Please use this identifier to cite or link to this item:
Development of Auto-Sorting Device for Green Coffee Bean Defects with Machine Vision
|關鍵字:||咖啡生豆瑕疵;選別裝置;機器視覺;貝氏分類器;green coffee bean defects;sorting device;machine vision;Bayes classifier||引用:||王翠華。我國咖啡市場分析。行政院農委會，農政與農情。257(2013)。 中華民國經濟部國際貿易局貿易統計查詢網站 https://cus93.trade.gov.tw/ Taniwaki M.H., A.A. Teixeira, A.R.R Teixeira, M.V. Copetti, B.T. Lamanaka. Ochratoxigenic fungi and ochratoxin A in defective coffee beans. Food Research International. 61(2014): 161-166. Franca A.S., L. S. Oliveira, J.C.F. Mendnça, X.A. Silva. Physical and chemical attributes of defective crude and roasted coffee beans. Food Chemistry, 90(2005):89-94. Arabica Green Coffee Defect Handbook 3rded. Specialty Coffee Association(SCA). 2018. 陳昆陽。應用機器視覺於咖啡生豆外觀瑕疵檢測之研究。碩士論文。2016。國立中興大學生物產業機電工程學系。 李松源。台灣咖啡種植。益全照相製版社。2011。再版，P.120-139。 Wu D., D.W. Sun. Colour measurements by computer vision for food quality control - A review. Trends in Food Science & Technology, 29(2013)5-20. 黃國益。應用機器視覺於蝴蝶蘭大苗幾何特徵與病害檢測之研究。博士論文。2002。國立中興大學農業機械工程系。 簡貿謙。應用機器視覺於臺灣五大水稻種子品種辨識之研究。碩士論文。2016。國立中興大學生物產業機電工程學系。 Craig A.P., A.S. Franca, L.S. Oliveira, J. Irudayaraj, K. Ileleji. Fourier transform infrared spectroscopy and near infrared spectroscopy for the quantification of defects in roasted coffees. Talanta. 134(2015): 379-386. Faridah F., G.O.F. Parikesit, Ferdiansjah F. Coffee Bean Grade Determination Based on Image Parameter. TELKOMNIKA, 9(2011): 547-554. Pinto C., J. Furukawa, H. Fukai, S. Tamura. Classification of Green Coffee Bean Images Based on Defect Types Using Convolutional Neural Network(CNN). Proceedings of the 2017 International Conference on Advanced Informatics, Concepts, Theory, and Applications (ICAICTA). Awate A., D. Deshmankar, G. Amrutkar, U. Bagul, S. Sonavane. Fruit Disease Detection using Color, Texture Analysis and ANN. Proceedings of the 2015 International Conference on Green Computing and Internet of Things(ICGCloT). C.E. Portugal-Zambrano, J.C. Gutiérrez-Cáceres, C.A. Beltrán-Castañón, J. Ramirez-Ticona. Computer vision grading system for physical quality evaluation of green coffee beans. Proceedings of the 2016 XLII Latin American Computing Conference (CLEI). Condori R.H.M., J.H.C. Humari, C.E. Portugal-Zambrano, J.C. Gutiérrez-Cáceres. Automatic classification of physical defects in green coffee beans using CGLCM and SVM. Proceedings of the 2014 XL Latin American Computing Conference. Apaza R.G., C.E. Portugal-Zambrano, J.C. Gutiérrez-Cáceres, C.A. Beltrañón. An approach for improve the recognition of defects in coffee beans using retinex algorithms. Proceedings of the 2014 XL Latin American Computing Conference. Ramirez-Ticona J., J.C. Gutiérrez-Cáceres, C.E. Portugal-Zambrano. Cell-phone based model for the automatic classification of coffee beans defects using White Patch. Proceedings of the 2016 XLII Latin American Computing Conference (CLEI). Carrillo E., A.A. Peñaloza. Artificial Vision to assure Coffee-Excelso Beans quality. Proceedings of the 2009 Euro American Conference on Telematics and Information Systems(EATIS). Gonzalez R.C., R.E. Woods. Digital Image Processing 3rded. Pearson Education Ltd. Jarvis R.A. On the identification of the convex hull of a finite set of points in the plane. Information Processing Letters, 2(1973): 18-21 Pudil P., J. Novovic ̌ová, J. Kittler. Floating Search Methods in Feature Selection. Pattern Recognition Letters, 15(1994): 1119-1125. Mahalanobis P.C. On the Generalized Distance in Statistics. 1936. Iglewicz B., D. Hoaglin. How to Detect and Handle Outliers. ASQC Quality Press, 16(1993): 73-78. Choudhury S.D., T. Tjahjadi. Silhouette-based gait recognition using Procrustes shape analysis and elliptic Fourier descriptors. Pattern Recognition, 45(2012): 3414-3426. Hu M.K. Visual pattern recognition by moment invariants. IRE Transactions on Information Theory, 8(1962): 179-187. 國際照明委員會網站International Commission on Illumination website http://www.cie.co.at/ 不同乾燥與儲存方式的生豆顏色差異。新北市大豐社福館，精品咖啡在我家課程照片。||摘要:||
本研究以七支生豆品項進行選別測試，再針對系統出料結果進行人工複檢。其中，本系統正確判別為瑕疵豆之重量，平均為人工複檢總瑕疵豆重量的79 %。本系統誤判為瑕疵豆的樣本重量平均為投入選別總重的1.6 %，於實際應用上可忽略此誤判數量，本系統生豆選別能有效降低人工選別的部分勞力成本。
In this study, an auto-sorting system was developed to detect washed green coffee beans with defects. Defects on green coffee beans are caused during the processes of harvesting, drying, hulling, and transporting. Beans with severe defects can be removed using a color-sorting machine. However, the sorting of high-quality coffee beans has to be conducted manually. Therefore, we developed a sorting system with an aim to reduce labor costs. The system performs functions such as auto-feeding, image capturing, color training and classification on a batch of green beans.
In this study, image processing techniques were used to extract contour features and color features based on the appearance of green coffee beans. Moreover, a method that detects defects caused to the beans by insects ( insect defect detection method) was developed. Top and bottom images of each bean were captured using cameras. First, 16 shape features were extracted, and coffee beans were classified into chipped defective beans and not chipped beans by using a back-propagation neural network. Second, the insect defect detection method was used to detect holes and very dark spots on beans. Third, color features were extracted from 8-bit gray images and RGB color images. Cluster-color features were extracted by applying K-means clustering on images in the CIELAB color space and by retrieving mean values from the corresponding areas in RGB images. In the offline training mode, clustering color features were used to train a new Bayes classifier for detecting defective coffee beans in various batches of coffee beans. In the online sorting mode, clustering color features were used to estimate the level of partial sourness and partial blackness in beans by using a pretrained Bayes classifier.
The auto-sorting device was tested for seven batches of green coffee beans, and samples were rechecked manually. Based on the sorting results, the weight of defective beans obtained after sorting with the proposed device was on average 79% that of the defective beans obtained through manual sorting. The verified false alarm rate of defective beans was less than 1.6% for the total weight on average.
|Appears in Collections:||生物產業機電工程學系|
Show full item record
Files in This Item:
TAIR Related Article
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.