Comparative Analysis of Different Image Augmentation Strategies for Automated Detection of European Mammals in Camera Trap Footage

Abstract

Artificial intelligence (AI) techniques are increasingly being used to automate image processing. However, state-of-the-art deep learning methods require large amounts of data and are sensitive to class imbalances. When these models receive insufficient or non-diverse training data, they can suffer from overfitting and reduced generalization ability. One solution is to implement image augmentation, which artificially increases dataset size and balance by simulating real-world variations of existing data points. Images can be augmented in various ways (resizing, rotating, mixing, coloring, etc.), and we refer to the combination of applied augmentations as an augmentation strategy. The optimal strategy varies depending on image content and research objectives.