A new publication finds that new image-based monitoring technology using deep learning (DL) algorithms can detect and classify small insects in complex scenes with unprecedented accuracy. The paper titled “Accurate detection and identification of insects from camera trap images with deep learning” is co-funded by MAMBO and co-authored by the project coordinator Toke Høye from Aarhus University in Denmark.
The reported decline of insect populations has increased the global demand for standardised insect monitoring data, which can be generated cost-efficiently and non-invasively using image-based monitoring. However, extracting ecological data from images for insects is more challenging than for vertebrate animals because of their small size and diversity.
To address this challenge, researchers have created a large annotated image dataset of functionally important insect taxa, including bees, hoverflies, butterflies, and beetles, across more than two million images recorded with ten time-lapse cameras mounted over flowers during the summer of 2019. The dataset was used to train and compare the performance of selected You Only Look Once (YOLO) deep learning algorithms, with the best performing model consistently identifying nine dominant insect species that play important roles in pollination and pest control across Europe. The model reached an average precision of 92.7% and recall of 93.8% in detection and classification across species.
Fig 1. Images of Sedum plants with one bumblebee (left) and honey bee (right) highlighted by red rectangles. The image size is 1920x1080 pixels and covers a field of view of approximately 35 x 22 cm.
The research achieved a detection and classification precision of over 90% across nine different types of insects. The open access image dataset contains over two million images featuring 29,960 annotated bees, hoverflies, butterflies, and beetles against various floral backgrounds. Furthermore, the study addresses the more challenging task of detecting multiple small insects in large images. Both the data and the models have been expanded and improved and are now publicly available.
An advanced non-destructive solution for automated in situ monitoring of insects, with good performance in detecting unfamiliar insect species is demonstrated through the research. Moreover, it provides a critical benchmark for future development and evaluation of deep learning models for insect detection and identification. This study is conducted with the vision that it will allow science on biodiversity monitoring to enter a new era of non-destructive studies on insect co-occurrence, phenology, and behaviour.
