Projects
AHDriFT-ID: species identification for drift fence camera traps
AHDriFT-ID is a wildlife classification model designed for close-focus, downward-facing camera traps used in the Adapted Hunt Drift Fence Technique. It addresses the unique challenges of these camera setups, including limited context and partial animal detections.
The model is based on a fine-tuned SpeciesNet architecture and trained on 118,554 images from 1,169 camera trap locations in Ohio, USA, covering 46 taxonomic classes. Data were split by camera location into train, validation, and test sets to ensure spatial independence. The model achieved 83.6% accuracy on the held-out test set. Taxonomic fallback is supported, allowing predictions at higher taxonomic levels when species-level confidence is low. The model will be updated as new labelled data become available, and users are encouraged to contribute AHDriFT datasets via lila.science.
Web application for automated analysis of drone imagery
Addax Data Science has developed a web-based application for DroneWild to process drone imagery and video recordings. The platform enables users to securely log in, upload datasets, and select species of interest before running automated analyses with AI models.
The system generates outputs such as heat maps, tabular summaries, and visualised results that can be downloaded in multiple formats. For large-scale datasets, users can schedule batch jobs to run overnight and return later to review the results.
Continuous bird sound identification system for real-time monitoring
This open-source bird sound identification system is built upon the BirdNET-Pi project. The system uses a Raspberry Pi with a microphone to automatically identify bird species 24/7 in the field.
The device runs continuously, detecting and identifying bird calls in real-time through a simple web dashboard. Users can track which birds are present, monitor activity patterns, and analyze data from multiple locations. Built as an open-source project with a non-commercial license, this system is freely available to researchers, conservation groups, and bird enthusiasts worldwide. It provides an affordable way to monitor bird populations and study wildlife behaviour without disturbing natural habitats.
Automate species recognition in camera trap images from British Columbia
Addax Data Science partnered with Wild Eyes Monitoring Solutions Ltd. to develop a species recognition model for camera trap imagery from Western Canada. Trained on over 5.3 million animals, the model can distinguish 87 terrestrial species or higher-level taxons and achieved 91.9% accuracy on an out-of-sample test set. It also identifies sex and age classes in five deer species, allowing for more detailed ecological insights.
To ensure robust performance, data were split by camera location to prevent shortcut learning and encourage true generalisation. Users can enable taxonomic or sample-based prediction aggregation to improve reliability in challenging cases.
Hippopotamus deterrent system for human-wildlife conflicts in Zimbabwe
Addax Data Science has partnered with Hack The Planet to develop a recognition model for detecting hippopotamuses (hippos) in the field, aiming to reduce human-hippo conflicts. The primary goal of this project was to create a lightweight, efficient system capable of running on microcontrollers, making it suitable for deployment in remote areas with limited resources.
In Zimbabwe, where human-hippo conflicts are a growing concern, the system will function as a real-time deterrent when hippos are detected nearby. By training and quantizing recognition models, Addax contributed to the development of a fast, compact system capable of running on low-energy edge devices. This project provides a practical, cost-effective solution to reduce human-wildlife conflicts in Zimbabwe.
Automate species recognition of UK mammals in drone imagery
DroneWild collaborated with Addax Data Science to develop an AI pipeline to identify UK mammals in drone imagery. Using a two-stage approach, a detection model was used to locate animals, while a classification model identified the species. The dataset consisted of 39,000 drone images provided by DroneWild, which contained approximately 325,000 individual animals.
To minimise data leakage, the dataset was split into training and validation sets based on image sequences. Performance varied by species, with red and fallow deer being the most accurately classified. Precision, recall, and F1-scores varied by species, indicating where the dataset could be expanded to improve performance on underrepresented species.
Teaching conservationists computer vision at Caltech university
In January 2025, Addax Data Science contributed to the three-week CV4Ecology workshop at Caltech in Los Angeles. It aimed at equipping ecologists with the skills to analyze large image, audio, and video datasets using computer vision. The course taught students how to train and evaluate computer vision models on their own data, combining classroom instruction with real-world projects.
Peter from Addax served as an instructor, teaching foundational concepts, mentoring students one-on-one, and lecturing to support their understanding of how computer vision can address ecological research questions. Participants left the workshop with tools that fit their projects, a strong foundation in computer vision, and a network of researchers working on conservation technology. More information about the workshop can be found here.
Study to investigate influence of tree cover on carrion partitioning
Carrion plays an important role in maintaining ecosystem stability and provides feeding opportunities for numerous scavenger species around the world. Vertebrate scavengers are responsible for the large majority of carrion consumption and compete with each other for these resources. The aim of this study was to investigate the partitioning among avian and mammalian scavengers. We examined the influence of ambient tree cover, carcass openness, and repeated use of carcass provision sites on i) the relative and absolute number of scavengers, and ii) time-to-detection, time-to-first-scavenging and time-to-depletion by monitoring the exploitation of large ungulate carcasses with motion-triggered cameras. We found that the proportion of vertebrates scavenging from carrion and their ability to detect it were strongly associated with tree cover. Read the full article here.
Automate invasive species recognition for New Zealand camera trap images
Addax Data Science has been asked by the Department of Conservation (DOC) to develop a species recognition model to differentiate between 17 species or higher level taxons present in New Zealand. It was trained on a set of approximately 2 million camera trap images from various projects across the country.
The model has an overall validation accuracy, precision, and recall of 98%. When tested on an out-of-sample test set, the model scored 95%, 96%, and 94%, respectively. The model was designed to expedite the monitoring of New Zealand’s invasive species (deer, possum, pig, cat, rodent, and mustelid). It is published open-source and can be deployed through our camera trap analysis desktop application, AddaxAI.
Continuous real-time monitoring and deterring system for wolves
After 150 years of extirpation, the wolf has returned to the Netherlands as a protected species. The fragmented Dutch landscape and densely populated countryside, however, poses challenges. The high number of roads increases the risk of collisions, whereas the wolf poses a threat to the abundantly present livestock.
Addax Data Science investigates the feasibility of a 24-hour monitoring system to continuously check for the presence of wolves. This real-time detection system, capable of deterring the animals with sounds or light flashes, aims to reduce human-wolf conflicts. By addressing potential encounters and protecting both wolves and livestock, this initiative can contribute to the sustainable coexistence of humans and wolves in the Netherlands.
Database management system to track cash grant payments
Inclusion foundation contributes to eradicating extreme poverty worldwide by distributing cash grants via mobile money. Through their basic income projects, they want to help people directly and develop an effective, scientifically tested approach. In order to do so, they need a reliable automated method to manage the payment information.
We have created a software management system that can track transactions, expenses, participants, households, family relationships, statuses, phone numbers, and other community demographics. The system calculates conversion rates, transactions and destinations for all participants and keeps track of administration. The information is exported to the desired format to be sent to the financial department.
Species recognition model for the Namib Desert biome
Desert Lion Conservation is a small non-profit organisation dedicated to the conservation of desert-adapted lions in the Northern Namib. Their main focus is to collect important base-line ecological data on the lion population and to study their behaviour, biology and adaptation to survive in the harsh environment. They use this information to collaborate with other conservation bodies in the quest to find a solution to human-lion conflict, to elevate the tourism value of lions, and to contribute to the conservation of the species.
Addax Data Science has been asked to develop a species recognition model to differentiate between 25 species or higher level taxons. The model can be deployed through our camera trap analysis desktop application, AddaxAI. Learn more about the project here.
Desktop application for automatic species detection in camera trap analysis
AddaxAI is an application designed to streamline the work of ecologists dealing with camera trap images. It’s an open source AI platform that allows you to analyse images with machine learning models for automatic detection, offering ecologists a way to save time and focus on conservation efforts. AddaxAI is a collaboration between Addax Data Science and Smart Parks to support open-source projects in nature conservation.
The software has the open-source MegaDetector model incorporated, which can filter out images containing animals, people, and vehicles. It also supports the deployment of custom project specific species recognition models to be used in conjunction with MegaDetector.
