Automatic Object Based Plant-level Crop Segmentation in Drone-based Hyperspectral Imagery

In today’s technologically advanced world, precision agriculture plays a pivotal role in enhancing crop yields and minimizing environmental impact by employing data-driven, site-specific farming practices. Agricultural fields can be closely and effectively monitored utilizing drone-mounted hyperspectral sensors with improved spatial, spectral, and temporal resolution. Drone-based spectral imaging has enabled the undertaking of studies aimed at field-level agriculture. Although research on field-level crop monitoring and mapping is well in progress, analyzing crop and soil environment dynamics at finer scales, such as within-field patches or individual plant levels, has yet to be addressed. Object-based image analysis (OBIA) is an approach for detecting objects with greater precision than traditional pixel-based technologies by utilizing both spectral and spatial information. Dimensionality reduction is critical in this context, as it aids in selecting the most discriminating bands for effective segmentation, the primary task of OBIA. In this study, we evaluate various segmentation approaches for discriminating crops in drone-based hyperspectral imagery at the individual plant level, followed by the classification of segmented objects.