STSM’s Experiences
Dmitry Schigel
Associate professor of Mycology, University of Helsinki;
Lead scientific officer, GBIF Secretariat / PhD / Entomology, mycology, metabarcoding, databases
GBIF Secretariat, Denmark
Host institute and country: LUOMUS – Finnish Museum of Natural History, Finland
Host researcher and/or team: Pasi Sihvonen
Time of the visit: 18 July – 8 August 2025
Testing the technology limits of time-lapse cameras in detecting saproxylic and fungicolous invertebrates, generating technical footage for image and motion recognition.
STSM was carried out in Finland in two sites – in Saimaa area near Puumala in SE Finland (field deployment) and in Helsinki (early footage seminar with the host staff). Field work took place during 12 field days using 5 cameras of 4 different types. This resulted in 52 unique deployments against 7 species of fungi and 3 species of myxomycetes, and uncountable insect visitation primarily of Coleoptera and of Diptera, with other invertebrates including mollusks, spiders, mites, and millipeds.
Funding for field work, technical footage, experience and lessons learn for the future work, feedback from the hosting institution
We are planning to resubmit a funding proposal for a larger scale project in Finland. LUOMUS has been very supporting in hosting and encouraging this work.
New grant schemas and flashy sites may look overwhelming triggering imposter syndrome from the first-time applicants. This should not happen – if the calls is open and suits your needs, apply. Do contact technical coordinators and subject matter expert in the WG – the proposal must help COST action own purposes.
Dr. Edina Török
HUN-REN Centre for Ecological Research, Hungary
Host institute and country: German Centre for Integrative Biodiversity Research (iDiv) Leipzig, Germany
Host researcher: Maximilian Sittinger
Time of the visit: 2 weeks (07/07/2025 to 07/18/2025)
As a pollination ecologist, I have always been interested in developing non-lethal, standardised monitoring methods for insect activity. Observing pollinators at all times of day is impossible, but an AI-powered camera trap offers a promising solution.
Through the Standardised European monitoring of plant-pollinator interactions (SEPPI) project, I had already worked with DIY camera traps to capture flower–pollinator interactions. However, I wanted to understand the deeper technical foundations of these systems, how do they actually work, and how could we improve them?
This curiosity led me to join the InsectAI COST Action, which supports interdisciplinary collaboration on AI for insect research. The STSM programme gave me the opportunity to visit Leipzig and work with Maximilian Sittinger at iDiv to take a step further: to build a “smart” bee hotel that uses real-time AI detection to monitor visiting bees.
During my STSM, we followed a structured and hands-on approach for developing and testing a bee detection model for use with artificial nesting structures.
Before the visit, I used DIY camera traps to collect thousands of images from my own reed insect hotel, capturing real-life insect activity. In preparation for model training, we also gathered additional image datasets: a publicly available dataset of a woody insect hotel from Roboflow, and photographs taken of a standardised board trap nest, kindly provided by Lasse Krüger from the Wild Bee Monitoring in Agricultural Landscapes project.
Once in Leipzig, I learned how to annotate all these images using CVAT, preparing well-structured and labelled datasets for model training.
Together with Maximilian Sittinger, we then trained separate YOLOv11 object detection models using Google Colab, one for each of the three insect hotel types: reed, woody, and board.
After training these individual models, we created a fourth “combined” model using a merged dataset that included images from all three hotel types. This mixed model showed a clear performance advantage: it achieved over 96% detection accuracy and was more robust across various datasets and conditions due to its higher generalization capability.
To test the practical application of this work, we converted the best-performing model into a deployable format and ran real-time detection tests using an OAK-1 AI camera. The model was able to detect insect activity in several real-world and controlled scenarios, demonstrating promising results. While it’s not yet perfect and requires further refinement, these initial tests show that real-time insect monitoring with AI cameras is a feasible and valuable approach for future ecological research. We are eager to continue developing and refining this method in the future.
Working with Maximilian Sittinger was a great experience, I learned a lot from him, and really appreciated the opportunity to collaborate in person. I had the chance to focus deeply on a single, hands-on technical project, and to learn every step of the AI model development pipeline. I also got to meet the Species Interaction Ecology research group and enjoy the beautiful city of Leipzig.
Maria Cholvi Simó
PhD Candidate, Faculty of Pharmacy, University of Valencia,
Host institute and country: The Cyprus Institute and CYENS Center of Excellence, Cyprus
Host researcher and/or team: Kelly Martinou
Time of the visit: 20/8/2025 to 1/9/2025
Being able to meet the project team face-to-face and see how people work in different countries is such a valuable experience. I seized the opportunity provided by the STSM to deepen my understanding of how different countries organize their vector monitoring programs. The project we are developing—an automated egg counting system for vectors—is a crucial tool to assist the authorities responsible for surveillance of vectors that may pose public health threats.
The objective of my STSM was to assist the computer scientists in developing the project, both by performing computational tasks and providing information about the vector’s life cycle that could help refine the protocols. I also helped demonstrate the potential applications of the project and facilitated the coordination of the multidisciplinary team. Additionally, during the stay, we reviewed scientific literature to support the project and conducted a KAP (Knowledge, Attitudes, and Practices) survey aimed at effectively translating our knowledge into the daily work of the laboratories.
Throughout life, we’re always learning, and often the most valuable lessons come from the people we meet along the way. STSMs offer a unique chance to connect with researchers from different countries, share ideas, and learn from their perspectives. For me, working on a project to automate the egg counting process not only allowed me to contribute my expertise but also helped me see how artificial intelligence can be applied to improve the monitoring of disease-carrying vectors across Europe. If you’re passionate about these topics, I encourage you to take advantage of COST actions to develop your ideas and collaborate with others in the field.
I am really grateful to the COST action for this opportunity, and to Kelly Martinou for the incredible hospitality she extended to me in Cyprus, along with her entire team and the CYENS team.
Petr Kriz
Researcher/PhD student/ computer vision
Brno University of Technology
Host institute and country: Universidad de Las Palmas de Gran Canaria
Host researcher and/or team: Carlos M. Travieso-González
Time of the visit: 25/06/2025 – 09/07/2025
I applied for an STSM at InsectAI because the project matches my interest in applying computer vision to ecological and real-world scenarios. The interdisciplinary focus, combining AI and biodiversity, felt both meaningful and technically challenging. I saw it as a great opportunity to contribute to a growing community and to collaborate internationally. I was also curious to gain hands-on experience with real data beyond lab conditions. The connection to lightweight AI models was particularly relevant to my PhD research.
The main purpose of my STSM was to support the development and evaluation of computer vision models for detecting and classifying butterflies.
I worked on collecting and filtering image data, with special care taken to reduce motion blur caused by butterfly movement. I manually labeled video frames using the CVAT tool and organized the data for training object detectors and classifiers. I tested several lightweight models from the YOLO and timm libraries.
The STSM gave me practical experience with dataset curation and model evaluation in a real-world context. I deepened my understanding of lightweight deep learning models and improved my skills in data preparation and annotation. I also had the chance to work in a collaborative and international research environment, which was both inspiring and motivating.
We are currently preparing a joint publication that includes the results of the detector and classifier evaluation. The datasets and code shared during the STSM will also serve as a basis for future experiments by both teams.
Don’t hesitate to apply — it’s a unique opportunity to grow technically, professionally, and personally. Enjoy the people, the culture, and the new perspectives along the way!
Andrey Malyshev
Postdoc/PhD/Plant and insect phenology
University of Greifswald (Germany)
Host institute and country: Aarhus University (Denmark)
Host researcher and/or team: Pr. Quentin Geissmann
Time of the visit: 02/06/2025 to 13/06/2025
The possibility to visit a research group specializing in AI-assisted insect monitoring was appealing as I am new in using such methodology but require such skills in my planned research on insect and flower phenology under climate change.
The purpose was to develop a machine learning method to detect wild bee emergence from insect hotels and from their cocoons.
Briefly describe the work you carried out and any results obtained:
An analytical method was successfully developed to detect the emergence timing and sex of masonry bees in different environments (cocoons/bee hotels). The model metrics revealed very good closed and open cocoon detection (82% precision, 99% recall, and mAP@0.5 of 86% for closed; 82% precision, 99% recall, and mAP of 86% for open). Male bee detection was ok to good (93% precision, 63% recall, and mAP@0.5 of 76%). Female bee detection was poor, likely due to the low number of training images of female bees to train the model (18% precision, 33% recall, and mAP@0.5 of 30%) . The research network of the applicant was extended and optimized by the ability to objectively quantify insect emergence in different locations by time-lapse video analysis.
A cooperation was established with the host to a) help the applicant complete the model training and validation at the home university (online) and b) carry out a joint insect overwintering experiment in the futur,e where a bee hotel filled with different bee species will be sent to the host, who will record their emergence with a time-lapse camera in the spring. The host location will be one of several European locations where such bee hotels will be sent to record insect emergence after different winters. The experiment will explore the influence of winter temperature on bee survival and phenology in the spring. Each participating partner will upload the time-lapse videos online, and I will be able to analyse them using the created model.
Close communication was agreed between the University of Greifswald and the University of Aarhus via online services such as GitHub and online meetings so that questions arising on automatic image detection can be answered on an ongoing basis.
Get in close contact with researchers who employ methods you envision using and develop a plan that would make a meeting mutually beneficial (planning of joint experiments after learning new methodology). Develop the research stay activities together with the host/host group.
Manisha Sanjay Sirsat
Senior Artificial Intelligence Researcher
InnovPlantProtect, Portugal
Host institute and country: University of Perugia, Italy
Host researcher and/or team: Maria Cristina Pinotti, Lorenzo Palazzetti, Francesco Betti Sorbelli
Time of the visit: 2 weeks (31.08.2024 to 14.09.2024)
I applied for a STSM of the InsectAI to collaborate with the team at the University of Perugia (UNIPG) in Italy, who are experts in using computer vision techniques for pest identification. Their work closely aligns with my research interests in AI-driven agricultural solutions. The main goal of my STSM was to integrate three AI-based models developed by InnovPlantProtect (InPP)—the trap-image based leafhopper detection model, the vine detection model, and the leafhopper symptoms detection model into a unified pipeline to improve the monitoring and management of the green leafhopper (Jacobiasca lybica), a key pest in Mediterranean vineyards. Additionally, I aimed to evaluate this pipeline’s adaptability for the Brown Marmorated Stink Bug (BMSB).
The work conducted during this STSM was highly interdisciplinary, combining expertise in AI, drone technology, and agricultural pest management. In addition to the proposed activities, we engaged in new discussions, including an in-depth exchange of project knowledge between the host and home institutions. We updated InPP’s previously developed practical framework by integrating the AI models into a single pipeline, aiming to adapt the pipeline for multi-species pest detection. The outcome of the STSM shows the successful integration of AI models for vine-specific leafhopper symptom detection. Currently, this AI pipeline is effectively working for tree-specific detection of pest symptoms.
I gained valuable interdisciplinary experience, which has led us to plan further collaboration on the identification of the BMSB in Portugal. I highly recommend that young researchers and innovators apply for a STSM with InsectAI, as it was an outstanding opportunity for me to collaborate with leading researchers and enhance my skills in AI and pest management technologies.
FERHAT SADAK
Assistant Professor, Ph.D., Artificial Intelligence
Bartin University, Türkiye
Host institute and country: Naturalis Biodiversity Centre, Netherlands
Host researcher and/or team: Dr Chantal HUIJBERS
Time of the visit: 19/08/2024 to 06/09/2024
I applied to InsectAI to address the class imbalance in insect population datasets and contribute to biodiversity monitoring. The opportunity to collaborate with experts at Naturalis aligned perfectly with my research interests.
The STSM aimed to improve insect population monitoring by evaluating techniques to handle class imbalance using the EfficientNetv2m model, focusing on enhancing classification accuracy and generalization.
I tested various class imbalance techniques, including SMOTEENN and SMOTETomek, which significantly improved test accuracy (up to 0.751). These methods proved most effective for balancing datasets and improving model performance.
Results: I gained technical expertise in handling imbalanced datasets, collaborative experience with leading researchers, and insights into biodiversity monitoring challenges. Planned follow-ups include collaborations with Naturalis, a webinar to share findings, and potential publications to advance biodiversity monitoring research.
I recommend ensuring your research aligns with InsectAI’s goals, engage actively with collaborators, and use the opportunity to develop skills and build long-term partnerships.
FEVZIYE HASAN
Entomology Curator and Insect Ecologist
Evolutionsmuseet, Uppsala University, Sweden
Host institute and country: Institute for Ecoscience, Aarhus University, Denmark
Host researcher and/or team: Professor Toke Høye
Time of the visit: 12 – 16 August 2024
I am inspired by the rapid advances in technology and AI, which can profoundly improve our understanding of ‘dark taxa’ like insects and contribute to big data in entomology, insect ecology and overall biodiversity science. I think that these new technologies will deliver data that will be vital in decision-making to find urgent solutions to the biodiversity crisis. Applying for the STSM at InsectAI was a perfect opportunity for me to engage with and learn from leading experts in this exciting field.
Collecting insect biodiversity data is traditionally time-consuming due to the challenges of fieldwork and sample processing. As novel monitoring technologies emerge, they promise to generate unprecedented amounts of insect data which will be important for understanding biodiversity, species distributions, ecological patterns, and responses to environmental change over time. However, without proper metadata standards established from the beginning, this wealth of information could be inaccessible or difficult to use effectively. To address this, my STSM focused on developing FAIR metadata standards (Findable, Accessible, Interoperable, and Reusable) for automated insect camera traps. One of our aims was to see how datasets from camera traps could be integrated with global databases like the Global Biodiversity Information Facility (GBIF), increasing their value for research, conservation, and policy-making. By proactively developing these standards, we can ensure new monitoring technologies maximise their data delivery to open infrastructures at scale.
This project allowed me to combine my knowledge of entomology curation and insect ecology with my interest of technology and AI. A highlight for me during the STSM was visiting the GBIF Secretariat in at the University of Copenhagen, where we met with Tim Roberts and Cecilie Svenningsen to discuss how metadata standards would work with GBIF’s infrastructure. As a result of the STSM, I received an invitation to the InsectAI writing retreat in Bergen, Norway, where I had the opportunity to collaborate with developers on guidelines for extending the camtrapDP metadata standard for insects. This experience greatly expanded my understanding of the development process and also made me realise how important it is for an entomologist to ask “silly” questions to developers, as these questions often lead to deeper understanding of vocabularies and clearer communication.
Learning from others is one of the greatest joys in science, and joining the InsectAI COST Action is a fantastic opportunity to connect with brilliant colleagues and collaborate across disciplines. If you’re considering applying for a STSM, my advice is to start by identifying the questions or challenges that ignite your curiosity. The initiative offers opportunities from data analysis and societal impacts to species identification, don’t be afraid to get creative and embrace the chance to explore new ideas. Thank you to InsectAI for this great opportunity, and to Toke Høye for warmly welcoming me into his lab, introducing me to the team, and hosting me for the week at Aarhus University.