Classification of Norway spruce drying states on the basis of reflection spectra

The article is devoted to the development of a method for classifying the reflection spectra of spruce needles of different health categories and assessing the quality of the classification. Such health categories of spruces have been chosen that include the initial stages of drying out, which is essential for early detection of disease foci, but makes it difficult to classify according to visual criteria by means of remote sensing of the Earth. An algorithm for untrained classification and visualisation of spectral data based on correlation and cluster analysis is proposed. The reflection spectra of spruce needles obtained under laboratory conditions were studied and the results were interpreted using the developed software. The analysis of various combinations of parameters within the proposed algorithm, as well as combinations of individual components of the algorithm with known classification methods, made it possible to determine the most effective combination of parameters and classification methods (projection of spectra into the space of principal components, elimination of the influence of the first principal component on spectra, the Ward cluster linking metric and the standardised Euclidean metric for calculating the spectral distance) for detection of different stages of spruce disease. Its use to made it possible to increase the F-score classification quality indicator for the 2nd health category (the most important category for the task of detecting drying in the early stages) up to 70.59 %.