Hyperspectral Imaging and Machine Learning for Automated Pest Identification in Cereal Crops.

The spectral characteristics of harmful insect pests in wheat fields were characterised using hyperspectral imaging for the first time. The analysis of spectral profiles revealed that reflectance is determined by the structure of the insect's chitin and the colouration of its body surface. Insects with lighter or more vivid colours (white, yellow, or green) showed higher reflectance values compared to those with predominantly dark pigmentation. Reflectance was also influenced by the presence of wings, surface roughness, and the age of the insect. Each species exhibited distinct spectral patterns that allowed for differentiation not only from other insect species but also from the plant background. A classification model using PLS-DA was developed and demonstrated high accuracy in identifying 12 pest species, confirming the strong potential of hyperspectral imaging for species-level classification. The results validate the PLS-DA method for differentiating insects based on spectral characteristics and underscore the reliability of this approach for automated monitoring systems to detect phytophagous pests in crop fields. This technology could reduce insecticide use by 30-40% through targeted application. The research has both scientific and economic significance, laying the groundwork for integrating machine learning and computer vision into agricultural monitoring. It supports the advancement of precision farming and contributes to improved global food security.