Satellite Image Generation

Module: Generative AI | Difficulty: Advanced

Super-Resolution for Satellite

Change Detection Training

Multi-Spectral Generation

Generate RGB + NIR + SWIR simultaneously.

class SatelliteSR(nn.Module):
    def __init__(self, unet, num_bands=4):
        super().__init__()
        self.unet = unet
        self.band_proj = nn.Linear(num_bands, 3)
    def forward(self, lr_multispectral, t):
        rgb = self.band_proj(lr_multispectral.permute(0,2,3,1).permute(0,3,1,2))
        return self.unet(rgb, t)

Research Insight: Satellite image generation must preserve geospatial accuracy. Standard FID metrics are insufficient — metrics like SSIM and spectral angle mapper are needed to ensure spectral fidelity.