本研究研發蘭花苗株自動化盤點系統，藉由影像辨識將植床上之蘭花苗株中心框選出來，並盤點植床上苗株之數量。在此系統中，使用自行架設在龍門架上之相機座，配合廠商之自動運輸軌道來採集蘭花苗株在植床上的影像。再使用影像縫合之技術重建完整的植床照片，同時藉由二維條碼將蘭花苗株分批，最後以物件偵測演算法進行辨識後，將植床上各苗株框選出來並且除去重複框選之苗株，最後盤點植床上的苗株個數。本研究之物件偵測演算法以多層卷積的架構為骨幹，並搭配遷移學習以增加辨別時的準確度。在資料庫的建立上，由於植床上之苗株分布密集，每張照片需標記的數量甚多，使得樣本數難以增加，且單張照片尺寸太大也會降低訓練網路的精度，因此，本研究將原始照片裁剪成多個小照片，有效地增加樣本數，使得網路訓練時照片較不容易被壓縮而失真，進而提高演算法辨識精度。以最適合本研究之演算法而論，各個盤點辨識指標都高於98%。
為了能實際應用於產業上，本研究資料庫建立中所拍攝的照片皆採集於所配合之蘭花廠商，確保未來在蘭花廠商之產線中可以實際使用，以達到減省人力、提高生產率。

This thesis develops an automated counting system for orchid seedlings. In this study, the center of the orchid seedlings on the plant tray is detected by image recognition, and the number of seedlings on the plant tray is counted. The pictures are taken by moving the automatic conveyor belt, using the self-erected camera mounted on a higher horizontal rack, and the pictures of the complete plan tray are reconstructed utilizing the image stitching technology. The 2D barcode is scanned for an identified batch of orchid seedlings, and the object detection algorithm is implemented to identify the orchid seedlings and to remove the repeated ones. Finally, the number of orchid seedlings on the plant tray is counted and summarized. The object detection algorithm uses a convolutional neural network as backbone, and the transfer learning is employed to increase the accuracy of discrimination. In building the database of orchid seedlings, due to the nature of dense distribution of orchid seedlings on the plant tray, the number of labels to be tagged in a photo is often very high. As a result, it is difficult to increase the number of samples, and the oversized photo will also greatly reduce the accuracy of network training output. Hence, in this study the original cropping photos with many seedlings are cut into multiple photos with small amount of seedlings in each photo to effectively increase the samples, to make the photos less likely to be compressed and distorted during network training, and thereby to improve the accuracy of identification algorithms. The experimental results show that all evaluation accuracies of counting are higher than 98% when the best algorithm is utilized in this study. Furthermore, all photos of orchid seedlings in the database are taken from the cooperative orchid plant to ensure that the developed system can be realized in its production lines of routine counting in the future. As a result, the company is able to reduce their labor working and to increase counting accuracy.

[1]行政院農業委員會. (2019). 農業統計資料查詢. August 25, 2020. [Online]. Available: https://agrstat.coa.gov.tw/sdweb/public/inquiry/InquireAdvance.aspx
[2]盧紀燁、陳哲銘. (2018). 高空影像辨識之農作物大盤點. August 25, 2020. [Online]. Available: https://issuu.com/pdis.tw/docs/20180327
[3]台灣儀器科技研究中心. (2019). 無人機於精準農業之應用. August 25, 2020. [Online]. Available: https://www.tiri.narl.org.tw/Publication/InstTdy_Full/0220020?IssueId=351
[4]Y. LeCun, L. Bottou, Y. Bengio, and P. Haffner, "Gradient-based learning applied to document recognition," Proceedings of the IEEE, vol. 86, no. 11, pp. 2278-2324, 1998.
[5]K. Simonyan and Andrew Zisserman. “Very Deep Convolutional Networks for Large-Scale Image Recognition.” CoRR abs/1409.1556 (2015): n. pag.
[6]Krizhevsky, A., Ilya Sutskever and Geoffrey E. Hinton. “ImageNet classification with deep convolutional neural networks.” CACM (2017).
[7]CNN Architectures : VGG, Resnet, InceptionNet, XceptionNet. August 25, 2020. [Online]. Available: https://www.kaggle.com/shivamb/cnn-architectures-vgg-resnet-inception-tl
[8]K. He, X. Zhang, S. Ren and J. Sun, "Deep Residual Learning for Image Recognition," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, 2016, pp. 770-778, doi: 10.1109/CVPR.2016.90.
[9]K. He, X. Zhang, S. Ren and J. Sun, (2016, October). "Identity mappings in deep residual networks." In European Conference on Computer Vision (pp. 630-645). Springer, Cham.
[10]K. He, X. Zhang, S. Ren, and J. Sun, "Spatial pyramid pooling in deep convolutional networks for visual recognition," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 37, no. 9, pp. 1904-1916, 2015.
[11]S. Lazebnik, C. Schmid, and J. Ponce, "Beyond bags of features: Spatial pyramid matching for recognizing natural scene categories," in 2006 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'06), 2006, vol. 2, pp. 2169-2178.
[12]大疆創新. (2016). Phantom 4. August 25, 2020. [Online]. Available: https://www.dji.com/tw/phantom-4?gclid=Cj0KCQiA0NfvBRCVARIsAO4930nVdNQB76Frqtb-iIuvDR1vCO6S6ZHBMfpgKZ39JEgY6LoJVHXaX1saAm_sEALw_wcB
[13]logitect. webcam - C922 pro. August 25, 2020. [Online]. Available: https://www.logitech.com/zh-tw/product/c922-pro-stream-webcam?crid=34
[14]S. G. Adam Paszke, Soumith Chintala, Gregory Chanan. (2016). PyTorch. August 25, 2020. [Online]. Available: https://pytorch.org/
[15]F. Chollet. (2015). Keras. August 25, 2020. [Online]. Available: https://keras.io/
[16](2017). 關於Keras. August 25, 2020. [Online]. Available: https://kknews.cc/zh-tw/news/yy84r8j.html
[17]G. B. Team. (2015). TensorFlow. August 25, 2020. [Online]. Available: https://www.tensorflow.org/
[18]G. Bradski and A. Kaehler, Learning OpenCV: Computer vision with the OpenCV library. " O'Reilly Media, Inc.", 2008.
[19]Images stitching. August 25, 2020. [Online]. Available: https://docs.opencv.org/4.1.2/d1/d46/group__stitching.html
[20]M. Brown and D. G. Lowe, "Automatic panoramic image stitching using invariant features," International Journal of Computer Vision, vol. 74, no. 1, pp. 59-73, 2007.
[21]S. Ren, K. He, R. Girshick and J. Sun, "Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks," in IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 39, no. 6, pp. 1137-1149, 1 June 2017, doi: 10.1109/TPAMI.2016.2577031.
[22]Z. Cai and N. Vasconcelos, "Cascade R-CNN: Delving Into High Quality Object Detection," 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Salt Lake City, UT, 2018, pp. 6154-6162, doi: 10.1109/CVPR.2018.00644.
[23]Evaluating detection (Intersection Over Union). August 25, 2020. [Online]. Available: https://www.oreilly.com/library/view/hands-on-convolutional-neural/9781789130331/a0267a8a-bd4a-452a-9e5a-8b276d7787a0.xhtml
[24]J. Redmon and A. Farhadi, "Yolov3: An incremental improvement," arXiv preprint arXiv:1804.02767, 2018.
[25]R. Girshick, J. Donahue, T. Darrell and J. Malik, "Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," 2014 IEEE Conference on Computer Vision and Pattern Recognition, Columbus, OH, 2014, pp. 580-587, doi: 10.1109/CVPR.2014.81.
[26]J. R. Uijlings, K. E. Van De Sande, T. Gevers, and A. W. Smeulders, "Selective search for object recognition," International Journal of Computer Vision, vol. 104, no. 2, pp. 154-171, 2013.
[27]J. Redmon, S. Divvala, R. Girshick and A. Farhadi, "You Only Look Once: Unified, Real-Time Object Detection," 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, 2016, pp. 779-788, doi: 10.1109/CVPR.2016.91.
[28]J. Redmon and A. Farhadi, "YOLO9000: Better, Faster, Stronger," 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Honolulu, HI, 2017, pp. 6517-6525, doi: 10.1109/CVPR.2017.690.
[29](2018). YOLO v3深入理解. August 25, 2020. [Online]. Available: https://zhuanlan.zhihu.com/p/49556105?fbclid=IwAR23SjXryPoEPc87AW1mgo51vKWXhvwfEYE_rDTWLy5xhz3BfUP3wui8Vyg