External links
Abstract
Honey bees have a crucial role in pollination, and this paper presents a simple, non-invasive system for pollen-bearing honey bee detection in surveillance video recorded at the hive entrance. The system is intended as part of a more complex tracking and counting pipeline for remote pollination monitoring. It runs in real time on embedded systems co-located with the hive, using background subtraction, color segmentation, and morphology to segment bees. Classification into pollen-bearing and non-pollen-bearing bees is performed using a nearest-mean classifier with color-variance and eccentricity features, achieving an 88.7% correct classification rate with 50 training images per class.
Relevancy to Gratheon
This older paper fills an important historical and practical gap for Gratheon's gate-tracker ideas: pollen-load detection can turn entrance video from a generic traffic counter into a pollination-service and forage-quality signal. Its lightweight embedded pipeline is also relevant for low-power field deployment, especially when deep networks are too expensive or unnecessary. Gratheon can use it as a baseline for comparing modern CNN/YOLO pollen classifiers against simpler color-and-shape methods on edge hardware.