Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/35351
標題: A Study of the Geometric Characteristics and Diseases for Phalaenopsis Seedlings with Machine Vision
應用機器視覺於蝴蝶蘭大苗幾何特徵與病害檢測
作者: 黃國益
Huang, Kuo-Yi
關鍵字: machine vision
機器視覺
Phalaenopsis Seedlings
disease
sorting mechanism
蝴蝶蘭大苗
病害
選別機構
出版社: 農業機械工程學系
摘要: 蝴蝶蘭為近年來最受歡迎的花卉之一,台灣為全球蝴蝶蘭苗最大的產銷國,為了使得蝴蝶蘭生產作業邁向自動化,提升台灣蝴蝶蘭產業的競爭力,本研究應用機器視覺建立蝴蝶蘭大苗選別系統,該系統包括幾何特徵之估算模組、病害檢測模組、選別機構等。為了建立完整系統,相關的研究範圍包括以下四部分 : 1、建立蝴蝶蘭大苗幾何特徵估算法則 (1)利用影像處理技術,建立估算蝴蝶蘭大苗幾何特徵之演算法則。 (2)撰寫幾何特徵之估算程式。 2、建立蝴蝶蘭病害檢測法則 (1)利用影像處理技術,建立蝴蝶蘭病害檢測法則。 (2)撰寫病害檢測程式。 3、建立蝴蝶蘭大苗選別系統 (1)設計製造蝴蝶蘭大苗選別機構。 (2)建立選別控制系統,整合影像處理及控制大苗選別系統程式。 4、建立蝴蝶蘭大苗選別系統軟體 (1)建立影像處理函式庫。 (2)撰寫使用者指南。 (3)撰寫指令參考資料。 本論文應用影像處理技術,建立蝴蝶蘭苗幾何特徵估算法則,包括苗莖中心與葉片端點搜尋法、葉片數估算法、苗盆與葉片分離法及葉片輪廓萃取法。針對44株蝴蝶蘭大苗進行估算試驗,同時以人工方式進行量測,人工量測所得之數據為正確值,針對估算的結果與實際量測結果進行比較,其中以葉片數之估算最為準確,其平均相對誤差為1.48%,其它特徵值的平均相對誤差分別為葉長為1.80%、葉幅為2.44%、葉片夾角為3.90%、葉寬為4.02%及葉片長寬比為7.04%。 本論文提出蝴蝶蘭病害檢測法則針對蝴蝶蘭之軟腐病、褐斑病及疫病病害進行檢測與分類。首先利用Rayleigh轉換與影像處理技術萃取蝴蝶蘭病害區域之特徵,這些特徵包括形心座標、面積、周長、平均直徑及平均灰階值 、 、 ,再進一步利用檢測線法則估算病害區域之最大灰階平均值 與灰階值陡峭係數 ,最後以 、 、 、 及 為特徵向量,利用貝氏分類法求得決策邊界 、 、 、 及 ,以這些決策邊界為判斷依據,針對軟腐病、褐斑病、疫病害區域進行檢測與分類。針對144個病害測試樣本進行檢測與分類,平均分類正確率為88.2%,每一株大苗的平均處理速度為1.78 sec。若僅考慮是否檢測出病害,而不論其屬於何種病害,則本系統之病害檢測能力可達到96.5%。 本論文完成蝴蝶蘭大苗選別系統之研製,該系統主要由四部份組成:(1)取像定位機構、(2)影像辨識系統、(3)分級顯示裝置、(4)選別控制系統。此選別系統作業方式係將蝴蝶蘭大苗藉由置料平台與升降機構送至取像平台,利用影像辨識系統估算蝴蝶蘭大苗的幾何特徵及檢測病害,並根據台糖外銷蝴蝶蘭大苗的選別標準進行品質選別。針對430株蝴蝶蘭大苗進行選別試驗,由試驗結果顯示,機器選別的正確率為90.0 %,每一株蝴蝶蘭大苗的平均選別作業時間為21.15 sec;人工選別的正確率則為97.2 %,每一株蝴蝶蘭大苗的平均選別作業時間為27.42 sec;機器選別的速度較人工選別約快22.3 %。 本論文完成蝴蝶蘭大苗選別系統軟體SSPS 1.0之開發,並建立影像程式庫,針對每一個副程式及函式的功能進行說明,可提供使用者參考。
Phalaenopsis is getting popular recently. In order to ensure competition in world market for Taiwanese floristic industry, an automatic production line is a key factor. In this study we used the image processing techniques to develop a sorting system for Phalaenopsis seedlings. The sorting system consists of the module of estimation of the geometric characteristics, the module of diseases detection, and the sorting machine. The scope of this research to develop the whole system includes: 1. Developing an algorithm for estimating characteristics of Phalaenopsis seedlings: The algorithm is developed to estimate the geometric characteristics of Phalaenopsis seedlings using the image processing techniques. 2. Developing an algorithm for detecting diseases of Phalaenopsis seedlings: The disease detection algorithm is established to detect bacterial soft rot (BSR), bacterial brown spot (BBS), and Phytophthora black rot (PBR) using image processing techniques. 3. Developing a sorting system: A sorting system for Phalaenopsis seedlings was designed and manufactured. The sorting system is composed of the sorting mechanism and control system. 4. Developing a software for the sorting system: The software includes: (1) image processing functions, (2) user guide, and (3) command reference. A methodology using machine vision to estimate the geometric characteristics of Phalaenopsis seedlings was established in this paper. The image processing techniques including the stem-center search method, the leaf-endpoint search method, the leaf number search method, the pot removing method, and the leaf-shape extraction procedure were applied to develop the algorithms that were used to estimate the geometric characteristics. Forty-four samples were investigated. Measurements taken manually and from estimation using our method were obtained and compared. The average relative errors between estimated values and measured results were 1.48% for the total number of leaves, 1.80% for the length, 2.44% and 3.90% for the span and the angle between two upper leaves, 4.02% for the width, and 7.04% for the length/width ratio. A novel system for detecting and classifying Phalaenopsis seedling diseases, including BSR, BBS, and PBR, was developed. The features of the lesion area of a Phalaenopsis seedling were extracted by Rayleigh transform and image processing techniques, such as hole-filling, erosion, dilation, opening, and closing operators. The detection line algorithm (DLA) was used to evaluate the lesion area. Five color features - Rmean, Gmean, Bmean, Gmax, and M were used in the classification procedure. A Bayes classifier was applied to classify BSR, BBS, and PBR of Phalaenopsis seedlings. One hundred and forty-four samples were used to evaluate the system. The methodology rapidly detected and classified these three Phalaenopsis seedlings diseases, at 1.78 sec/pot, to an accuracy of 88.2%. The disease detection capability of the system, without classifying the disease type, was as high as 96.5%. A sorting system for Phalaenopsis seedlings was designed and manufactured. This sorting system consists of four major parts: (1) image grabbing and positioning mechanism, (2) pattern recognition system, (3) display panel, and (4) control system. Four hundred and thirty pots of Phalaenopsis seedlings were used to test the sorting system. According to the results, we were able to achieve a rapid sorting of 21.15 sec/pot compared to 27.42 sec/pot by manual sorting, to an accuracy of 90.0% compared to 97.2% when sorting manually. Our machine can save up to 22.3% of the time used for manual sorting. The sorting system for Phalaenopsis seedling (SSPS) software 1.0 was developed. The SSPS 1.0 library had been established. The sub-programs and functions were described in the SSPS 1.0 reference manual.
URI: http://hdl.handle.net/11455/35351
Appears in Collections:生物產業機電工程學系

文件中的檔案:

取得全文請前往華藝線上圖書館



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.