Biology subjects

Stefan, V.

Publications and source records attributed to Stefan, V..

3 recordsLinked to original sources

bioRxiv2025

Utilizing CNNs for classification and uncertainty quantification for 15 families of European fly pollinators

Pollination is essential for maintaining biodiversity and ensuring food security, and in Europe it is primarily mediated by four insect orders (Coleoptera, Diptera, Hymenoptera, Lepidoptera). However, traditional monitoring methods are costly and time consuming. Although recent automation efforts have focused on butterflies and bees, flies, a diverse and ecologically important group of pollinators, have received comparatively little attention, likely due to the challenges posed by their subtle morphological differences. In this study, we investigate the application of Convolutional Neural Networks (CNNs) for classifying 15 European pollinating fly families and quantifying the associated classification uncertainty. Our dataset comprises a wide range of morphological and phylogenetic features, such as wing venation patterns and wing shapes. We evaluated the performance of three state-of-the-art CNN architectures, ResNet18, MobileNetV3, and EfficientNetB4, and demonstrate their effectiveness in accurately distinguishing fly families. In particular, EfficientNetB4 achieved an overall accuracy of up to 95.61%. Furthermore, cropping images to the bounding boxes of the Diptera not only improved classification accuracy but also increased prediction confidence, reducing misclassifications among families. This approach represents a significant advance in automated pollinator monitoring and has promising implications for both scientific research and practical applications.

ecology↗

bioRxiv2024

Perceived biodiversity: is what we measure also what we see and hear?

1. Biodiversity is crucial for human health and well-being. Perceived biodiversity - peoples subjective experience of biodiversity - seems to be particularly relevant for mental well-being. 2. Using photographs and audio recordings of forests that varied in levels of species richness, we conducted two sorting studies to assess how people perceive visual and acoustic diversity and whether their perceptions align with measured tree and bird species richness ( actual diversity). Per study, 48 participants were asked to sort the stimuli according to any similarity-based sorting criteria they liked ( open sorts) and perceived diversity ( closed sorts). 3. The main perceived visual forest characteristics identified by participants in the open visual sorts were vegetation density, light conditions, forest structural attributes and colours. The main perceived acoustic forest characteristics identified in the open acoustic sorts comprised bird song characteristics, physical properties such as volume, references to the time of day or seasonality and evoked emotions. 4. Perceived visual and acoustic diversity were significantly correlated with actual tree and bird species richness, respectively. Notably, the relationship was twice as strong for the acoustic sense. The acoustic sense may thus be crucial to obtain a more thorough understanding of perceived biodiversity. 5. We further computed several visual and acoustic diversity indices from the photos and audio recordings, e.g. colourfulness or acoustic complexity, and assessed their relevance for perceived and actual diversity. While all acoustic diversity indices were significantly associated with perceived acoustic diversity and bird richness, we could not identify a visual diversity index that captured perceived visual diversity and tree richness. 6. Our results suggest that people can perceive species richness. Our identified visual and acoustic forest characteristics may help to better understand perceived diversity and how it differs from actual diversity. We present acoustic diversity indices that quantify aspects of perceived and actual acoustic diversity. These indices may serve as cost-efficient tools to manage and plan greenspaces to promote biodiversity and mental well-being.

ecology↗

bioRxiv2024

Utilising affordable smartphones and open-source time-lapse photography for monitoring pollinators

Monitoring plant-pollinator interactions is crucial for understanding factors that influence these relationships across space and time. While traditional methods in pollination ecology are time-consuming and resource-intensive, the growing availability of photographic technology, coupled with advancements in artificial intelligence classification, offers the potential for non-destructive and automated techniques. However, it is important that the photographs are of high enough quality to enable insects to be identified at lower taxonomic levels, preferably genus or species levels. This study assessed the feasibility of using smartphones to automatically capture images of insects visiting flowers and evaluated whether the captured images offered sufficient resolution for precise insect identification. Smartphones were positioned above target flowers from various plant species to capture time-lapse images of any flower visitor in urban green areas around Leipzig and Halle, Germany. We present the proportions of insect identifications achieved at different taxonomic levels, such as order, family, genus, and species, and discuss whether limitations stem from the automated approach (e.g., inability to observe distinguishing features in images despite high image quality) or low image quality. Practical recommendations are provided to address these challenges. Our results indicate that for bee families, nearly three quarters of all cases could be identified to genus level. Flies were more difficult, due to the small size of many individuals and the more challenging features needed for identification (e.g., in the wing veins). Overall, we suggest that smartphones are an effective tool when optimised by researchers. As technology continues to advance, smartphones are becoming increasingly accessible, affordable, and user-friendly, rendering them an appealing option for pollinator monitoring.

ecology↗