Automatic Flower and Visitor Detection System

Dat Thanh Tran, Toke Thomas Høye, Moncef Gabbouj, Alexandros Iosifidis

Abstract: The visit patterns of insects to specific flowers at specific times during the diurnal cycle and across the season play important roles in pollination biology. Thus, the ability to automatically detect flowers and visitors occurring in video sequences greatly reduces the manual human efforts needed to collect such data. Data-dependent approaches, such as supervised machine learning algorithms, have become the core component in several automation systems. In this paper, we describe a flower and visitor detection system using deep Convolutional Neural Networks (CNN). Experiments conducted in image sequences collected during field work in Greenland during June-July 2017 indicate that the system is robust to different shading and illumination conditions, inherent in the images collected in the outdoor environments.