Physiognomic vegetation types and their identification by using the decryption of digital images

The algorithm of the physiognomic vegetation types and the dead grass cover and the soil surface decryption using digital images is presented for further quantitative assessment of projective cover. The collection of material was held at the remediation site within Nikopol manganese ore Basin in city Pokrov. As objects of study were chosen following tehnosols: pedozems, sod-lithogenic soils on losses-like loam, on red-brown clay and gray-green clay. The visual analysis of the digital images of the surface areas studied revealed several types of images. This open surface soil, dead plants, grasses, plants Seseli campestre, Lactuca tatarica and legumes. The discriminant analysis allowed to accurately classify these objects by color characteristics. In the whole sample classification accuracy was 65.39%. The analysis only color without spatial context (especially form) reduces the accuracy of classification. In addition, structurally homogeneous object can be represented significant range of color values, reflections, shadows, mutual superposition of different objects, which significantly reduces the quality of classification. The following algorithm of the classification was proposed: 1) it is necessary to conduct cluster analysis (classification without training) a plurality of pixels. The number of clusters established must exceed the number of physiognomic types; 2) to analyze the correspondence between physiognomic types and clusters. Stop at that decision, when each physiognomic type corresponds to at least one cluster; 3) the decision to hold the cluster discriminant analysis, on which to perform differentiation pixels in images (classification of training); 4) conduct a segmentation of the image ‒ to unite in clusters corresponding physiognomic types; 5) evaluate physiognomic structure cover experimental plots. The accuracy of the proposed classification algorithm was 91.66%. The physiognomic types of vegetation can act as quantitative characteristics of the vegetation and can be considered as ecogeographic variables to describe the environmental conditions that other components of the ecosystem.