The Novel Camouflaged False Color Composites for the Vegetation Verified by Novel Sample Level Mirror Mosaicking Based Convolutional Neural Network

Abstract

Remote sensing is the analytics of sensor data modalities to capture the earth's surface characteristics. The hyperspectral data widely used for surface material identification by using pixel-wise unique signature patterns. The true-color-composite (RGB) or/and a variety of false-color-composites (FCCs) used to classify various objects and features. In this paper, three novel FCCs have been proposed and compared with already existing popular FCCs. These FCCs have been analyzed using three different approaches viz., (i) k-means (ii) patch-based deep network and (iii) sample level mirror mosaicking (SLMM)-based deep network; for the classification of various objects or features viz., Vegetation, Soil, and Road. The open-source dataset provided by the National Ecological Observatory Network (NEON) has been used to show the efficacy of proposed FCCs and SLMM-based deep-network. Our proposed FCCs and SLMM-based deep networks outperform over all other considered FCCs and classification methods.