摘要
為了應對氣候變遷帶來的挑戰及其對全球糧食安全的影響，精準農業變得越來越重要。這項研究的重點是自動檢測小麥葉子中的氣孔和毛狀體，這是了解植物生理學和提高作物復原力的關鍵任務。氣孔是葉子表面的微小孔隙，對於調節氣體交換和光合作用至關重要，因此對其分析對於培育更具彈性的作物品種至關重要。傳統的氣孔計數方法是勞力密集的，因此需要自動化解決方案。 為了驗證設計，本研究探討了 YOLOv8 物件導向偵測模型的應用，結合半監督學習技術（Soft Teacher）和整合兩個管道，來檢測小麥葉子中的氣孔和毛髮。Yolov8 模型比之前的迭代有所進步，它在帶有註釋的小麥氣孔圖像上進行訓練，然後使用半監督方法來利用未標記的數據進行改進。此方法旨在提高檢測精度，同時減少大量手動註釋的需要。該模型的性能在未見過的小麥圖像上進行了評估，表明與完全監督的模型相比，精確度和召回率指標顯著提高。半監督 YOLOv8 方法在有效檢測氣孔方面顯示出前景，特別是在具有不同氣孔大小和形狀的挑戰性場景中。 這份研究主要用 YOLOv8 模型與半監督學習相結合並使用整體學習，為小麥中的自動化氣孔檢測提供了穩健的解決方案，結果表明，整合模型可以取得更好的 mAP@50:5:95 結果，為小麥氣孔自動檢測奠定了基礎。未來的工作將集中在進一步優化不同作物品種的模型，並探索整合額外的環境數據以提高檢測精度。
並列摘要
In response to the challenges posed by climate change and its impact on global foodsecurity, precision agriculture is becoming increasingly vital. This research focuses on automating the detection of stomata and hair in wheat leaves, a critical task for understandingplant physiology and improving crop resilience. Stomata, the tiny pores on leaf surfaces,are essential for regulating gas exchange and photosynthesis, making their analysis crucial in breeding more resilient crop varieties. Traditional methods of stomata counting arelabor-intensive, prompting the need for automated solutions. To verify the design, this study explores the application of the YOLOv8 oriented object detection model, coupled with semi-supervised learning techniques (Soft teacher) and an ensemble two pipelines, to detect stomata and hair in wheat leaves. The Yolov8 model, an advancement over previous iterations, was trained on annotated wheat stomata images and then refined using a semi-supervised approach to leverage unlabeled data. This method aimed to enhance detection accuracy while reducing the need for extensive manual annotations. The model’s performance was evaluated on unseen wheat images, demonstrating significant improvements in precision and recall metrics compared to fully supervised models. The semi-supervised YOLOv8 approach shows promise in efficiently detecting stomata, particularly in challenging scenarios with varying stomatal sizes and shapes. The findings suggest that the YOLOv8 model, combined with semi-supervised learning, offers a robust solution for automating stomata detection in wheat,the results show that the integrated model can achieve better results of mAP@50:5:95, paving the way for its application in broader agricultural contexts. Future work will focus on further optimizing the model for different crop species and exploring the integration of additional environmental data to enhance detection accuracy.
參考文獻
[1] Farquhar GD and Sharkey TD. Stomatal conductance and photosynthesis. AnnuRev Plant Physiol., 33:317-345, 1982.
[2] Jiaxin Wang, Heidi J Renninger, Qin Ma, and Shichao Jin. Measuring stomatal and guard cell metrics for plant physiology and growth using stomanager1. Plant Physiology, 195(1):378–394, 01 2024.
[3] Angela Casado-García, Arantza del Canto, Alvaro Sanz-Saez, Usue Pérez-López,Amaia Bilbao-Kareaga, Felix B. Fritschi, Jon Miranda-Apodaca,AlbertoMuñozRueda, Anna Sillero-Martínez, Ander Yoldi-Achalandabaso, Maite Lacuesta, and Jónathan Heras. Labelstoma: A tool for stomata detection based on the yolo algorithm. Computers and Electronics in Agriculture, 178:105751, 2020.
[4] Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. You only look once: Unified, real-time object detection. Computer Vision and Pattern Recognition Conference (CVPR), pages 779–788, 2016.
[5] Wei Liu, Dragomir Anguelov, Dumitru Erhan, Christian Szegedy, Scott Reed,Cheng-Yang Fu, and Alexander C. Berg. Ssd: Single shot multibox detector.