Early identification of potato storage disease using an array of metal-oxide based gas sensors

• Detection of potato soft rot was explored by means of metal-oxide gas sensors.
• Detection was tested at symptomatic and pre-symptomatic stages.
• A subset of sensors was shortlisted from the original array.
• Unsupervised techniques and predictive models were employed for analysis.
• The metric of ‘sensitivity’ was employed for predictive modelling.

Soft rot is a widespread potato tuber disease that causes substantial losses each year to the UK potato industry. In this work, it was explored the possibility for the early detection and monitoring of this disease by means of gas sensing in a laboratory setting. Potato tubers were inoculated with one type of bacterium which can cause soft rot, Pectobacterium carotovorum, and stored in controlled conditions conducive to rapid disease progression. Two time points were selected for sampling; one for pre-symptomatic early disease detection and the other corresponding to when symptoms of infection first become evident. In both cases, results showed discrimination between uninfected and diseased tubers following analysis of 40 potato tuber samples for each of the two time points with a commercial array of 12 MOX sensors (AlphaMOS Fox3000). A subset of sensors has been identified from the original array while retaining the same results. Data processing was carried out with PCA and k-means clustering for exploratory data analysis, followed by predictive models with LDA, MARS, RBF SVM, Random forests and C5.0. The conditional predictive model metric of sensitivity also been successfully adopted to assess model performance in discriminating between healthy and infected potatoes.