A Honey Bee In-and-Out Counting Method Based on Multiple Object Tracking Algorithm.

Simple Summary: The honey bee (Apis mellifera) is of great significance to both the ecological environment and human society. Systematic monitoring of honey bee activities is a useful method for determining the colony condition and can facilitate pre-emptive measures against detrimental conditions. It is relatively easy to digitally monitor and analyze the incoming and outgoing actions of bees in current research. Cameras within beehives were used to collect the activity data of bees. Our work mainly involved the application and evaluation of algorithms and methods. Object detection algorithms and multiple object tracking algorithms were used to track the honey bees' incoming and outgoing actions. Based on the object detection algorithm and the multiple object tracking algorithm, in-and-out counting methods (the single-line method and the box method) were designed. After three levels of evaluation, the results showed that the in-and-out counting model consists of YOLOv8m, OC-SORT, and the box method, which achieved the best performance, and the model can be used to analyze the activity and colony condition. The bees' in-and-out activity is an essential indicator of bee behavior and can be useful to determine the colony condition. The traditional counting of bees' in-and-out activity mainly relies on manual work, which is time-consuming and inefficient. Therefore, there is a need to devise some more efficient alternative methods. For this purpose, the present study was conducted. This study proposed a bee in-and-out activity counting method based on video detection, including object detection, multiple object tracking (MOT), and an in-and-out activity counting algorithm. Images and video stream data were captured using a camera (Huiboshi model X20, Shenzhen Huiboshi Technology, Shenzhen, China) in a smart beehive. Two bee in-and-out counting methods, single-line and box methods, were designed. After three levels of evaluation, the results showed that the best counting model consists of YOLOv8m, OC-SORT, and the box method, which achieved F1in of 91.49% and F1out of 89.08%. The model can track and count honey bees in a complex environment and can be used to analyze their activity and colony condition. [ABSTRACT FROM AUTHOR]