Identifying optimal waveband positions for discriminating Parthenium hysterophorus using hyperspectral data

Parthenium hysterophorus (an alien weed) is posing serious threat to crop yields, livestock's and human health and is considered the seventh most devastating weed across the globe. Early identification and mapping of Parthenium hysterophorus is essential for the timely eradication and control of this alien invasive species. Hyperspectral remote sensing (due to its high spectral details) is highly valuable in discriminating vegetation species and mapping its distribution. However, the use of high-dimensional hyperspectral data possesses the threat of multi-collinearity (i.e., contiguous wavelength-bands exhibits strong spectral correlation) which in-turn yields unstable parameter estimation. This study aims to explore the potential of hyperspectral (spectroscopic) data for discerning Parthenium hysterophorus and to identify optimal wavebands that are sensitive for species discrimination. In this study, the spectral signatures of Parthenium hysterophorus and four co-occurring plant species were acquired using portable hand held spectrometer. Spectral Angle Mapper (SAM) in conjunction with Genetic Algorithms (GA) were used discern the measured species based on their spectral profiles. The analysis yielded high classification accuracies for both the training (overall accuracy = 99%) and testing (overall accuracy = 97%) datasets. The SAM-GA picked a meaningful subset of spectral bands from different parts of electromagnetic spectrum (i.e., centering at 0.47 μm, 0.715 μm, 1.12–1.25 μm and 1.8–1.9 μm) which carries highest information for the spectral discrimination of Parthenium hysterophorus. In conclusion, this study confirms the capability of hyperspectral data in discerning Parthenium hysterophorus from other crops/plant species and also highlights the importance of few wavelength positions for the spectral discrimination of Parthenium hysterophorus weed.