Automated Rhinoceros Detection in Satellite Imagery using Deep Learning Dataset

This repository contains a dataset of satellite images for white rhino (Ceratotherium simum simum) detection. The dataset was released in “Automated Rhinoceros Detection in Satellite Imagery using Deep Learning” Scientific Reports. The dataset consists of 512x512 sub-images extracted from World-View-3 satellite images (c) Maxar Technology acquired between 2015 and 2022 in a rhino farm in South Africa. Each sub-image is manually labelled with bounding boxes around individual rhinos. – Real_Labels.zip

The repository also contains synthetic labels generated with Blend library – SyntheticLabel.zip.

Both types of labels are stored in text files with the same name as the corresponding image files and have the following format: label  xmin ymin width height. Label is 0 for all detections as only one class ‘rhino’ is present. xmin, ymin, width, height are relative to the image width and height, i.e. to know the pixel coordinates one needs to multiply them by image width or height: e.g., xmin_pixel = xmin * image_width, ymin_pixel = ymin * image_height.

Textures.zip contains the set of background images extracted from the original TIF files. These textures were used as environmental backdrops in the generation of synthetic data.

blend_scene.blend11 is the accompanying Blender project file. It includes the full procedural setup for synthetic image creation, specifically:

• Random placement of rhino models within each scene,

• Randomized background selection from the provided textures, and

• Rendering scripts for generating the final synthetic imagery.

Together, these files document both the source textures and the reproducible Blender workflow used to generate the synthetic dataset.

The elephant data can be found here.

 How to cite:

I. Duporge, X. Lin, A. Palnitkar, A. Suresh, O. Isupova, D. Rubenstein, and J. Y. Aloimonos. 2025. Automated Rhinoceros Detection in Satellite Imagery using Deep Learning. Scientific Reports.”