Evaluation of an artificial olfactory system for grain quality discrimination

電子鼻系統用于谷物品質評價
S. Balasubramanian, S. Panigrahi,, B. Kottapalli, C.E. Wolf-Hall
Department of Agricultural & Biosystems Engineering, North Dakota State University (NDSU), USA
Department of Agricultural & Biosystems Engineering, NDSU, 1221 Albrecht Blvd., PO Box: 5626, Fargo, ND 58105, USA
cDepartment of Veterinary & Microbiological Sciences, NDSU, USA
dDepartment of Veterinary & Microbiological Sciences, NDSU, USA
Received 25 December 2005; received in revised form 19 December 2006; accepted 21 December 2006

Abstract
A commercially available Cyranose-320TM conducting polymer-based electronic nose system was used to analyze the headspace from stored barley samples. Three types of barley samples were analyzed, namely, clean barley, naturally Fusarium infected barley and Fusarium inoculated clean barley. The barley samples were stored at moisture contents of 13, 18, 20 and 25 g of water/100 g sample. The raw signals obtained from the electronic nose system were pre-processed by various signal-processing techniques to extract area-based features. Principal component analysis was subsequently performed on the processed signals to further reduce the dimensionalities. Classification models using linear (LDA) and quadratic discriminant analyses (QDA) were developed using the extracted features. The performance of the developed models was validated using leave-1-out cross validation and bootstrapping method. The models classified the barley samples stored into two groups based on the ergosterol content, i.e., ‘‘acceptable’’ (ergosterol content o3.0 mg/g) and‘‘unacceptable’’ (ergosterol content X3.0 mg/g). Overall, the total maximum classification accuracy obtained was 86.8% by both LDA and QDA when leave-1-out cross-validation was used. By bootstrapping validation the maximum total classification accuracy obtained was 86.4% and 86.1% respectively, by QDA and LDA. The study proves that there is potential in using an electronic nose system for indicating mold spoilage in stored grains, and necessitates future studies in this direction.
@2007 Swiss Society of Food Science and Technology. Published by Elsevier Ltd. All rights reserved.

使用Cyranose-320導電聚合物基電子鼻系統分析了儲存的大麥樣品。分析了三種大麥樣品，即干凈的大麥、自然鐮刀菌感染的大麥和鐮刀菌接種干凈的大麥。大麥樣品以13、18、20和25 /100 g樣品的含水量儲存。這個從電子鼻系統獲得的原始信號通過各種信號處理技術進行預處理，以提取基于面積的信號。特征。隨后對處理后的信號進行主成分分析，以進一步減小尺寸。利用提取的特征建立了基于線性(LDA)和二次判別分析(QDA)的分類模型。這個利用左1出交叉驗證和自舉方法對所開發模型的性能進行了驗證。分類的模型大麥樣品根據麥角固醇含量分為兩組，即“可接受”(麥角固醇含量O3.0 mg/g)和“不可接受”(麥角固醇含量x3.0 mg/g)。總體而言，兩個LDA獲得的大分類準確率為86.8%。使用“1-out”交叉驗證時的qda。通過引導驗證，獲得大的總分類精度 qda和lda分別為86.4%和86.1%。研究證明，使用電子鼻系統指出儲存顆粒中的霉菌變質，這一方向的研究勢在必行。