參數選擇對水稻品種鑑別的影響

The Effect of Parameter Selection 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1期 (2005/03) Pp. 17-32

https://doi.org/10.30062/JAM.200503.0003

摘要

探討五種水稻的外觀形狀及顏色作為參數，以倒傳遞類神經網路訓練鑑別模式對品種鑑別的影響，使用全部60個參數，平均驗證與鑑別率分別為92.24%與92.0%。選出在第一主成分軸上的Loading值較大的50個參數，平均驗證與鑑別率分別為91.77%與90.0%。依參數的相關矩陣選取30個相關性較低的參數，平均驗證與鑑別率分別為89.18%與91.4%。以參數對網路訓練的影響度，由大到小挑選20個參數，平均驗證與鑑別率分別為90.59%與91.8%。在參數的選擇方法中，以參數對網路訓練的影響度，挑選的參數所建立的鑑別模式，不僅可使用較少的參數，同時也具有較佳的穩定性。

關鍵字

機器視覺 ； 類神經網路 ； 水稻 ； 參數選擇 ； 鑑別

並列摘要

Considering different shape factors and color intensities of five varieties of paddy rice as parameters, four models were established by a back-propagation neural network and were used to study the validation and classification rates affected by choosing different parameters. With 60 parameters, the average validation and classification rates were 92.24% and 92.0% respectively. If the most effective 50 parameters were chosen by loading values in the first principal component, the average validation and classification rates were 91.77% and 90.0% respectively. 30 parameters selected from the correlation coefficient matrix to build up the model, the average validation and classification rates were 89.18% and 91.4% respectively. If the most effective 20 parameters were chosen from model training, the average validation and classification rates were 90.59% and 91.8% respectively, which could be the best model for classifying due to its less parameters and better stability.

並列關鍵字

Machine vision ； Artificial neural network ； Paddy rice ； Parameter selection ； Classification