Second, the mapping between the texture-structure cube of the input HSI and that of the desired HSI is learnt via two parallel subnetworks. Finally, the deep learnt texture cube is interfused with the ...