Connecting the public: This mini program aims to promote the mainstream of biodiversity conservation by desensitizing current monitoring data in the industry and designing low threshold interactions for the traditional data labeling process. This allows the public to participate in the training process of biodiversity models in a more accessible and intuitive way through the mini program. On the one hand, the public can enjoy and learn about the most authentic protection monitoring images through the form of "playing games"; On the other hand, the power of the public can be utilized to continuously train a universal model of biodiversity, achieving the goal of citizen science in the process.
Through product design, 'Wild Friends' breaks down the process of annotating and verifying institutional data into tool based tasks, reducing the initial training costs of institutions. With simple guidance, volunteers or the general public can complete basic annotation content.
The first step is to check for the presence of animals (manually identified or judged by AI);
Step two, estimate the number of animals (manually determined);
Step three, select animals (manually or through AI evaluation of selection accuracy);
Step four, identify the name of the animal (manually selected or judged by AI);
Step five, randomly allocate cross validation in the background. Ensure the accuracy and consistency of data.
 

AI species recognition: This product uses AI recognition as the underlying technology, with endangered species as the core recognition object. It trains a large biodiversity recognition model that can support monitoring of mountains, rivers, forests, fields, lakes, grasses, and sands systems. The model is free and open to public welfare organizations dedicated to biodiversity conservation, such as research institutes, conservation organizations, and individuals. The reason why "wild friends" are so powerful is because they have a powerful "engine": YOLO World.
As the underlying universal model of 'wild friends', its primary characteristic is strong learning ability. It has powerful multimodal zero sample recognition and few sample recognition capabilities, which means it can quickly identify animal location regions and species information of multiple species through a small number of samples. For example, to recognize a new species, traditional models require thousands of photos and several days of training; YOLO World only requires a small number of photos and training iterations to achieve rapid adaptation.
Secondly, it has a high degree of tolerance. No longer limited to training and prediction of specific species, it has strong open vocabulary recognition ability and zero sample recognition ability, and can accurately identify and locate untrained species. For example, traditional models can only recognize trained species such as tigers and antelopes; The new model can also recognize snow leopards and foxes simultaneously - even if it has never trained these two animals before.
Another advantage of "wild friends" is that they spend less money. Common AI models heavily rely on high-performance acceleration cards, which result in high costs for both hardware environment and maintenance operations.
 

To preserve natural resources, artificial intelligence must be introduced to preserve them, and automation must be used to preserve environmental diversity by linking to the use of the Internet today, which is everywhere, controlling it, and following up. It was made into a real reserve and controlled using connected surveillance cameras. Transporting animals to a safe environment protected by surveillance cameras to reduce poaching.

Advancements in on-board technologies and AI integration hold great potential to further enhance the existing drone-based crocodilian monitoring method. Improvements in drone hardware, such as hybrid models with extended flight times and enhanced camera resolutions, allow for broader habitat coverage and the capture of more detailed imagery in complex environments. Integrating artificial intelligence (AI) represents a significant opportunity to streamline image analysis by automating crocodile detection and size estimation using allometric models. These AI-driven enhancements could provide near real-time data processing, reducing reliance on time consuming manual analysis.

This improvements are currently under development. We conducted an experimental study in Cameroon in April 2025 with students and young researchers from the University of Ngaoundéré and local NGOs, using drones equipped with thermal cameras and searchlights, and including AI-assisted automated data processing.

This building block on capacity building on local stakeholders including Indigenous including Peoples and Local Communities (IPLCs) to operate drones, enabling them to take active roles in conservation. 

Ease of Use of the method devellopted:

1. Minimal Technical Skills Required:
   Users only need basic training in drone operation and measurement extraction from high-resolution images. The process is straightforward:
   • Fly the drone following the standardized flight protocol.
   • Marke crocodilians on overhead images.
   • Measure the visible head length using accessible image analysis tools (e.g., ImageJ, QGIS).
   • Apply the corresponding allometric equation or lookup from pre-prepared tables (abaques) for total length estimation.
2. Readily Adaptable:
   The framework uses easy-to-read tables (abaques), making it accessible to both specilialists and non-specialists for operators can quickly apply the method without requiring advanced scientific expertise.
3. Accessible Equipment:
   The approach relies on consumer-grade drones and widely available software, ensuring affordability and reducing barriers to adoption.

Why It’s Effective:

The framework’s simplicity, scalability, and reliability make it ideal for diverse contexts, from remote wetlands to urban-adjacent habitats. It empowers a broad range of users to generate scientifically robust data.

The allometric framework is a non-invasive tool designed to estimate the total body length of crocodilians based on the measurement of their head length, captured through high-resolution drone imagery. By leveraging established species-specific head-to-body length ratios, this method eliminates the need for physical capture or handling, reducing risks for both researchers and wildlife. Validated for 17 of the 27 crocodilian species, the framework allow to provides reliable demographic data essential for population monitoring and conservation management.

The framework uses easy-to-read tables (abaques), making it accessible to non-specialists, operators can quickly apply the method without requiring advanced scientific expertise.

Understanding the demographic structure is vital for wildlife research and conservation. For crocodylians, accurately estimating total length and demographic class usually necessitates close observation or capture, often of partially immersed individuals, leading to potential imprecision and risk. Drone technology offers a bias-free, safer alternative for classification. This study evaluated the effectiveness of drone photos combined with head length allometric relationships to estimate total length, and propose a standardized method for drone-based crocodylian demographic classification. 

This building block establishes standardized flight parameters for effective crocodilian monitoring