For wildlife rescue and species conservation

Free Artificial Intelligence

Want to detect wildlife and ground-nesting birds in real time with your drone? It's possible thanks to our free AI model for the Matrice 4T, 4TD, and 400RTK. Detections are automatically displayed on the RC Plus 2 during flight and can be marked and transferred to our POI app.

The latest beta version of our AI model is available for download now.

Download instructions

What is AI?

Artificial intelligence – AI for short – is a term that has become increasingly prevalent in recent years and is influencing more and more products in our daily lives. It is intended to simplify and accelerate processes, making our lives easier and more efficient.

At thermal DRONES, we have been working for years on an AI model for wildlife conservation. We have made it our mission to contribute to reducing animal suffering through the most efficient workflows and (partially) automated processes possible. For example, by making the best possible use of the short morning hours when searching for fawns.

But what exactly is our AI? In short: Our AI is a complex statistical model that recognizes comparable patterns from training data in new data.

This means that, based on thousands of images, it works out how the pixels of a fawn are represented in a meadow, and new images are then searched for these patterns. The model then indicates the probability with which a pattern corresponds to the already known patterns.

Darstellung der Kameraansicht einer Matrice 4T mit aktivierter KI-Detektion
The AI model can already reliably detect clutches of eggs from a height of 35 m.

Challenges and limitations

The model was trained using images from fawn rescue operations; that is, the images were captured during very specific types of flights under highly defined conditions. Therefore, the model only functions correctly when used under the right operational conditions with the correct flight parameters and settings.

Most AI models—unlike humans—cannot contextualize information because they operate purely on probabilities and lack an understanding of what they are detecting. It is therefore crucial to verify that the operating conditions match the patterns being sought.

Danger!

The use of AI leads to a loss of competence, motivation, and attention when analyzing images. This can result in animals being overlooked. Furthermore, our model is still in the beta phase and may contain errors. Therefore, the provided model is currently intended only for testing and learning purposes, not for independent practical application!

Artificial intelligence directly on the drone

Mit der DJI Matrice 4T hat DJI erstmals einen OnBoard-KI-Rechner in einer Drohne verbaut, über den man erstellte KI-Modelle lokal und ohne externe Ressourcen laufen lassen kann. Auch die Folgemodelle aus der Matrice-Reihe (Matrice 4(T)D & Matrice 400 RTK) kommen von Haus aus mit eingebauten KI-Rechnern, sodass unser Modell ohne Probleme auf ihnen installiert werden kann.

A joint project for the future of wildlife rescue

The AI model we provide is based on years of collaboration with our customers, who have kindly provided us with image data from several years of operations in the areas of fawn rescue and breeding bird protection. This resulted in a dataset of more than 100,000 infrared and color images, which could be used as training data for the AI model. At this point, we would like to express our sincere thanks to everyone who contributed to the development of the model! Without the close exchange and collaboration, such a comprehensive dataset would hardly have been possible. Based on this, we are already able to provide a very precise and reliable artificial intelligence that is continuously trained and improved with more data.

Support is welcome!

To make the model even more accurate and reliable, we would still greatly appreciate receiving photos (thermal and color images). Images can be uploaded directly via the POI Cloud interface, or you can contact us about this. Please refer to our guide for information on the correct image settings.