臺灣博碩士論文加值系統

洋菇是全世界食用最多的菇種，在台灣也是極為重要的經濟作物。洋菇產業在採收人力的高流動率和採收判斷不一致狀況下，世界各國的洋菇農企業均有朝向自動化採收的趨勢。而洋菇採收系統的技術瓶頸之一在於採收的判斷時機。從高光譜技術上來看，高光譜影像技術過去常用在遙測領域上，後來也逐步應用到農作物非破壞性的品質檢測上，像是農產品瑕疵檢測、病害檢測、甜度，和成熟度等檢測上。因此本研究利用高光譜影像和類神經網路技術來辨別洋菇4個成熟度階段以作為採收的判斷依據。首先將拍攝後的洋菇高光譜影像透過Mask-RCNN方法擷取出單顆洋菇以建立單顆洋菇高光譜影像資料集。並透過本研究所提出的特徵萃取方式來降低資料維度，再送入2層式類神經網路模型中，最後取得4個成熟度階段的分類結果。本研究結果顯示辨識的交叉驗證平均正確率為98.35% ± 3.51%。

The Agaricus bisporus is the most consumed mushroom species in the world, and it is also an extremely important economic crop in Taiwan. In the mushroom industry, the mushroom agricultural enterprise of all countries in the world are trending towards automatic harvesting. One of the technical bottlenecks of the mushroom harvesting system lies in the timing of harvesting. From the perspective of hyperspectral technology, hyperspectral imaging technology has been gradually applied to the field of non-destructive quality detection of crops, such as agricultural product defect detection, disease detection, sweetness, and maturity detection, and so on. Therefore, this study uses hyperspectral imaging and neural networks to identify the four maturity stages of mushroom and use it as the basis for harvesting determination. First, a single mushroom is extracted separately through Mask-RCNN method from the captured hyperspectral images, and then create data sets. Through feature extraction method to reduce the data dimension, and then feed into the 2-layer neural network model, the classification results of 4 maturity stages are finally obtained. The results show that the detection average accuracy rate with cross validation is 98.35% ± 3.51%.

摘要 i
Abstract ii
目錄 iii
圖目錄 vi
表目錄 ix
第一章 緒論 1
1.1 研究背景 1
1.2 研究動機與目的 1
1.3 論文架構 2
第二章 相關文獻探討 3
2.1 破壞性檢測洋菇成熟度 3
2.2 非破壞性檢測法 4
2.3 高光譜影像技術 4
2.3.1 高光譜影像在農業的應用 4
2.4 Otsu二值化演算法 5
2.5 類神經網路 Neural Networks 8
2.6 Mask-RCNN網路架構 9
第三章 研究設備與方法 12
3.1 整體流程 12
3.2 實驗器材 14
3.2.1 高光譜成像系統介紹 14
3.2.2 高光譜成像系統設定 15
3.3 資料集準備及前處理 16
3.3.1 製作高光譜影像之流程 20
3.3.2 產生訓練和測試資料集之流程 24
3.4 洋菇高光譜成熟度階段分類之流程 31
3.5 本文方法 33
3.5.1 黑白校正 33
3.5.2 影像正規化（Image Normalization） 33
3.5.3 Otsu使用方法 34
3.5.4 Mask-RCNN使用方法 34
3.5.5 Neural Network架構 35
3.5.6 轉換特徵方法 37
第四章 實驗結果 38
4.1 實驗資料集 38
4.1.1 洋菇群高光譜影像（高光譜儀影像） 39
4.1.2 洋菇群高光譜影像（Mask-RCNN mask標註之影像） 40
4.1.3 單顆洋菇高光譜資料集影像 41
4.1.4 洋菇資料集特徵圖 43
4.2 實驗評估指標 44
4.3 實驗項目 44
4.3.1 NIR/VNIR+（只取前景影像+mean+variance） 46
4.3.2 Model層數+（NIR+只取前景影像+mean+variance） 47
4.3.3 只取前景影像/取整張圖+（NIR+mean+variance） 48
4.3.4 Epochs+（只取前景影像+NIR+mean+variance） 49
4.3.5 比較5種轉換特徵組合+（NIR+只取前景影像） 50
4.3.6 在最高正確率的組合之下做11摺交叉驗證 52
4.4 實驗結果分析 53
第五章 結論與建議 54
5.1 結論 54
5.2 建議 54
參考文獻 56

[1] Kashkoush, M., ＆ Avigad, G. (2018, September). A Decision Support System for Automated Mushroom Harvesting. In Proceedings of SAI Intelligent Systems Conference (pp. 1178-1184). Springer, Cham.
[2] 鄭燮（民57）。不同成熟度採收洋菇對產量之影響。中國園藝，3＆4，177-187。
[3] 邱相文、石信德、呂昀陞（民108）。菇類栽植盛盤自動化上下架系統之開發。菇類智慧化生產與農場經營管理研討會專刊:農業試驗所特刊，216，53-60。
[4] Tillett, R. D., ＆ Batchelor, B. G. (1991). An algorithm for locating mushrooms in a growing bed. Computers and electronics in agriculture, 6(3), 191-200.
[5] Reed, J. N., ＆ Tillett, R. D. (1994). Initial experiments in robotic mushroom harvesting. Mechatronics, 4(3), 265-279.
[6] Reed, J. N., Miles, S. J., Butler, J., Baldwin, M., ＆ Noble, R. (2001). AE—Automation and emerging technologies: Automatic mushroom harvester development. Journal of Agricultural Engineering Research, 78(1), 15-23.
[7] Rowley, J. H. (2009). Developing flexible automation for mushroom harvesting (Agaricus bisporus)(Doctoral dissertation). University of Warwick.
[8] Masoudian, A.(2013). Computer vision algorithms for an automated harvester(Master''s thesis). University of Western Ontario.
[9] Rapila, T., Kantola, J., ＆ Rapila, T. (2011). U.S. Patent No. 8,033,087. Washington, DC: U.S. Patent and Trademark Office.
[10] Janssen, J. J., Van Nunen, J. J., ＆ Giebels, M. M. (1995). U.S. Patent
No. 5,471,827. Washington, DC: U.S. Patent and Trademark Office.
[11] Russell, T. A., ＆ Wheeler, G. C. (1993). U.S. Patent No. 5,185,989. Washington, DC: U.S. Patent and Trademark Office.
[12] Li, B., Lecourt, J., ＆ Bishop, G. (2018). Advances in non-destructive early assessment of fruit ripeness towards defining optimal time of harvest and yield prediction—a review. Plants, 7(1), 3.
[13] Tsai, S. Y., Wu, T. P., Huang, S. J., ＆ Mau, J. L. (2007). Nonvolatile taste components of Agaricus bisporus harvested at different stages of maturity. Food Chemistry, 103(4), 1457-1464.
[14] Taghizadeh, M., Gowen, A. A., ＆ O’Donnell, C. P. (2011). The potential of visible-near infrared hyperspectral imaging to discriminate between casing soil, enzymatic browning and undamaged tissue on mushroom (Agaricus bisporus) surfaces. Computers and electronics in agriculture, 77(1), 74-80.
[15] Qin, J., Chao, K., Kim, M. S., Lu, R., ＆ Burks, T. F. (2013). Hyperspectral and multispectral imaging for evaluating food safety and quality. Journal of Food Engineering, 118(2), 157-171.
[16] 吳柏輝(2019)。基於機器學習對於高光譜檢測水果甜度研究(碩士論文)。國立雲林科技大學。
[17] Zou, S., Tseng, Y. C., Zare, A., Rowland, D. L., Tillman, B. L., ＆ Yoon, S. C. (2019). Peanut maturity classification using hyperspectral imagery. Biosystems Engineering, 188, 165-177.
[18] Qiao, S., Wang, Q., Zhang, J., ＆ Pei, Z. (2020). Detection and Classification of Early Decay on Blueberry Based on Improved Deep Residual 3D Convolutional Neural Network in Hyperspectral Images. Scientific Programming, 2020. 
[19] Li, X., Wei, Y., Xu, J., Feng, X., Wu, F., Zhou, R., ... ＆ He, Y. (2018). SSC and pH for sweet assessment and maturity classification of harvested cherry fruit based on NIR hyperspectral imaging technology. Postharvest Biology and Technology, 143, 112-118.
[20] Otsu, N. (1979). A threshold selection method from gray-level histograms. IEEE transactions on systems, man, and cybernetics, 9(1), 62-66.
[21] Da Silva, I. N., Spatti, D. H., Flauzino, R. A., Liboni, L. H. B., ＆ dos Reis Alves, S. F. (2017). Artificial neural network architectures and training processes. Springer, Cham.
[22] Goodfellow, I., Bengio, Y., ＆ Courville, A. (2016). Deep learning. MIT press.
[23] Murtagh, F. (1991). Multilayer perceptrons for classification and regression. Neurocomputing, 2(5-6), 183-197.
[24] He, K., Gkioxari, G., Dollár, P., ＆ Girshick, R. (2017). Mask r-cnn. In Proceedings of the IEEE international conference on computer vision (pp. 2961-2969).
[25] Jiongnima(民107年1月21日)。實例分割模型Mask R-CNN詳解：從R-CNN，Fast R-CNN，Faster R-CNN再到Mask R-CNN【部落格文字資料】。取自https://blog.csdn.net/jiongnima/article/details/79094159
[26] 蕈菌的構造【部落格圖形資料】。取自http://learn.foodmate.net/class/study/point_part?block_id=109
[27] 羅榮豪(2013)。添加營養劑對洋菇產量的影響(碩士論文) 。國立中興大學。
[28] Waleed Abdulla(2018, Mar 20). Splash of Color: Instance Segmentation with Mask R-CNN and TensorFlow [Web blog message]. Retrieved from https://engineering.matterport.com/splash-of-color-instance-segmentation-with-mask-r-cnn-and-tensorflow-7c761e238b46
[29] Matterport(2018, March 20). Mask_RCNN [Web blog message]. Retrieved from https://github.com/matterport/Mask_RCNN