Low Cost Machine Vision for Insect Classification

TL;DR

This work presents a low‑cost, open‑source multisensor system (KInsecta21) for automated insect monitoring that can be integrated with classical traps. Through standardized imaging with diffuse illumination and motion‑blur suppression, the authors demonstrate that high‑quality images support strong taxonomy‑level classification using standard CNNs (ResNet‑50, MobileNet) and a compact custom network, especially after cropping via segmentation. A U‑Net based semantic segmentation pipeline achieves robust bounding boxes (IoU ≈ 0.766), enabling effective automatic cropping and improved classifier performance on edge devices. The findings suggest practical applicability for field deployment and citizen‑science data collection, with future work including multi‑sensor fusion (e.g., wingbeat, environmental metadata) and few‑shot learning to handle rare species.

Abstract

Preserving the number and diversity of insects is one of our society's most important goals in the area of environmental sustainability. A prerequisite for this is a systematic and up-scaled monitoring in order to detect correlations and identify countermeasures. Therefore, automatized monitoring using live traps is important, but so far there is no system that provides image data of sufficient detailed information for entomological classification. In this work, we present an imaging method as part of a multisensor system developed as a low-cost, scalable, open-source system that is adaptable to classical trap types. The image quality meets the requirements needed for classification in the taxonomic tree. Therefore, illumination and resolution have been optimized and motion artefacts have been suppressed. The system is evaluated exemplarily on a dataset consisting of 16 insect species of the same as well as different genus, family and order. We demonstrate that standard CNN-architectures like ResNet50 (pretrained on iNaturalist data) or MobileNet perform very well for the prediction task after re-training. Smaller custom made CNNs also lead to promising results. Classification accuracy of $>96\%$ has been achieved. Moreover, it was proved that image cropping of insects is necessary for classification of species with high inter-class similarity.

Figures (8)

• Figure 1: Image of a European paper wasp taken with the Imaging Unit with zoomed-in characteristic details of body pattern, tabia-tarsus junction and wing details.
• Figure 2: Comparison of 3 wasp species. (Top) Images acquired from the imaging unit. (Bottom) Schematic representation according to Ripberger koch1993ripberger. From left to right: Vespula germanica, Vespa vulgaris, Polistes dominula
• Figure 3: Image acquisition unit (left) and principle of the insect arena (right). The red region indicates the light barrier in the center of the arena that starts the image recording.
• Figure 4: Distribution of the species in the total data set used for the experiments. Notice the long tail of the histogram which is typical for insect monitoring data.
• Figure 5: Example of inferring a bounding box: From left to right: original image, manually annotated segmentation mask (used as label for training the U-Net in a supervised context), U-Net predicted mask, bounding box inferred from the predicted mask.