COS 40-4 - Sticky Pi 2.0, a deep learning-based smart insect trap for all ecologists

Abstract: Traditionally, the study of insect ecology and behaviour has relied on the manual collection, identification and direct observation of wild specimens, which is extremely tedious and requires expert knowledge. However, in recent years, the development of technologies such as remote sensing and deep learning has led to increasingly compelling and complementary alternatives to manual labour. Together, these methods promise to increase both the throughput and the content of entomological data collection. However, we are still in the early stages of the transition to computational entomology. In particular, the adoption of such tools has been limited by the considerable gaps between proof-of-principle prototypes (developers) and tools that can effectively serve the general community (practitioners). Therefore, it has become a priority to turn our computational entomology methods into robust, applicable and usable solutions.

Last year, we released the Sticky Pi, an open-source prototypical smart insect trap based on the Raspberry Pi. By imaging trapped insects every 20 minutes, we recorded the high-frequency variations in the activity and abundance of wild insect communities, and ultimately characterised their "chronoecology", an aspect that is usually overlooked. Departing from a "lab-centric DIY resource", we have since significantly improved the performance reproducibility and usability of our methods. Specifically, we have improved the device (optics, power management, sensors, ...), developed a novel phone application to wirelessly retrieve the data, have written complete documentation and have deployed a cloud platform to aggregate and visualise data from dozens of smart traps. Here, we present Sticky Pi 2.0 and show how it addresses critical practical limitations in computational entomology.