A Band Grouping Based Approach for Phenotype-Class Mapping of Tree Genotypes Using Spectro-Temporal Information in Hyperspectral Time-Series UAV Data

Modern genomic tree breeding studies and forest health monitoring programs demand accurate mapping of tree phenotype. However, conventional field-based approaches to map phenotype are costly in terms of time and money. The recent phenomenal advances in the low-flying Unmanned Airborne Vehicle (UAV) remote sensing platforms together with the availability of high-resolution hyperspectral cameras allow to periodically capture a huge amount of crown spectral details of individual trees. These details can be exploited to map phenotypic class of tree genotypes. State-of-the-art methods that maps tree phenotype often underexploit the information in hyperspectral time-series data by using only a few specific band-based remote sensing (RS) indices to model phenological tree parameters that define the phenotypic response. Thus, we propose a wavelet-based approach to map tree phenotype of trees that a) maximally exploits the spectral and temporal information in band-groups by addressing data redundancy problem, and b) uses spectro-temporal/phenological information in the hyperspectral time-series data to map phenotype class of tree genotype. The improved performance of the proposed method over a RS index-based state-of-the-art one to map trees to a phenotypic class, on a set of 100 trees from 10 genotypes, proves the method to be performing.