The morphology and structural properties of galaxies constitute key tracers of the physical processes that shaped these systems throughout their history. Upcoming galaxy Big Data surveys will deliver data scales in the peta-exabyte range in the coming decade. Even with automated galaxy morphology measurements such as supervised machine learning these techniques are time consuming since they are reliant on labeled training datasets. However, unsupervised machine learning does not require labeling which allows for large scale galaxy classification over shorter timescales. We present an unsupervised galaxy classification technique comprised of a series of clustering and size reduction algorithms which has the capability of compressing hundreds of thousands of galaxies to \(\sim \) 180 morphological clusters via the extraction of radial power spectra from multi-band galaxy images, where each cluster contains galaxies of specific visual properties. We are able to validate this method by reproducing well known trends in the colour-mass and sfr-mass planes as a function of morphological type. Lastly, we show how this technique can identify specific objects such as dwarf blue elliptical galaxies. This work acts as a pilot study for upcoming data at unprecedented scales and depth from surveys such as LSST, Euclid and SKA.