DonaDataset: A Camera Trap Image Dataset for Wildlife Monitoring in Doñana National Park

DonaDataset is a camera trap image dataset maintained by the wildintelproject, released on GitHub on May 26, 2026 (link: https://github.com/wildintelproject/donadataset). This dataset is used to train the DonaNet neural network based on the YOLO architecture for detecting and classifying mammals in Spain's Doñana National Park. Its core value lies in combining real field data with AI technology to enable efficient and accurate wildlife monitoring, supporting ecological conservation and research.

Wildlife population monitoring is a core task in ecological research and conservation. However, traditional manual observation has issues such as being time-consuming and labor-intensive, and being susceptible to subjective influences—especially for nocturnal or alert species, it is difficult to obtain accurate data. Although camera trap technology can record wildlife images 24/7, processing and analyzing massive amounts of data has become a new challenge. As an important wetland ecosystem in Europe, Doñana National Park is a habitat for endangered species like the Iberian lynx, and its monitoring is crucial for conservation decisions.

DonaDataset's features include:

1. Authoritative data source: All images come from actual monitoring projects in Doñana National Park, reflecting real field environments;
2. Targeted species coverage: Focuses on common mammals in the park, with special attention to protected species, and the sample size is sufficient to support model training;
3. Rigorous annotation quality: Each image is reviewed by professionals to ensure accurate species classification, providing a reliable foundation for DonaNet training.

DonaDataset is mainly used to train DonaNet, which is based on the YOLO architecture. The reasons for choosing YOLO include:

1. Real-time detection capability: Achieves fast inference while maintaining accuracy, suitable for processing large numbers of camera trap images;
2. End-to-end training: Simplifies the development process, allowing direct training with the dataset without complex preprocessing;
3. Good scalability: Supports different scale model variants (e.g., YOLOv5s/m/l), which can balance precision and speed requirements.

The technical characteristics of DonaDataset have important value for machine learning research:

1. Complexity of field environments: Includes real scenarios such as light changes, weather effects, occlusions, and variable animal postures, enhancing the model's generalization ability;
2. Class imbalance: Reflects differences in species occurrence frequency in the wild, providing real data for studying class imbalance issues;
3. Long-tailed distribution: Rare species have few samples, which conforms to ecological laws and provides challenges and opportunities for training reliable detection models.

The application prospects of DonaDataset and DonaNet include:

1. Automated species identification: Replaces manual identification to improve efficiency;
2. Population dynamics monitoring: Establishes long-term databases to support ecological research and conservation decisions;
3. Invasive species detection: Quickly identifies alien species to assist early warning;
4. Behavior pattern research: Can further analyze animal activity rhythms, social behaviors, etc.

The significance of opening DonaDataset: Promotes research collaboration, promotes method standardization, provides educational resources, and raises public conservation awareness. Key considerations for technical implementation include:

1. Continuous learning: Adapts to changes in field environments and species updates;
2. Edge deployment optimization: Suitable for local real-time analysis of camera traps in remote areas;
3. False positive handling: Distinguishes between real animals and false triggers;
4. Privacy and ethics: Handles possible human activity data captured.