The aim of this project is to develop a user-friendly hardware and software system for the automated generation of augmented training data and the detection and identification of insects in images. The system will significantly reduce the time and effort required for the manual processing and analysis of insect samples, enabling more efficient and large-scale biodiversity monitoring studies.

Key tasks include:

• Creating a workflow for the automatic generation of synthetic training data using 3D models of target species.
• Developing workflows and tools for the automated image acquisition of insect samples.
• Designing and optimizing deep learning methods for insect detection and species identification.
• Adapting an existing, web-based software system for the user-friendly application of AI models for automated image analysis of insect biodiversity surveys.

Insect populations are declining at an alarming rate due to factors such as climate change and intensive land use. Monitoring insect biodiversity is essential to understand these changes, but current methods are time-consuming and resource-intensive. Traditional approaches rely on manual sorting and identification of insect samples, which is impractical for large-scale studies.

Recent advancements in deep learning and computer vision offer promising solutions, but the generation of high-quality training data remains a challenge. Synthetic data, particularly from 3D models, can help bridge this gap by forming diverse and scalable training datasets. Building on the results of a previous digitization project (DISC3D), the 3DTRAB project aims to develop innovative tools for automated insect identification, utilising thousands of insect 3D models from museum collections.

GFBio contributes to the project by developing and extending the software infrastructure for the management and application of synthetic training data and AI models. Specifically, GFBio will:

• Enhance the existing data portal for 3D models of insects to include functionalities for managing and searching synthetic training data.
• Develop a web-based interface for researchers to apply AI models to their insect images without requiring extensive IT expertise.
• Ensure the long-term availability of the data and the sustainability of the developed tools.

The 3DTRAB project is a collaborative effort involving the following partners:

Universität Marburg (UMR): Contributing expertise in environmental informatics, deep learning, and software development.

TU Darmstadt (TUDa): Providing the DISC3D scanner technology and expertise in 3D modeling and synthetic data generation.

GFBio e.V.: Developing the software infrastructure, ensuring metadata standardisation and data availability according ot the FAIR data principles.