Silkworm cocoon classification using fusion of Zernike moments-based shape descriptors and physical parameters for quality egg production

This paper proposes a novel non-destructive vision-based system to perform automated gender classification of silkworm cocoons for the purpose of improving the quality of egg production. Gender classification is achieved by discriminative learning on samples of male and female cocoons acquired from CSR2 and Pure Mysore silkworm cocoon breeds. Features composed of different combinations of weight, volume, geometric and Zernike moments-based shape properties of cocoons are used for training classifiers of types k-nearest neighbour (kNN), linear discriminant analysis (LDA), neural networks (NNs) and support vector machine (SVM). The experimental results show superiority of the proposed method with respect to the state-of-the-art methods. Specifically, the experimental results indicated the excellent performance of NN and SVM classifiers. An overall classification accuracy of 91.3% was achieved with NN for CSR2 cocoon breed and 100% was attained with SVM for pure Mysore cocoon breed.