Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/35698
標題: 應用機器視覺於穴盤蔬菜苗之辨識系統
Feature Inspection for Vegetable Seedlings with Machine Vision
作者: 林群翔
關鍵字: 機器視覺;影像處理;傅立葉描述子;輪廓追蹤;蔬菜穴盤苗
出版社: 生物產業機電工程學系
摘要: 
本研究應用機器視覺配合三種不同實驗方法,分別為: 1.檢測圓法則2.面積判別方式3.應用傅立葉傅立葉描述子,針對包心白菜種苗在生長天數上進行判別與比較及針對種苗子葉與本葉特徵進行解析與區別。
本研究利用檢測圓法則針對40 株種苗進行子葉時期葉寬之量測,以累積葉寬來作為生長天數判別之依據,由實驗結果顯示,檢測圓之檢測方式是成立的,但由於檢測圓之設定只針對種苗之子葉時期,其適用性範圍較小,因此檢測圓之實驗方法在本研究中為種苗之初期判別法則。在後續實驗中,針對10~17天之種苗利用累積種苗本葉之葉片面積總像素點來作為生長天數判別之依據,並以隨機取樣之方式,來驗證此方法之效益。經隨機取樣後,以每天取10次,每次取5株之方式,共取80次,套用此法則所得之辨識率可達80%。此外更利用八方鏈碼編碼與輪廓搜尋之方式,針對種苗子葉與本葉特徵進行區別,經驗証後其區別成功率可達79.38%。後續應用傅立葉描述子之概念針對種苗本葉及子葉之邊界輪廓座標序列進行傅立葉轉換,觀察在不同生長天數下,本葉及子葉之邊界頻譜變化,藉由兩者邊界在頻譜上之差異來進行區別。經實驗結果顯示,以子葉邊界頻譜為基準,選擇不同頻率區間所產生之波峰個數作為比較之值,經比較後完整本葉之辨識率在頻率區間0~30為72.9%、區間0~40為66%、區間0~50為57.2%,由辨識結果顯示,隨頻率區間取樣範圍之擴大,其辨識率逐漸下滑。進一步利用本葉輪廓變化顯著之特性,來進行生長天數之判別,並將結果對照葉片面積LUT所得結果,以增加生長天數判定之正確性。

To differentiate the growth days and to distinguish cotyledons from leaves of Chinese cabbage acrospires, this study incorporated machine vision in three experimental methods: 1. circular detection rule, and 2. area determination, and 3. Fourier descriptor
By circular detection rule, the cotyledon width of 40 acrospires was measured with days to investigate their relationships. The results showed that the circular detection rule was tenable. However, the circular detection rule was only applicable to cotyledon period. Therefore, the circular detection method was the judgment rule only in primary stage of acrospires. In the follow-up experiments, the accumulated leaf area of 10 to 17-day acrospires was totally transformed into pixels as a judgment basis of growth days. To assess the effect of this method, five of the samples were randomly chosen and investigated 10 times per day. Totally 80 times of investigation were performed in this study. The identification rate was 82.5% maximally by using circular detection rule. Besides, eight-adjacency chain code and outline searching were applied to distinguish cotyledon from leaf. The identification rate was 79.38% maximally. The outline coordinates of the cotyledons and leaves of acrospires were transformed by fast Fourier transformation to investigate the spectra variation with growth days. The cotyledons and leaves can be identified by their differences in outline spectra. Based on the outline spectra of cotyledons, the peak number in different frequency intervals was compared. The identification rate of intact leaves was 72.9% in channel 0-30, 66% in channel 0-40, and 57.2% in channel 0-50, respectively. The results indicated that the wider the interval of sampling, the lower the identification rate.
URI: http://hdl.handle.net/11455/35698
Appears in Collections:生物產業機電工程學系

Show full item record
 

Google ScholarTM

Check


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