Division of Condica capensis Larval Instar Based on Four Clustering Algorithms

Purpose To investigate the division of Condica capensis larval instar by four classic clustering algorithms, and clarify the optimal indicators.

Methods A total of 420 laboratory-reared C. capensis larvae were used as experimental materials. The head capsule width, head capsule length, and antenna spacing were measured. Gaussian mixture model (GMM), K-means clustering, DBSCAN clustering, and OPTICS clustering were used to divide the larval instar, and traditional instar classification indicators and regression analysis were used to validate the clustering results.

Results All four clustering algorithms identified six larval instars in C. capensis. GMM and K-means clustering performed better than DBSCAN clustering and OPTICS clustering. The head capsule width of 1-6 instar were 0.260-0.300 mm, 0.374-0.484 mm, 0.594-0.836 mm, 0.899-1.273 mm, 1.406-1.645 mm, and 1.801-2.254 mm, respectively; head capsule length were 0.228-0.296 mm, 0.319-0.416 mm, 0.515-0.631 mm, 0.733-0.909 mm, 0.944-1.250 mm, and 1.305-1.664 mm, respectively; antenna spacing were 0.110-0.163 mm, 0.217-0.309 mm, 0.382-0.543 mm, 0.523-0.734 mm, 0.828-1.093 mm, and 1.061-1.480 mm, respectively.

Conclusion The GMM and K-means clustering are suitable methods for determining the larval instars of C. capensis. Head capsule width and length are reliable morphological indicators for larval stage division, while antenna spacing can be used as an auxiliary indicator.