Hyperspectral Data Compression Using Fully Convolutional Autoencoder

In space science and satellite imagery, better resolution of the data information obtained makes images clearer interpretation more accurate. However, huge volume gained by complex on-board instruments becomes a problem that needs to be managed carefully. To reduce stored transmitted on-ground, signals received should compressed, allowing good original source representation in reconstruction step. Image compression covers key role imagery and, recently, deep learning models have achieved remarkable results computer vision. this paper, we propose spectral compressor network based on convolutional autoencoder (SSCNet) conduct experiments over multi/hyperspectral RGB datasets reporting improvements all baselines used as benchmarks than JPEG family algorithm. Experimental demonstrate effectiveness ratio signal robustness with type greater 8 bits, clearly exhibiting using PSNR, SSIM, MS-SSIM evaluation criteria.