應用機器視覺於火鶴花切花自動分級系統之研究

A Study Of Automatic Anthurium Cut Flower Grading System With Machine Vision

林皇杉(Huang-Shan Lin)

指導教授 ： 黃惠藩

亞洲大學/資訊學院/資訊工程學系碩士班/碩士(2008年)

摘要

火鶴花是台灣第二大的出口花卉，從2001年到現在，共有3161公噸的火鶴花出口到世界上各個國家，對台灣的花卉產業創造了1866萬美元的產值，在火鶴花出口之前要先進行火鶴花的分級及包裝，且火鶴花不可有破損或是污點，現在的火鶴花分級篩選還是依靠人力，因此需要一套火鶴花切花的自動分級系統。
本研究提出一套機器視覺組成的自動分級系統，利用彩色影像分割及特徵分析的技術來量測火鶴花切花花瓣的寬度，並針對市面上的分級標準予以分級，並加入破損偵測及表面汙點偵測來協助分級標準的判斷。在後端加入了自動出料機構的設計，讓火鶴花的分級達到自動化的目的。
本研究的離線測試利用100朵紅色火鶴花切花作為樣本，針對人工量測及電腦量測的結果分析，兩者的誤差率僅2.3%。花瓣破損的偵測可達97%，花瓣上的汙點偵測可達86%，線上測試對影像分析的準確度可達98%，自動出料的正確率在50次出料作業中，出料準確率可達92%。

關鍵字

火鶴花 ； 機器視覺 ； 自動分級 ； 自動出料

並列摘要

Anthurium is the second largest flower export in Taiwan. Since 2001, about 3161 metric tons of anthurium cut flowers have been exported to various countries around the world, mainly Japan, and creating over 18 million US dollars revenue for Taiwan’s flower and plant industries. Before each anthurium cut flower is exported, it is graded for size and sheathed with plastic bag. It must also be free of cuts and bruises, and cannot contain any surface defects. At the current stage, anthurium cut flower grading and inspection is performed manually, which is very time consuming and inconsistent. Therefore, there is a need for an objective and automatic grading and inspection system for anthurium cut flowers.
We propose a system that uses machine vision algorithm for grading of anthurium cut flowers. Color image segmentation and blob analysis techniques are used to measure the spathe width of anthurium cut flowers and the measurements are then used to grade the flowers according to official grading standard. A cut detection algorithm and a surface defect classification scheme are also proposed to facilitate the detection of cuts and surface detects on the flowers. An automatic discharging mechanism is designed and added at the backend of the system thus making the grading system fully automatic.
Of the 100 flower samples used in the off-line experiment, the average measurement error of spathe width for the machine vision system against manual measurement is 2.3%. The overall accuracy of the cut detection algorithm on the 100 sample is 97%. The surface defect classification algorithm is able to identify all of the defect samples but reports 14 false detections on the good samples, resulting in an overall accuracy of 86%. In the on-line experiment, the grading accuracy of the machine vision system is 98%, and the success rate of the discharging mechanism is 92% based on 50 discharge operations.

並列關鍵字

Anthurium ； machine vision ； automatic grading ； automatic discharging

參考文獻

[10] 張文宏、陳世銘、連豐力、許豐益、謝廣文，“水果智慧型選別之研究”，農業機械學刊，第三卷，第四期，1994年，25-35頁。

[11] 楊清富、李芳繁，“應用機器視覺進行蕃茄顏色分級之研究”，農業機械學刊。第三卷，第一期，1994年，15-29頁。

[12] 柯建全、黃膺任、艾群，“應用影像處理檢測荔枝果皮之顏色變化”，農業機械學刊，第八卷，第一期，1999年，59-68頁

[14] 李芳繁，蔡玉芬，“以機器視覺與類神經網路分級玫瑰花沏花之研究”，農業機械學刊，第六卷，第一期，1997年，57-69頁。

[15] 張文宏、陳世銘，“以機器視覺引導機器人選別水果”，農業機械學刊，第二卷，第三期，11-24頁。