Spectral Image Mapping for Low-Cost Plant Identification

The patent creates a spectral mapping model that copies spectral information from a single multispectral/hyperspectral reference image to multiple RGB images. The reference camera captures high spectral resolution data, while RGB cameras capture low spectral resolution data. Through image registration and spectral unmixing algorithms, the system copies and transfers spectral characteristics from the reference image to RGB images, enabling virtual hyperspectral imaging without requiring expensive cameras at every nozzle position.

The patent introduces a spectral mapping model as an intermediary between RGB cameras and the desired spectral information. This model includes a spectral unmixing module that acts as a mediator to decompose the spectral information from the reference camera and reconstruct it for RGB camera images. The mapping model serves as an intermediary layer that translates between different spectral resolutions, allowing the system to achieve high spectral resolution output from low spectral resolution input cameras.

The patent segments the imaging area into multiple zones, each covered by a different camera (RGB or reference camera). The system processes images from multiple cameras and integrates them through the spectral mapping model. By segmenting the field into manageable zones and processing them individually, the system can cover a large area using one reference camera and multiple RGB cameras, with each camera responsible for a specific spatial segment.

The patent makes the single multispectral or hyperspectral camera serve multiple functions: it provides spectral reference data for the spectral mapping model, captures images for plants directly in its field of view, and enables spectral enhancement for RGB camera images outside its direct coverage. This multi-functional use of one reference camera maximizes its utility and extends the effective imaging coverage area beyond what a single camera could directly capture.