Abstract: The hyper-spectral image compression is focused on compression quality and time
owing to the application fields of hyper-spectral image. The application of BP Neural Network in
the hyper-spectral image compression can get better compression quality. A joint-optimized
improved algorithm on the premise of good renewed qualities is presented. It uses Cauchy error
estimator and shape factor in Sigmoid function, upgrades the differential coefficient and each lay
adapts different step adjustment. Simulation results prove that the algorithm can reduce
compression time and improve coding efficiency.

Key words: image compression, hyper-spectral image, neural network, BP algorithm