Abstract—Data collection in animal ecology has become more common, and extracting information can be done using deep learning methods. Challenges arise when using the dataset as the training input, including balancing the data quality and quantity. This research aims to investigate the effect of data balancing techniques on the optimizer—specifically, Root Mean Square Propagation (RMSProp) and Adam—in a deep learning image recognition model. Once recognition capability is obtained, it provides feedback for an Image-Based Visual Servoing (IBVS) system. The wildlife dataset is from Serengeti National Park and contains 41,762 images across 7 classes. Before applying data-balancing techniques, we determined that 1,100 images were needed, which we considered a balanced distribution across all classes since it was close to the original dataset. The next step was applying the balancing technique. We offered two balancing techniques: the first was to augment three classes with each class’s 1,100 images, while the other four classes retained their original number of images. The second involved reducing the number of larger classes to 1,100 images. The result showed that combining the deep learning model AlexNet with RMSProp achieved the highest Macro F1-Score of 0.8118, outperforming the other deep learning models with the Adam optimizer. The best balancing technique reduced the larger class to a fixed number of images in the smaller dataset, and AlexNet with RMSProp is a good choice for real-time visual servoing.
 
Keywords—data balancing techniques, wildlife animal recognition, image-based visual servoing, Root Mean Square Propagation (RMSProp), Adam, deep learning


Copyright © 2026 by the authors. This is an open access article distributed under the Creative Commons Attribution License which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited (CC BY 4.0).