Phenotyping of tortoise beetles: Automated extraction of morphology from thousands of beetle pictures

Illustration (click to hide):

Project Description

We want to investigate why populations differ in their ability to evolve and adapt. For this, we need to understand how and why the two main factors of evolution, selection and genetic variation, fluctuate in populations over time and space. Estimation of these factors requires measurements of phenotypic traits. This is why our aim is to extract morphological traits, such as body size and shape, from pictures of our study species the green tortoise beetle Cassida viridis, sampled in multiple wild populations in Sweden. We already have over 10,000 pictures and are planning to sample more beetles in the coming years. Deep learning would make it possible for us to analyze all pictures in a manageable time frame. With help of training data, we hope to extract specific measurements of the beetle’s protective shield through excluding features that add noise to the measurements, such as legs and antennae.