國際間水果銷售常因國與國間貿易政策或天然災害等原因導致產量、銷量不穩定，此情況影響到一個國家的經濟穩定性更影響到果農與經銷商的生計。台灣在過去三年多種水果產量穩定上升，但因我國的水果外銷政策改變導致外銷市場緊縮加上許多果農因某種水果銷量高而一窩蜂開始種植同一種水果導致某些種類的水果產量過剩。
為解決水果產量過剩的問題，透過尋找其他外銷出路、與觀光產業結合等方法都可以有效提升水果需求。自動化獲得水果產量可讓政府單位儘 早討論出相關解決措施、果農得以提早估計成本與利潤關係以避免投入過多成本。
本研究所提出的視覺式圓形水果產量估計系統主要透過 image division、 fruit detection、 sub-image combination三個步驟來估計水果產量。首先將影像分割為較小尺寸影像避免原始影像像素丟失並使用 Faster R-CNN架構偵測較小尺寸影像中的椪柑並拼接回原始影像，在拼接過程中去除同一顆柑橘因影像分割導致之被重複計算的情況。
本研究對自建椪柑資料集之偵測正確率，對完整不被遮蔽之椪柑可達98.65%，對各種被遮蔽狀況之椪柑正確率皆可維持於93%以上。對於面積 小於30×30像素之椪柑偵測正確率達90%以上。

International fruit sales often lead to unstable production and sales due to national and inter-country trade policies or natural disasters. This situation affects the economic stability of a country and affects the livelihood of farmers and distributors.
The fruit production has steadily increased in the past three years. However, many fruit farmers have grown a large number of fruits, and plant the same kind of fruit in large quantities, and this leads to overproduction of certain types of fruits. Besides, the Taiwan government has changed the fruit export policy, leading to a tightening of the export market.
In order to prevent the surplus of farm products during peak seasons, an automated fruit yields system may help government agencies to discuss relevant solutions as early as possible, and farmers can early estimate cost and profit relationships to avoid excessive costs.
This study proposes the vision-based round fruit yield estimation system through three stages: image division, fruit detection, and sub-image combination. First, to avoid loss quality of the original image, when the video is input into the system, the video of a frame image is divided into smaller size images before fruit detection. Second, this system presents an improved Faster R-CNN architecture to detect the location of the citrus. Final, after image division and fruit detection stages, the divided citrus images are possible to calculate repeatedly. Therefore, in the sub-image combination, the divided citrus images need to merge back to the original image, and remove the duplicated citrus images.
This study presents a round fruit yield estimation system based on an approved neural model and collects a self-built citrus dataset contained a total of 1590 images. In the work, there are four categories of the citrus dataset includes: complete, 1/2 shaded area, over 3/4 shaded area, and less than 30×30 pixels. The accuracy rate of the complete class is 98.65%. The accuracy rate of 1/2 occluded and 3/4 occluded area classes are above 93%. For the class of less than 30 × 30 pixels, the accuracy is more than 90%.

[Bar17] S. Bargoti and J. Underwood, ”Deep Fruit Detection in Orchards,” Proceedings of 2017 IEEE International Conference on Robotics and Automation (ICRA), Singapore, pp. 1-8, 2017.
[Ren17] S. Ren, K. He, R. Girshick, and J. Sun, “Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Network,” IEEE Transaction on Pattern Analysis and Machine Intelligence, Vol. 39, pp. 1-14, No. 6, 2017.
[Cak13] Y. Çakır, M. Kırcı, and E. Güneş, “Detection of Oranges in Outdoor Conditions”, Proceedings of 2013 Second International Conference on Agro-Geoinformatics (Agro-Geoinformatics), pp.500-503, 2013.
[Sen12] S. Sengupta and W.Lee, “Idenficatioin and determination of the number of immature green citrus fruit in a canopy under different ambient light conditions,” Proceedings of International Conference of Agricultural Engineering CIGR-AgEng2012, Spain, pp. 51-61, 2012.
[Li16] H. Li, W. Lee, and K. Wang, “Immature green citrus fruit detection and counting based on fast normalized cross correlation (FNCC) using natural outdoor colour images,” Precision Agriculture, USA, Vol. 17, Issue 6, pp 678–697, 2016.
[Mal16] W. Maldonado and J. Barbosa, “Automatic green fruit counting in orange trees using digital images,” Computers and Electronics in Agriculture, Vol. 127, pp. 572-581, 2016.
[Dai16] J. Dai, Y. Li, K. He, and J. Sun, “R-FCN: Object Detection via Region-based Fully Convolutional Networks,” Proceedings of 30th Conference on Neural Information Processing Systems (NIPS 2016), Spain, pp. 1-9, 2016.
[Wan18] C. Wang, W. Lee, X. Zou, D. Choi, and J. Diamond, “Detection and counting of immature green citrus fruit based on the Local Binary Patterns (LBP) feature using illumination-normalized images,” Precision Agriculture, Vol. 19, Issue. 6, pp. 1062–1083, 2018.
[Fel04] P. Felzenszwalb and D. Huttenlocher, “Efficient Graph-Based Image Segmentation,” International Journal of Computer Vision, Vol. 59, Issue. 2, pp. 167-181, 2004.
[Gir15] R. Girshick, “Fast R-CNN,” Proceedings of International Conference on Computer Vision (ICCV), USA, pp. 1440-1448, 2015.
[Gir13] R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp.1-21, 2013.
[He15] 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 (TPAMI), Vol. 37, Issue. 9, pp. 1904-1916, 2015.
[Roh17] M. Roh and J. Lee, “Refining Faster-RCNN for Accurate Object Detection,” Proceedings of 2017 Fifteenth IAPR International Conference on Machine Vision Applications (MVA), Japan, pp. 514-517, 2017.
[Mai18] X. Mai, H. Zhang, and Q. Meng, “Faster R-CNN with Classifier Fusion for Small Fruit Detection ,” Proceedings of IEEE International Conference on Robotics and Automation (ICRA), Australia, pp. 7166-7172, 2018.
[Sun17] X. Sun, P. Wu, and S. Hoi, “Face detection using deep learning: An improved Faster-RCNN approach,” Proceedings of 2017 Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-11, 2017.
[Zha16] C. Zhao, W. Lee, and D. He, “Immature green citrus detection based on colour feature and sum of absolute transformed difference (SATD) using colour images in the citrus grove,” Computers and Electronics in Agriculture, USA, pp. 243-253, 2016.
[Liu16] W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. reed, c. fu and a. Berg, “ SSD: Single Shot MultiBox Detector,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-17, 2016.
[Bel15] S. Bell, C. Zitnick, K. Bala, and R. Girshick, “Inside-Outside Net: Detecting Objects in Context with Skip Pooling and Recurrent Neural Networks,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-11, 2015.
[Kon16] T. Kong, A. Yao, Y. Chen, and F. Sun, “HyperNet: Towards Accurate Region Proposal Generation and Joint Object Detection,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-9, 2016.
[Bod17] N. Bodla, B. Singh, R. Chellappa, and L. Davis, “Improving Object Detection With One Line of Code,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-9, 2017.
[Kon17] T. Kong, F. Sun, A. Yao, H. Liu, M. Lu, and Y. Chen, “RON: Reverse Connection with Objectness Prior Networks for Object Detection,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-10, 2017.
[Zhu16] C. Zhu, Y. Zheng, K. Luu, and M. Savvides, “CMS-RCNN: Contextual Multi-scale Region-based CNN For Unconstrained Face Detection,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-12, 2016.
[Li18] B. Li, T. Wu, L.Zhang, and R. Chu, “Auto-Context R-CNN,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-18, 2018.
[Egg17] C.Egg, D. Zecha, S.Brehm, and R. Lienhart, “Improving Small Object Proposals for Company Logo Detection,” Proceedings of Computer Vision and Pattern Recognition (CVPR), USA, pp. 1-8, 2017.
[48]農委會農業貿易統計表http://agrstat.coa.gov.tw/sdweb/public/trade/TradeReport.aspx
[49]行政院農業委員http://agrstat.coa.gov.tw/sdweb/public/trade/tradereport.aspx
[50]火龍果的新聞 https://www.newsmarket.com.tw/blog/72960/