Methods of increasing the accuracy of classification of agrophytocenoses by satellite images

Abstract

The article is devoted to the problem of the agrophytocenoses classification with satellite observation data and noise filtration methods. The model of agricultural crop types determination with multispectral data and the support vector machine method was conducted. The modified algorithm for agricultural crops classification approbation was carried out on the satellite imageries of the “Steppe” farm of the Kamyansko-Dniprovsky district of the Zaporizhzhya region. Ground-based reference data on crop types, normative base of agricultural development phases for heavy-loamy and medium-loamy soils characteristics of the southern Ukraine steppe zone and remote sensing data of the studied territory gathered by Landsat ТМ and ETM+ satellites for 2001 and 2003 were used for digital processing. It was investigated that a significant number of false-detected pixels (>5%) belong to the interface zones (field boundaries) and chaotic scattered segments of incorrectly classified crops within the field (≈12). This error is mainly due to spatial variations in soil moisture, plant health, type of projective surface, cultures spectral similarity, phenological stage of plants development, time of satellite observation. Median and rank filtration methods were used to minimize the effect of the above-mentioned problems on the final classification result. The method of median filtration allowed to increase the classification accuracy up to 3%, and the rank filtration method up to 2% compared with the simple procedure of the support vector machine method. The classification error and the disadvantages of filtration are largely due to the interface zones of the different crops, the fields boundaries. Since the spatial resolution is 30×30 meters, it would be advisable to reject the pixels of the transition (on the boundary) from the training sample to identify more accurate classification model.

Keywords

• remote sensing
• estimation
• vegetation state
• classification
• filtration