior abdominal region ,the handcrafted morphological descriptors were integrated
with CNN-derived deep features for binary classification.This work achieved 100% accuracy,recall, precision, and F1score on
an independent test set. Statistical tests confirmed the robustness of these results: the Wilson score 95% CI for accuracy was
[0.981, 1.000], the AUC was 1.00 (95% CI: [1.000, 1.000]), and the binomial test against chance yielded 𝑝 < 0.000001. By
combining handcrafted morphological features with CNN-derived features, the model achieved 100% accuracy in five-fold
cross-validation. Five-fold stratified cross-validation further demonstrated consistency, with a mean accuracy of 1.000 ± 0.000.
Additional reliability metrics (MCC, Kappa, specificity, and Brier score) consistently supported perfect agreement between
predictions and ground truth. Such high accuracy is not only important from a scientific standpoint but is quite important in
prac tical applications in the silk seed production process, where slight inaccuracies might result in huge economic impact and
loss of yield quality.

Jyoti Sharma, Pradeep Chouksey, “Feature Engineering and Hybrid Deep Learning for Gender Classification of Bombyx mori Pupae”, International Journal of Advance Engineering and Research Development (IJAERD), Vol. 13, Issue 8, pp. 10-21, August 2026.