Why do you want a bit depth of 32? It seems a bit extreme.
I believe the NI resampling routines will maintain the existing 10 bits of grayscale levels.
The highest bit depth NI images can handle is 16 bits, unless you go to floating point. I'm not even sure if you could get higher bit depths that way.
The simplest way to increase your bit depth is to use a custom summing filter. Convert your image to floating point, then create a 15x15 filter with all 1's. Use convolution and set the divisor to 1 also. This will add all the values in the 15x15 block and increase your bit depth to about 18. When you resample to 64x64, you will keep the increased bit depth.
Bruce
Bruce Ammons
Ammons Engineering