Field margins, floral plantings, and crops offering nectar and pollen to insects may contribute to ecosystem service delivery (pollination) and ecological intensification of agriculture. The application of machine learning on data collected using low cost RGB cameras represents a new opportunity to investigate, evaluate and optimize the attractiveness of such cropping systems to beneficial insects. In this demonstrator, we applied an innovative technology to detect and identify pollinators in the field and to automatize the detection and quantification of pollinators in addition to classical measurements. Furthermore, we developed and tested a new maize-based cropping system that includes phacelia or a commercial flowering strip mixture in a maize field (intercropping) to allow for a large area of connected fodder supply to pollinators, while aiming for minimized maize yield losses due to competition.
The video shows the field experiment and the observations in the year 2023. First results show that the maize-based intercropping system significantly increased insect abundance as compared to sole maize, and could potentially connect isolated areas, while providing maize yield. Our machine learning approach to non-invasively and automatically detect and count pollinators in RGB images has a high potential to evaluate cropping systems (e.g. flower strip seeds) or management impacts (e.g. weeding) on insect abundance.