WAVELENGTH AND VISUAL FEATURE SELECTION FOR NON-DESTRUCTIVE EVALUATION OF FUSARIUM HEAD BLIGHT IN WHEAT USING HYPERSPECTRAL IMAGING

Abstract

Fusarium head blight (FHB) is a fungal disease affecting the heads of cereal crops and has caused significant economic losses in the Canadian prairie wheat industry. Although wheat breeding efforts to enhance resistance genes are ongoing, efficient and accurate detection of FHB remains critical for breeding programs and crop protection. This thesis presents the development of a custom hyperspectral scanner optimized for seed and kernel data acquisition and evaluates an extensive range of classification models trained on spatial–spectral features. The scanner captured hyperspectral scans of over 1,500 healthy and Fusarium-damaged kernels, from which statistical features were extracted at each wavelength. Classification models were trained to assess the contribution of individual wavelengths and statistical features to overall performance, and an exhaustive combinatorial analysis was conducted to identify high-performing feature pairs. The best-performing model achieved an accuracy of 99.7% using a large feature set. Lightweight models identified through feature analysis achieved accuracies of up to 97.3% and correctly classified up to 99.4% of Fusarium-damaged kernels using as few as two features.

Abstract from the paper’s OpenAlex record, shown for indexing and research. Copyright remains with the authors or the publisher. Read at the source ↗

The text the model read, with the source record’s inline markup (italics, sub- and superscripts) rendered and its character entities decoded; quote offsets refer to the raw text.