隨著人工智慧技術的發展，植物病害的自動化識別已成為提升農業生產效率的關鍵。儘管深度學習模型在病害診斷中表現出色，但在複雜田間環境下，圖像若包含非葉片背景或多片葉子，現有方法難以精準判別葉片，亦無法對個別葉面的病癥進行細部分析。為克服此挑戰，本研究旨在開發一套更具穩健性與多功能性的智慧農業服務系統。
本文提出一種創新的雙階段深度學習識別策略：首先，利用 MobileNetV3 模型對輸入圖像進行葉片存在性預判；隨後，再結合 YOLOv11 模型對圖像中經確認的個別葉片進行精準的疾病檢測。此策略有效解決了複雜田間背景下病害識別的挑戰，能精確識別健康、赤星病 (Rust)、黑星病 (Scab) 及黑斑病 (Black rot) 等多種病癥。
本研究的學術貢獻在於首次整合葉片預判與病害物件偵測的雙階段深度學習模型（MobileNetV3 + YOLOv11），顯著提升了模型在非標準化輸入圖像下的病害識別穩健性與精準度。實務上，本研究開發的系統透過整合至 LINE Bot 平台，為農民提供了一站式的智慧農業解決方案。這不僅包含即時的病害診斷與地理位置結合的病害數據可視化，能有效呈現病害分佈趨勢，更拓展至提供即時天氣查詢、農產品價格查詢、雷達回波圖、系統使用說明等多樣化資訊服務。實驗結果表明，本系統能顯著提升現實農業場景中病害偵測的準確性、應用廣度與整體農業管理效率，為智慧農業的推動奠定堅實的技術基礎。

With the advancement of artificial intelligence technology, automated identification of plant diseases has become crucial for enhancing agricultural production efficiency. While deep learning models, such as convolutional neural networks and their variants, excel in disease diagnosis, existing methods struggle to accurately distinguish leaves from complex field backgrounds or when images contain multiple leaves, making detailed analysis of individual leaf symptoms challenging. To overcome these limitations, this study aims to develop a more robust and versatile smart agricultural service system.
This paper proposes an innovative two-stage deep learning identification strategy: first, a MobileNetV3 model performs preliminary leaf presence detection on input images to filter out irrelevant backgrounds; subsequently, a YOLOv11 model conducts precise disease detection on the confirmed individual leaves. This strategy effectively addresses the challenge of disease identification in complex field environments, accurately identifying various symptoms such as healthy, Rust, Scab, and Black rot.
The academic contribution of this research lies in its first-time integration of a two-stage deep learning model (MobileNetV3 for leaf pre-judgment and YOLOv11 for disease object detection). This significantly enhances the robustness and accuracy of disease identification for non-standardized input images. Practically, the system developed in this study, integrated into the LINE Bot platform, offers farmers a one-stop smart agricultural solution. This solution not only provides real-time disease diagnosis combined with geospatial visualization of disease data, effectively showcasing disease distribution trends, but also expands to include diverse information services such as real-time weather queries, agricultural product price inquiries, radar echo maps, and system usage instructions. Experimental results demonstrate that this system significantly improves the accuracy, application breadth, and overall agricultural management efficiency of disease detection in real-world agricultural scenarios, laying a solid technical foundation for advancing smart agriculture.

[1] Ait NasserA., & AkhloufiAM. (2024). A hybrid deep learning architecture for apple foliar disease detection. Computers, 13(5), 116. 擷取自 https://doi.org/10.3390/computers13050116
[2] HowardSandler, M., Chen, B., Wang, W., Chen, L.-C., Tan, M., Chu, G., Vasudevan, V., Zhu, Y., Pang, R., Adam, H., & Le, Q.A.,. (2019). Searching for MobileNetV3. CVF International Conference on Computer Vision (ICCV) (頁 1314–1324). Seoul, Korea (South): IEEE.
[3] JocherG., & QiuJ. (2024). Ultralytics YOLO11, Version 11.0.0. 擷取自 Ultralytics YOLO11: https://github.com/ultralytics/ultralytics
[4] LiJ., ZhuX., JiaR., LiuB., & YuC. (2022). Apple-YOLO: A novel mobile terminal detector based on YOLOv5 for early apple leaf diseases. 2022 IEEE 46th Annual Computers, Software, and Applications Conference (COMPSAC), (頁 352-361). 擷取自 https://doi.org/10.1109/COMPSAC54236.2022.00056
[5] LINE. (2024年12月17日). 【LINE App 2024使用數據】. (LINE Plus 公司) 擷取自 全台每日LINE通話數達1億次 AI生成應用接受度高 數位生活有LINE更便利: https://linecorp.com/tw/pr/news/2024/1217/
[6] LiT., ZhangL., & LinJ. (2024). Precision agriculture with YOLO-Leaf: advanced methods for detecting apple leaf diseases. Frontiers in Plant Science, 15, 1452502. 擷取自 https://doi.org/10.3389/fpls.2024.1452502
[7] MaL., ZhaoL., WangZ., ZhangJ., & ChenG. (2023). Detection and counting of small target apples under complicated environments by using improved YOLOv7-tiny. Agronomy, 13(5), 1419-. 擷取自 https://doi.org/10.3390/agronomy13051419
[8] SalamNaznine, M., Jahan, N., Nahid, E., Nahiduzzaman, M., & Chowdhury, M. E. H.A.,. (2024). Mulberry Leaf Disease Detection Using CNN-Based Smart Android Application. IEEE Access(12), 83575–83588. 擷取自 https://doi.org/10.1109/ACCESS.2024.3407153
[9] SB., & JT. (2024). Deep learning-based detection of fungal diseases in apple plants using YOLOv8 algorithm. 2024 International Conference on Advances in Data Engineering and Intelligent Computing Systems (ADICS), (頁 1–6). 擷取自 https://doi.org/10.1109/ADICS58448.2024.10533474
[10] SQLite. (2025年04月16日). About SQLite. 擷取自 https://www.sqlite.org/about.html
[11] ThapaSnavely, N., Belongie, S., & Khan, A.R.,. (2020). The Plant Pathology 2020 challenge dataset to classify foliar disease of apples. 擷取自 https://doi.org/10.48550/arxiv.2004.11958
[12] TianY., YangG., WangZ., WangH., LiE., & LiangZ. (2019). Apple detection during different growth stages in orchards using the improved YOLO-V3 model. Computers and Electronics in Agriculture, 157, 417–426. 擷取自 https://www.sciencedirect.com/science/article/abs/pii/S016816991831528X
[13] Ultralytics. (2024年9月30日). Ultralytics YOLO11. 擷取自 Ultralytics Docs: https://docs.ultralytics.com/zh/models/yolo11/
[14] VishnoiK.V., KumarK., KumarB., MohanS., & KhanA.A. (2023). Detection of apple plant diseases using leaf images through convolutional neural network. IEEE Access, 11, 6594–6609. 擷取自 https://doi.org/10.1109/ACCESS.2022.3232917
[15] WangA., ChenH., LiuL., ChenK., LinZ., & HanJ. (2024). Yolov10: Real-time end-to-end object detection. Advances in Neural Information Processing Systems, 37, 107984–108011. 擷取自 https://doi.org/10.48550/arxiv.2405.14458
[16] WangAugustin. (2017年07月31日). 開發LINE聊天機器人不可不知的十件事. 擷取自 https://engineering.linecorp.com/zh-hant/blog/line-device-10
[17] WangWang, Y., & Zhao, J.Y.,. (2022). MGA-YOLO: A lightweight one-stage network for apple leaf disease detection. Frontiers in Plant Science, 13, 927424. 擷取自 https://doi.org/10.3389/fpls.2022.927424
[18] WuQiusheng. (2021). Leafmap: A Python package for interactive mapping and geospatial analysis with minimal coding in a Jupyter environment. Journal of Open Source Software, 3414. 擷取自 https://doi.org/10.21105/joss.03414
[19] XieX., LiuZ., WangY., FuH., LiuM., ZhangY., & XuJ. (2024). Puppet dynasty recognition system based on MobileNetV2. Entropy, 26(8), 645-. 擷取自 https://doi.org/10.3390/e26080645
[20] XinmingW., & HongS.T. (2023). Comparative study on leaf disease identification using YOLO v4 and YOLO v7 algorithm. AgBioForum, 25(1). 擷取自 https://agbioforum.org/menuscript/index.php/agb/article/view/192
[21] ZhangL., SunZ., TaoH., WangM., & YiW. (2025). Research on mine-personnel helmet detection based on multi-strategy-improved YOLOv11. Sensors, 25(1), 170. 擷取自 https://doi.org/10.3390/s25010170
[22] 中華民國交通部中央氣象署. (2025年07月08日). 交通部中央氣象署1 周預報. 擷取自 https://www.cwa.gov.tw/V8/C/W/week.html
[23] 交通部中央氣象署. (2025年07月06日). 雷達回波圖. 擷取自 https://www.cwa.gov.tw/V8/C/W/OBS_Radar.html
[24] 沈原民 (Shen黃冬青 (Huang, T.-C.), 洪挺軒 (Hung, T.-H.)Y.-M.),. (2016). 殺菌劑對梨赤星病感染源之抑制效果評估. 植物醫學 (Plant Medicine), 58(2), 79–84. 擷取自 https://doi.org/10.6716/JPM.201606_58(2).0004
[25] 教育學習網STEAM. (2021-2023). STEAM 教育學習網. 擷取自 STEAM 教育學習網: https://steam.oxxostudio.tw/category/python/example/line-bot.html
[26] 谢文远陈子林, 陈锋, & 陈征海张方钢,. (2019). 浙江蔷薇科苹果亚科植物分类修订. 杭州师范大学学报（自然科学版）, 18(3), 261–267. 擷取自 https://doi.org/10.3969/j.issn.1674-232X.2019.03.007
[27] 蔡宜潔、葉志忠、李昭蓉、黃淑玲. (2024). 113 年9 月農業產銷概況. 農業情報(389), 68-73. 擷取自 https://www.moa.gov.tw/ws.php?id=2515529