Abstract: 【Aim】 Aiming at the low efficiency of manual sorting pupae in silkworm breeding and the susceptibility to subjective factors, a recognition method for female and male domestic silkworm (Bombyx mori) pupae based on texture features and improved VGG was proposed. 【Methods】 The transmission transformation was used to correct the direction of B. mori pupae, and the head and tail images of B. mori pupae were intercepted. Bchannel image was used as the basis of profile extraction. The profile complexity was analyzed by Douglas-Peucker (DP) algorithm to identify and obtain the tail image of B. mori pupae. The background interference was eliminated with a mask and the texture information was enhanced by multi-channel feature fusion image. The Inception module was improved, and the residual network and the improved Inception module were added to the VGG model. The data set was expanded by data enhancement technology, and three kinds of input images and four recognition models were evaluated and compared by using the precision, recall, harmonic average F1-score of the precision and recall, and accuracy as the evaluation indexes. 【Results】 The results showed that the precision, recall and F1-score of the improved VGG model of feature fusion images for female pupae of B. mori were 98.017%, 94.794% and 96.375%, respectively, while those for male pupae were 95.342%, 98.231% and 96.762%, respectively, and the accuracy in identifying female and male pupae of B. mori was 96.580%. The accuracy of the feature fusion image in identifying female and male pupae of B. mori was 18.093% higher than that of the original gray scale image, and the accuracy of the improved VGG in identifying female and male pupae of B. mori was 2.257% higher than that of the original VGG. 【Conclusion】 The recognition method for female and male B. mori pupae based on texture features and improved VGG can reduce the labor time, providing a basis for the realization of automatic sorting of female and male pupae of B. mori.

Key words:  Silkworm pupa, sex, texture features, Douglas-Peucker algorithm, Inception model, VGG network