AI assisted analysis of sticky-traps as monitoring tool for localized insect biodiversity and biomass

  • Ate Boerema (Speaker)
  • Rutger Boersma (Contributor)
  • Arjen Strijkstra (Contributor)

Activity: Talk and presentationAcademic presentationAcademic

Description

Background: Insects are a diverse and relevant group of animals in the ecosystem. They are fast responders to changes in, management of, the environment and provide important ecosystem services. Localized monitoring of small scale (functional) biodiversity with sticky-traps is easy and cheap and can be very informative.
We developed a method to automatically quantify insects and biomass on yellow and blue sticky-traps. Sticky-traps are digitized using a flat-bed scanner (easily accessible) or in high quality on our sticky-trap-imager. An object detection algorithm (YOLOv7) was trained on >400 annotated traps, resulting in good automatic detection of insects (mAP@0.5 = 0.96; mAP@0.5:0.95 = 0.7). Subsequently a segmentation algorithm (U-NET) was trained (Dice/F1 value = 0.88) to detect body-contours, from which length of insects is calculated and biomass estimated. Insects are annotated the nearest attainable taxonomic unit in our web accessible labeling platform. Based on these labels a classification algorithm is in development. We have so far quantified thousands of sticky-traps with this method. The method is capable of describing differences in small scale biodiversity in rural, agricultural and urban settings and is sensitive enough to describe differences in small scale insect biodiversity between crops, and in different urban areas. It is a very useful tool that can be deployed by both experts and farmer- or citizen-scientists.
Period14-Feb-2024
Event titleNetherlands Annual Ecology Meeting 2024
Event typeConference
Conference number16
Organiser Netherlands Ecological Research Network (NERN)
LocationLunteren, NetherlandsShow on map
Degree of RecognitionNational