Förbättring av ett program för att räkna bladskärarmyror

Abstract

his project explored how to improve upon an existing program for counting leaf cutting ants. The original program could count the number of ants but could not distinguish between the different types of leaf-cutting ants (soldier, worker and mi nis). The main goal of this project was to count and classify each of the different ant types. Both the original program and our expansion of it analysed videos of ants filmed from above to count ants. A leaf-cutting ant colony at Universeum in Gothenburg was filmed and used as source material. Due to uncertainties if the mini ant type appeared in the source material, ants were only classified as worker or soldier. The classification was done with two methods, one based on analysing the ant’s size and the dominant colour, the other based on machine learning where three different models was used. Results from the different methods varied, with no method able to discern the soldier ants with great precision. The reason for the classification issues with soldier ants could be because of low film quality, low ant type variation or a lack of training data for the machine learning models.