Evaluating the Impact of Modern Data Augmentation Techniques on UAV-Based Livestock Detection
Çetin Yalçın Yusuf Yargı BAYDILLI
AbstractIntegrating deep learning-based object detection models with unmanned aerial vehicles (UAVs) enables faster, more efficient, and cost-effective livestock monitoring. However, deep learning models require large and diverse datasets to achieve high accuracy. Traditional data augmentation techniques may be inadequate for complex tasks like object detection. Therefore, this study evaluates the performance of deep learning models on goat and cattle images using Cutout, CutMix, MixUp, and Mosaic data augmentation techniques. Ablation experiments revealed that Mosaic augmentation contributed the most to model success. These findings highlight the critical role of selecting the right data augmentation strategy in enhancing the stability and scalability of UAV-based livestock analysis.