ASABE Technical Library - Abstract

Abstract. In the last decades certain fungal infections caused great losses for the mushroom industry, the most significant pathogenic agents are Cladobotryum dendroides and Trichoderma aggressivum. The industry needs a proper method against these diseases, with the detection of the pathogenic fungi and with the selection of the infected mushrooms the risk of the infection can be mitigated.

Hyperspectral imaging is an exciting new technique, it is the combination of spectroscopy and imaging, therefore it can provide information about the composition of the object and about the distribution of the components. As a consequence it became more popular in agricultural applications too.

Previous studies demonstrated that the presence of Cladobotryum and Trichoderma species can be detected using hyperspectral imaging method.

The aim of this work was to select the most significant wavelength values and find a reliable classification method to detect the infected samples at the earliest stages.

White button mushroom (Agaricus bisporus) samples were inoculated with Cladobotryum dendroides and Trichoderma aggressivum (respectively). The development of the infection was followed by hyperspectral imaging method.

Images were segmented and the spectra were pre-processed using Savitzky-Golay smoothing.

The data of the spectra were processed with principal component analysis and the most significant wavelength values were selected to discriminate between infected and control samples. Discriminant analysis and support vector machine classification methods were applied on the data of the selected wavelength.

The infected samples were clearly discriminated from control samples, the results of the Cladobotryum and Trichoderma infected samples showed less effective discrimination.