Spatial filtering in remote sensing involves applying a kernel (a moving window) over the image to modify pixel values based on the values of their neighbors. This technique selectively preserves or suppresses certain spatial frequencies in the image, which can enhance particular features like edges or reduce noise.