Identification of the Spectral Patterns of Cultivated Plants and Weeds: Hyperspectral Vegetation Indices

Abstract

The accurate recognition of weeds on crops supports the spot application of herbicides, the high economic effect and reduction of pesticide pressure on agrocenoses. We consider the approach based on the quantitative spectral characteristics of plant objects to be the most appropriate for the development of methods for the spot application of herbicides. We made test plots with different species composition of cultivated and weed plants on the experimental fields of the scientific crop rotation of the Federal Research Center of Biological Plant Protection. These plants form the basis of the agrocenoses of Krasnodar Krai. Our primary subjects are sunflower crops (Helianthus annuus L.), corn (Zea mais L.) and soybean (Glycine max (L.)). Besides the test plots, pure and mixed backgrounds of weeds were identified, represented by the following species: ragweed (Ambrosia artemisiifolia L.), California-bur (Xanthium strumarium L.), red-root amaranth (Amaranthus retroflexus L.), white marrow (C. album L.) and field milk thistle (Sonchus arvensis L.). We used the Ocean Optics Maya 2000-Pro automated spectrometer to conduct high-precision ground-based spectrometric measurements of selected plants. We calculated the values of 15 generally accepted spectral index dependencies based on data processing from ground hyperspectral measurements of cultivated and weed plants. They aided in evaluating certain vegetation parameters. Factor analysis determined the relationship structure of variable values of hyperspectral vegetation indices into individual factor patterns. The analysis of variance assessed the information content of the indicators of index values within the limits of the selected factors. We concluded that most of the plant objects under consideration are characterized by the homogeneity of signs according to the values of the index indicators that make up the selected factors. However, in most of the cases, it is possible to identify different plant backgrounds, both by the values of individual vegetation indices and by generalized factorial coefficients. Our research results are important for the validation of remote aerospace observations using multispectral and hyperspectral instruments.