CNN based Optimized Autoencoder for Satellite Image Compression

This paper introduces an innovative Convolutional Neural Network (CNN) based Optimized Autoencoder for the compression of satellite images, with a focus on high resolution Sentinel-2 and Landsat imagery. In this method, the Satellite images are resized and normalized before an optimal autoencoder architecture is implemented. The encoder compresses the image, while the decoder reconstructs it with minimum quality loss. Convolutional and down sampling layers in the encoder and up sampling layers in the decoder preserve image features. MSE, PSNR, SSIM, and CR were used to assess compression performance. The simulation outcomes show that the CNN-based Autoencoder can retain image quality across images with a high compression ratio of 512, an average MSE of 0.0064, an average PSNR of 25.214 dB, and an SSIM of 0.62. This satellite image compression approach is efficient enough for storage and transmission in satellite data applications.