應用深度學習追蹤蜜蜂進出巢活動與花粉計數

Abstract

近年來，受氣候變遷和環境污染影響，蜜蜂養殖的年損失率逐年增加。因此，對蜂群活動力和花粉覓食行為的監測日益受到蜂產業管理的重視。蜜蜂的進出巢活動及其花粉覓食行為是該領域研究的關鍵資訊。然而，傳統的人工觀察巢內進出活動及使用花粉捕捉器來收集蜂箱花粉量的方法非常耗費人力。因此，本研究提出了一套創新的蜜蜂觀察盒，用於收集蜜蜂巢內進出的影像，以及一個能夠同時計數蜜蜂進出數量與花粉攜帶量的蜜蜂計數演算法。該演算法結合了YOLOv8物件偵測模型、PP-LiteSeg 語義分割模型和ByteTrack 追蹤演算法。經過測試，此蜜蜂計數演算法在MAPE 上達到5.43% 的計數表現，且與人工計數結果的相關性達到0.9909。此外，通過識別每隻蜜蜂攜帶的花粉面積，每小時自動計數的花粉總面積與花粉陷阱收集的花粉總重量之間的相關性達到0.9506。 本研究提出了一種新穎的蜂群監控方法，它使研究人員能高效且節省勞力的自動監測蜜蜂進出巢數據，以深入研究蜜蜂的覓食行為。這種監控方法基於不僅可用於監測蜜蜂的巢內進出活動，還能以基於影像監測的方式量測蜜蜂的花粉採集量。

In recent years, due to climate change and environmental pollution, the annual loss rate of beekeeping has been climbing. As a result, monitoring the activity and foraging behavior of honeybee colonies has become increasingly important for honeybee management. Among these, the activity of honeybees entering and leaving the beehive and honey pollen foraging behavior is critical information in related honeybee research areas. However, traditional methods of manually observing beehive entering-and-leaving activity and collecting pollen amounts from pollen traps are labor-intensive. Therefore, this research designs a novel honeybee observation box for capturing honeybee videos. Additionally, this research proposes an innovative honeybee counting algorithm for simultaneously counting the honeybee numbers entering and leaving the beehive and the amount of pollen they bear. This honeybee counting algorithm is composed of the YOLOv8 object detection model, the PP-LiteSeg semantic segmentation model, and the ByteTrack tracking algorithm. After testing, this honeybee counting algorithm achieved an automatic counting performance with a MAPE of 5.43% and showed a high correlation (r=0.9909) with manual counts. Additionally, by identifying the area of pollen carried by each honeybee, the total predicted pollen area has a high correlation (r=0.9506) with the total measured pollen weight collected from pollen traps. This research offers a novel method for monitoring honeybee colonies, providing researchers with efficient and labor-saving automated monitoring data to further study honeybee foraging behavior. This monitoring method not only allows for tracking honeybee activity but also enables the measurement of pollen foraging amounts from honeybees.