近紅外線光譜的波長選擇對水稻品種鑑別的影響

The Effect of Wavelength Selection of Near Infrared Spectra on Classifying Paddy Rice

劉昌群(Chang-Chun Liu) ； 蕭介宗(Jai-Tsung Shaw) ； 彭敬益(Keen-Yik Poong) ； 洪梅珠(Mei-Chu Hong) ； 沈明來(Ming-Lai Shen)

《農業機械學刊》 14卷2期 (2005/06) Pp. 27-38

https://doi.org/10.30062/JAM.200506.0003

摘要

本研究探討五種水稻的近紅外線吸收光譜波長的選擇以該波長所對應之吸收光譜值做爲變數對品種鑑別的影響。所使用的三種變數選擇方法分別爲逐步排除法、變數之間的相關矩陣法，以及變數在主成分軸上的Loading值法。以全部351個變數建立判別分析以及倒傳遞類神經網路鑑別模式，平均鑑別率分別爲99.69%及97.69%。以逐步排除法選取62個變數建立判別分析以及倒傳遞類神經網路鑑別模式，平均鑑別率分別爲98.0%及92.76%。以變數之間的相關性選取62個變數建立判別分析以及倒傳遞類神經網路鑑別模式，平均鑑別率分別爲90.15%以及84.26%。以變數在主成分上的Loading值選取62個變數建立判別分析以及倒傳遞類神經網路鑑別模式，平均鑑別率分別爲89.38%及82.25%。在波長選擇方法中，以逐步排除法挑選的變數其鑑別率優於相關矩陣法及loading值法所選取的變數，且具顯著差異。所建立的鑑別模式，不僅能減少變數的數目，同時鑑別率仍可達到使用全部變數的準確性。使用相同的變數時，判別分析法的鑑別能力優於類神經網路法，且具顯著差異。

關鍵字

近紅外線 ； 類神經網路 ； 水稻 ； 變數選擇 ； 鑑別

並列摘要

Five varieties of paddy rice were examined using the reflectance spectra corresponding to a selected wavelength from 1100 to 2500 nm in 3-nm steps to determine the classification rate effect. Three hundred fifty-one variables were used to develop the discriminant analysis and neural network models. The average classification rates were 99.69% and 97.69%, respectively. Sixty-two variables were selected using stepwise discrimination to develop the discriminant analysis and neural network models. The average classification rates were 98.0% and 92.76%, respectively. Sixty-two variables were selected using the correlation matrix to develop the discriminant analysis and neural network models. The average classification rates were 90.15% and 84.26%, respectively. Sixty-two variables were selected by loading the first and second principal components to develop the discriminant analysis and neural network models. The average classification rates were 89.38% and 82.25%, respectively. The stepwise discrimination method was more effective in classifying the five varieties of paddy rice using near infrared spectra.

並列關鍵字

Near infrared ； Artificial neural network ； Paddy rice ； Variable selection ； Classification