Onion sour skin detection using a gas sensor array and support vector machine

電子鼻技術用于洋蔥采后染病揮發性氣體檢測

Changying Li ， Ron Gitaitis， Bill Tollner，Paul Sumner ， Dan MacLean

Received: 13 April 2009 / Accepted: 21 July 2009 / Published online: 7 August 2009
Springer Science+Business Media, LLC 2009

Abstract

Onion is a major vegetable crop in the world. However, various plant diseases, including sour skin caused by Burkholderia cepacia, pose a great threat to the
onion industry by reducing shelf-life and are responsible for significant postharvest losses in both conventional and controlled atmosphere (CA) storage. This study investigated a new sensing approach to detect sour skin using a gas sensor array and the support vector machine (SVM). Sour skin infected onions were put in a concentration chamber for headspace accumulation and measured three to six days after inoculation. Principal component analysis (PCA) score plots showed two distinct clusters formed by healthy and sour skin infected onions. The MANOVA statistical test further proved the hypothesis that the responses of the gas sensor array to healthy onion bulbs and sour skin infected onion bulbs are significantly different (P\0.0001). The support vector machine was employed for the classification model development. The study was undertaken in two phases: model training and cross-validation within the training datasets and model validation using new datasets. The performances of three feature selection schemes were compared using the trained SVM model. The classification results showed that although the six-sensor scheme (with 81% sensor reduction) had a slightly lower correct classification rate in the training phase, it significantly outperformed its counterparts in the validation phase (85% vs. 69% and 67%). This result proved that effective feature selection strategy could improve the discrimination power of the gas sensor array.
This study demonstrated the feasibility of using a gas sensor array coupled with the SVM for the detection of sour skin in sweet onion bulbs. Early detection of sour skin will help reduce postharvest losses and secondary spread of bacteria in storage.

洋蔥是世界上主要的蔬菜作物。然而，各種各樣的植物疾病，包括由洋蔥伯克霍爾德菌引起的酸性皮膚病，對洋蔥行業通過縮短貨架期，在傳統和控制氣氛(CA)的儲存中對收獲后的重大損失負責。研究了一種利用電子鼻和支持向量機(SVM)檢測酸性皮膚的新方法。酸性皮膚感染的洋蔥放在濃縮室中頂空積累，接種后3-6天進行測定。主成分分析(PCA)評分圖顯示健康和酸性皮膚感染的洋蔥形成了兩個不同的集群。Manova統計檢驗進一步證明了氣體傳感器陣列對健康洋蔥鱗莖和酸性皮膚感染洋蔥鱗莖的響應存在顯著差異的假設(p\0.0001)。采用支持向量機進行分類模型的開發。研究分為兩個階段進行：培訓數據集中的模型培訓和交叉驗證以及使用新數據集的模型驗證。利用訓練后的支持向量機模型對三種特征選擇方案的性能進行了比較。分類結果表明，雖然六傳感器方案(傳感器減少81%)在培訓階段的正確分類率略低，但在驗證階段(85%對69%和67%)明顯優于對應方案。結果表明，有效的特征選擇策略可以提高電子鼻的識別能力。

本研究證明了利用電子鼻結合支持向量機檢測甜洋蔥鱗莖酸性表皮的可行性。及早發現酸性皮膚將有助于減少采后損失和儲存中細菌的二次傳播