HuBot Dataset: Annotated Data for Non-Disruptive Bird Behavior Study

This dataset was created as part of the HuBot project, which focuses on developing a biomimetic mobile robot for non-invasive ecological research and wildlife monitoring. It contains a curated collection of 5,000 high-quality images of Houbara bustards, annotated for object detection tasks. The dataset aims to support research in ecological conservation and robotic applications, enabling accurate and reliable detection of Houbara bustards in diverse environments.

Dataset Details:
Size: 5,000 images, divided into:
Training set: 4,000 images.
Validation set: 500 images.
Testing set: 500 images.
Annotations: Each image includes bounding box annotations for detecting and localizing Houbara bustards.
Sources: Images were collected from in-house footage, online repositories, and globally deployed camera traps.
Features:
Bias Mitigation:

Includes diverse lighting conditions, backgrounds, and challenging scenarios to avoid overfitting.
Balanced representation of Houbara behaviors and habitats to ensure unbiased performance.
Data Augmentation:

Applied techniques such as horizontal flips, cropping, blurring, and noise addition to enhance model generalization.
Diversity and Representation:

The dataset captures a range of environmental conditions to support real-world applications and improve model robustness.
Future Scaling:

Plans are in place to expand the dataset to 50,000 images, covering rare Houbara behaviors, multiple subspecies, and more geographical locations.
Applications:
Object detection in biomimetic robotics.
Ecological research on endangered species.
Non-invasive data collection for long-term wildlife monitoring.
This dataset has been instrumental in optimizing HuBot’s object detection algorithms, allowing reliable performance in diverse conditions while minimizing disturbance to wildlife. It serves as a valuable resource for researchers and developers working on ecological robotics, conservation, and artificial intelligence.

License: This dataset is shared under the CC BY 4.0 license, allowing use for non-commercial purposes with appropriate attribution.

本数据集系HuBot项目的配套研究成果，该项目专注于研发用于无创生态研究与野生动物监测的仿生移动机器人。数据集包含经精心筛选的5000张高质量波斑鸨（Houbara bustard）图像，已针对目标检测任务完成标准化标注。本数据集旨在为生态保护与机器人应用相关研究提供支撑，助力在多样环境中精准可靠地检测波斑鸨。

数据集详情：
图像总量为5000张，划分为以下子集：
- 训练集（Training set）：4000张
- 验证集（Validation set）：500张
- 测试集（Testing set）：500张

标注说明：每张图像均包含用于检测与定位波斑鸨的边界框标注。

数据来源：图像采集自内部影像素材、在线资源库以及全球部署的野外触发相机（camera traps）。

特征设计：
1. 偏差缓解：涵盖多样的光照条件、背景环境与复杂场景，以避免模型过拟合；同时均衡呈现波斑鸨的行为模式与栖息生境，保障模型性能无偏倚。
2. 数据增强：采用水平翻转、裁剪、模糊处理与噪声添加等技术，以提升模型的泛化能力。
3. 多样性与代表性：数据集覆盖多种环境条件，可支撑实际应用场景并提升模型鲁棒性。
4. 未来扩容计划：团队计划将数据集扩容至50000张图像，涵盖稀有波斑鸨行为、多个亚种以及更多地理区域。

应用场景：
- 仿生机器人领域的目标检测研究
- 濒危物种生态研究
- 用于长期野生动物监测的无创数据采集

本数据集已有效助力优化HuBot的目标检测算法，使其可在多样环境中实现可靠检测，同时尽可能减少对野生动物的干扰。其可为从事生态机器人、物种保护与人工智能相关研究的科研人员与开发者提供宝贵的研究资源。

授权许可：
本数据集采用知识共享署名4.0（CC BY 4.0）许可证进行共享，允许在注明原作者的前提下用于非商业用途。