Tracking all members of a honey bee colony over their lifetime

Abstract: Computational approaches to the analysis of collective behavior in social insects increasingly rely on motion paths as an intermediate data layer from which one can infer individual behaviors or social interactions. Honey bees are a popular model for learning and memory. Previous experience has been shown to affect and modulate future social interactions. So far, no lifetime history observations have been reported for all bees of a colony. In a previous work we introduced a tracking system customized to track up to $4000$ bees over several weeks. In this contribution we present an in-depth description of the underlying multi-step algorithm which both produces the motion paths, and also improves the marker decoding accuracy significantly. We automatically tracked ${\sim}2000$ marked honey bees over 10 weeks with inexpensive recording hardware using markers without any error correction bits. We found that the proposed two-step tracking reduced incorrect ID decodings from initially ${\sim}13\%$ to around $2\%$ post-tracking. Alongside this paper, we publish the first trajectory dataset for all bees in a colony, extracted from ${\sim} 4$ million images. We invite researchers to join the collective scientific effort to investigate this intriguing animal system. All components of our system are open-source.

• The paper introduces a two-step algorithm combining SVM and random forest classifiers to reduce incorrect ID rates from 13% to 1.9%.
• The methodology tracks around 2000 bees over ten weeks, achieving near-complete trajectories in 70.4% of cases with high recall and precision.
• The approach leverages affordable hardware and advanced computational techniques, paving the way for in-depth, long-term analysis of honey bee behavior.

Tracking Honey Bee Colonies Over a Lifetime

Introduction

The paper presents a comprehensive approach to tracking every member of a honey bee colony over an extended period, leveraging computational models to decode and track bee markers despite potential error-prone detections. Honey bees, known for complex social behaviors and communication, pose a significant challenge in terms of individual identification and long-term tracking due to their numbers and frequent hive departures. This paper builds upon prior work to overcome these hurdles by implementing a sophisticated two-step algorithm that enhances marker decoding accuracy, significantly reducing incorrect ID decodings from approximately 13% to around 2%.

Methodology

The tracking methodology employed is designed to manage the intrinsic challenges of tracking within dense honey bee colonies, where conventional human tracking methods fall short. The core approach involves:

1. Recording Setup: Using affordable hardware without error-correction bits, the system tracks around 2000 individually marked bees over ten weeks. The markers are localized and decoded through convolutional neural networks, resulting in a high recall and precision rate.

1. Two-Step Tracking Algorithm:
   • Step 1 - Linking Consecutive Detections: This involves creating reliable short tracklets by connecting consecutive detections using a Support Vector Machine (SVM). This step takes into account Euclidean distance, angular orientation, and Manhattan distance of ID probabilities among detections.
   • Step 2 - Merging Tracklets: Building upon the tracklets, a random forest classifier merges these into longer trajectories, enabling gap handling for more robust long-term tracking. Features such as tracklet motion vectors, orientation differences, and ID probability confidences form the predictive basis for this step.
2. ID Assignment: The final step involves determining the bee's ID by aggregating and binarizing ID probabilities across a track, thus refining ID accuracy without relying on error correction codes.

Results

The system achieves notable enhancements in ID accuracy and tracking reliability. After enabling the two-step tracking methodology, the incorrect ID rate decreases drastically, with only 1.9% of detections marked incorrectly, almost reaching the theoretical lower bound of 0.6% as determined by ideal ground truth tracking. Moreover, it constructs virtually complete tracks for 70.4% of cases, approximating near-perfect tracking accuracy. These results were validated against manually labeled datasets, ensuring robustness across varying hive activity and environmental conditions.

Discussion

The proposed solution of a multi-step machine learning-based tracking algorithm represents a substantial leap toward accurately monitoring individual interactions and behaviors in honey bee colonies. This success, achieved without expensive hardware, underscores the potential for further extending such approaches to other densely populated animal studies. It opens venues for enriched analysis of social behaviors and network interactions within colonies by providing unprecedented long-term data. The research invites further enhancements such as integrating knowledge of marker postures and dynamic ID reliability weighting to bolster system performance further.

Conclusion

In conclusion, this work offers a pivotal advancement in tracking honey bee colonies, facilitating in-depth behavioral studies previously hindered by technical and logistical limitations. The system's contributions extend to reducing the reliance on costly equipment and furnishing extensive datasets critical for studying individual and collective behaviors in social insects. The authors hope to catalyze collaborative research endeavors to probe the expansive field of honey bee behavioral ecology and collective intelligence.