Fig. 4
From: A bimodal image dataset for seed classification from the visible and near-infrared spectrum
Example of image processing pipeline for BiSID-5k dataset24. The pipeline involves segmentation of both RGB and hyperspectral (HS) modalities. Initially, a white square box is detected in both modalities. The region containing the seed inside the white box is cropped. Subsequently, the Segment Anything Model (SAM)2 with a ViT-B backbone is applied to segment the image, isolating the object likely near the centre. Finally, a bounding box is drawn around the identified object and subsequently cropped. Notably, unlike the RGB modality, segmentation for the HS modality is performed on synthetic RGB images derived from the hyperspectral cubes. The segmentation results are extrapolated to the original hyperspectral image.
