thanks for your quick reply,
I test this approach, and fix the fish eye as the master channel.
The 5 others bands (frame type) have very low overlapping and I have very bad results on some CPs (others are ok)
Placing the GCPs for in the fish-eye images is quite imposible...
On an other hand, i test a project with one chunck ans two sensors : a fish eye and a normal frame (both in the green band), the accuracy of the alignement is really good. The problem is that I have the orientation of juste 2 sensors. But using a melling fourier transform I find the translation rotation and scaling between the different sensors.
So i think I'll give up the multispectral project.
What is exactly the master channel ? Its looks like it is not posible to do a multispectral project using different frames (fish-eye, normal, spherical .. ) Am I right ?